CN117114760B - Intelligent recommendation system and method for point exchange based on user consumption behavior analysis - Google Patents

Intelligent recommendation system and method for point exchange based on user consumption behavior analysis Download PDF

Info

Publication number
CN117114760B
CN117114760B CN202310488076.2A CN202310488076A CN117114760B CN 117114760 B CN117114760 B CN 117114760B CN 202310488076 A CN202310488076 A CN 202310488076A CN 117114760 B CN117114760 B CN 117114760B
Authority
CN
China
Prior art keywords
semantic understanding
consumption behavior
matrix
understanding feature
dimensional
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202310488076.2A
Other languages
Chinese (zh)
Other versions
CN117114760A (en
Inventor
曹王强
黄静
黄一帆
林炜
赵颖武
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang Kawin Information Technology Co ltd
Original Assignee
Zhejiang Kawin Information Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang Kawin Information Technology Co ltd filed Critical Zhejiang Kawin Information Technology Co ltd
Priority to CN202310488076.2A priority Critical patent/CN117114760B/en
Publication of CN117114760A publication Critical patent/CN117114760A/en
Application granted granted Critical
Publication of CN117114760B publication Critical patent/CN117114760B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0207—Discounts or incentives, e.g. coupons or rebates
    • G06Q30/0208—Trade or exchange of goods or services in exchange for incentives or rewards
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2431—Multiple classes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0202—Market predictions or forecasting for commercial activities

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Data Mining & Analysis (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Economics (AREA)
  • Computing Systems (AREA)
  • General Business, Economics & Management (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Marketing (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Game Theory and Decision Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The application relates to the technical field of intelligent recommendation, and particularly discloses a point exchange intelligent recommendation system and a method thereof based on user consumption behavior analysis, wherein the system comprises the steps of firstly acquiring consumption behavior data of a user to be recommended and text description of an exchange object to be recommended, then respectively word-dividing the consumption behavior data of the user to be recommended and the text description of the exchange object to be recommended, carrying out two-dimensional arrangement by a context encoder containing an embedded layer to obtain a consumption behavior semantic understanding feature matrix and an exchange object semantic understanding feature matrix, and then carrying out difference, and then obtaining a classification result by a difference extractor based on a convolutional neural network model and a classifier; the method integrates consumption behavior data of the user to be recommended and text description of the exchange object to be recommended, and utilizes deep learning and artificial intelligence technology to realize personalized custom service of the point exchange system, thereby improving interest of the user in exchanging objects, optimizing user experience and further enhancing user viscosity.

Description

Intelligent recommendation system and method for point exchange based on user consumption behavior analysis
Technical Field
The application relates to the technical field of intelligent recommendation, in particular to an intelligent recommendation system and method for point exchange based on user consumption behavior analysis.
Background
Consumer behavior analysis and research is one of the most common and often implemented research in market research, and refers to quantitative and qualitative research on various actions taken by consumers to acquire, use and process consumer goods and decision processes for determining the actions in advance. The method is generally the basis of marketing decisions, is indistinguishable from marketing activities, analyzes and researches consumer behaviors, and has great significance in improving marketing decision level and enhancing effectiveness of marketing strategies.
In the prior art, the combination of the mall and the point system encourages consumers to consume to a certain extent, thereby bringing more opportunities for merchants. However, many point mall systems are single and simple to redeem, resulting in less attractive to consumers. Personalized customization service of the point exchange system is realized, the interest of the user for exchanging articles is improved, and the experience of the user is optimized to improve the viscosity of the user, so that the problem to be solved by the point exchange system is urgent.
Accordingly, a point redemption intelligent recommendation based on user consumption behavior analysis is desired.
Disclosure of Invention
The application provides an intelligent point exchange recommendation system and a method based on user consumption behavior analysis, which are used for integrating consumption behavior data of a user to be recommended and text description of an object to be recommended, and realizing personalized custom service of the point exchange system by utilizing deep learning and artificial intelligence technology, improving the interest of the user in exchanging objects, optimizing user experience and further enhancing user viscosity.
In a first aspect, there is provided a point redemption intelligent recommendation system based on user consumption behavior analysis, the system comprising: the data acquisition module is used for acquiring consumption behavior data of a user to be recommended and text description of an exchange object to be recommended; the consumption behavior semantic understanding module is used for obtaining a plurality of consumption behavior semantic understanding feature vectors through a first context encoder comprising an embedded layer after the consumption behavior data are subjected to word segmentation; the exchange object description semantic understanding module is used for obtaining a plurality of exchange object semantic understanding feature vectors through a second context encoder comprising an embedded layer after word segmentation processing of the text description of the exchange object to be recommended; the data structure adjustment module is used for respectively carrying out two-dimensional arrangement on the plurality of consumption behavior semantic understanding feature vectors and the plurality of exchange object semantic understanding feature vectors to obtain a consumption behavior semantic understanding feature matrix and an exchange object semantic understanding feature matrix; the difference module is used for calculating a difference matrix between the consumption behavior semantic understanding characteristic matrix and the conversion object semantic understanding characteristic matrix; the differential feature capturing module is used for enabling the differential matrix to pass through a differential extractor based on a convolutional neural network model to obtain a differential feature matrix; and the matching result generation module is used for passing the difference feature matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the matching degree of the to-be-recommended exchange object and the current user reaches a preset threshold value.
With reference to the first aspect, in an implementation manner of the first aspect, the consumption behavior semantic understanding module includes: the first word segmentation processing unit is used for carrying out word segmentation processing on the consumption behavior data to obtain a first word sequence; the first embedding vectorization unit is used for mapping each word in the first word sequence into a first word embedding vector by using an embedding layer of the first context encoder so as to obtain a sequence of the first word embedding vector; and a first context coding unit, configured to perform global-based context semantic coding on the sequence of first word embedded vectors using a Bert model based on a converter of the first context encoder to obtain a plurality of consumption behavior semantic understanding feature vectors.
With reference to the first aspect, in an implementation manner of the first aspect, the redeeming object describes a semantic understanding module, including: the second word segmentation processing unit is used for carrying out word segmentation processing on the text description of the to-be-recommended exchange object to obtain a second word sequence; a second embedding vectorization unit, configured to map each word in the second word sequence into a second word embedding vector by using an embedding layer of the second context encoder to obtain a sequence of second word embedding vectors; and a second context encoding unit, configured to perform global-based context semantic encoding on the sequence of second word embedded vectors using a Bert model based on a converter of the second context encoder to obtain a plurality of redeemed semantic understanding feature vectors.
With reference to the first aspect, in an implementation manner of the first aspect, the differential module includes: the unfolding unit is used for respectively conducting feature matrix unfolding on the consumption behavior semantic understanding feature matrix and the exchange object semantic understanding feature matrix to obtain a one-dimensional consumption behavior semantic understanding feature vector and a one-dimensional exchange object semantic understanding feature vector; the Helmholtz free energy factor calculation unit is used for calculating the Helmholtz free energy factors of the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector respectively to obtain a first Helmholtz free energy factor and a second Helmholtz free energy factor; the weighting unit is used for weighting the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector based on the first Helmholtz class free energy factor and the second Helmholtz class free energy factor to obtain a weighted one-dimensional consumption behavior semantic understanding feature vector and a weighted one-dimensional conversion object semantic understanding feature vector; the dimension reconstruction unit is used for respectively carrying out feature dimension reconstruction on the weighted one-dimensional consumption behavior semantic understanding feature vector and the weighted one-dimensional conversion object semantic understanding feature vector so as to restore the weighted consumption behavior semantic understanding feature matrix and the weighted conversion object semantic understanding feature matrix which correspond to the consumption behavior semantic understanding feature matrix and the conversion object semantic understanding feature matrix; and the differential operation unit is used for calculating the differential matrix between the weighted consumption behavior semantic understanding characteristic matrix and the weighted conversion object semantic understanding characteristic matrix.
With reference to the first aspect, in an implementation manner of the first aspect, the helmholtz free energy factor calculating unit is configured to: calculating the Helmholtz type free energy factors of the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector by the following factor calculation formula to obtain a first Helmholtz type free energy factor and a second Helmholtz type free energy factor; wherein, the factor calculation formula is: Wherein/> Representing the feature value of each position in the one-dimensional consumption behavior semantic understanding feature vector,/>The feature values representing the locations in the one-dimensional redeemed article semantic understanding feature vector,And/>Classification probability values respectively representing the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional exchange semantic understanding feature vector, and/>Is the length of the feature vector,/>Represents a logarithmic function value based on 2,/>Representing an exponential operation,/>And/>Representing the first helmholtz-like free energy factor and the second helmholtz-like free energy factor, respectively.
With reference to the first aspect, in an implementation manner of the first aspect, the differential operation unit is configured to: calculating the differential matrix between the weighted consumption behavior semantic understanding feature matrix and the weighted conversion object semantic understanding feature matrix according to the following differential formula; wherein, the difference formula is: Wherein/> Representing the weighted consumption behavior semantic understanding feature matrix,/>Representing difference by location,/>Representing the weighted redemption semantic understanding feature matrix, and/>Representing the differential matrix.
With reference to the first aspect, in an implementation manner of the first aspect, the differential feature capturing module is configured to: each layer of the difference extractor based on the convolutional neural network model is used for respectively carrying out input data in forward transfer of the layer: performing convolution processing on the input data based on the convolution check to generate a convolution feature map; pooling the convolution feature map along a channel dimension to generate a pooled feature map; performing nonlinear activation on the feature values of all positions in the pooled feature map to generate an activated feature map; the input of the difference extractor based on the convolutional neural network model is the difference matrix, and the output of the last layer of the difference extractor based on the convolutional neural network model is the difference feature matrix.
With reference to the first aspect, in an implementation manner of the first aspect, the matching result generating module includes: the matrix unfolding unit is used for unfolding the difference feature matrix into a classification feature vector according to a row vector or a column vector; the full-connection unit is used for carrying out full-connection coding on the classification feature vectors by using a full-connection layer of the classifier so as to obtain full-connection coding feature vectors; the probability unit is used for inputting the fully-connected coding feature vector into a Softmax classification function of the classifier to obtain probability values of the difference feature matrix belonging to various classification labels, wherein the classification labels comprise a preset threshold value for the matching degree of the to-be-recommended exchange object and the current user and a preset threshold value for indicating that the matching degree of the to-be-recommended exchange object and the current user does not reach the preset threshold value; and the classification unit is used for determining the classification label corresponding to the largest probability value as the classification result.
In a second aspect, there is provided a point redemption intelligent recommendation method based on user consumption behavior analysis, the method comprising: acquiring consumption behavior data of a user to be recommended and text description of an exchange object to be recommended; the consumption behavior data is subjected to word segmentation processing and then passes through a first context encoder comprising an embedded layer so as to obtain a plurality of consumption behavior semantic understanding feature vectors; word segmentation processing is carried out on the text description of the to-be-recommended exchange object, and then a plurality of exchange object semantic understanding feature vectors are obtained through a second context encoder comprising an embedded layer; respectively carrying out two-dimensional arrangement on the plurality of consumption behavior semantic understanding feature vectors and the plurality of exchange object semantic understanding feature vectors to obtain a consumption behavior semantic understanding feature matrix and an exchange object semantic understanding feature matrix; calculating a differential matrix between the consumption behavior semantic understanding feature matrix and the exchange object semantic understanding feature matrix; the difference matrix passes through a difference extractor based on a convolutional neural network model to obtain a difference feature matrix; and the difference feature matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the matching degree of the to-be-recommended exchange object and the current user reaches a preset threshold value.
With reference to the second aspect, in an implementation manner of the second aspect, calculating a differential matrix between the consumption behavior semantic understanding feature matrix and the redeemed semantic understanding feature matrix includes: respectively expanding the consumption behavior semantic understanding feature matrix and the exchange object semantic understanding feature matrix to obtain a one-dimensional consumption behavior semantic understanding feature vector and a one-dimensional exchange object semantic understanding feature vector; respectively calculating the Helmholtz type free energy factors of the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector to obtain a first Helmholtz type free energy factor and a second Helmholtz type free energy factor; weighting the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector based on the first Helmholtz class free energy factor and the second Helmholtz class free energy factor to obtain a weighted one-dimensional consumption behavior semantic understanding feature vector and a weighted one-dimensional conversion object semantic understanding feature vector; respectively carrying out feature dimension reconstruction on the weighted one-dimensional consumption behavior semantic understanding feature vector and the weighted one-dimensional exchange semantic understanding feature vector to restore to a weighted consumption behavior semantic understanding feature matrix and a weighted exchange semantic understanding feature matrix corresponding to the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix; and calculating the differential matrix between the weighted consumption behavior semantic understanding feature matrix and the weighted conversion object semantic understanding feature matrix.
In a third aspect, there is provided a chip comprising an input-output interface, at least one processor, at least one memory and a bus, the at least one memory to store instructions, the at least one processor to invoke the instructions in the at least one memory to perform the method in the second aspect.
In a fourth aspect, a computer readable medium is provided for storing a computer program comprising instructions for performing the method of the second aspect described above.
In a fifth aspect, there is provided a computer program product comprising instructions which, when executed by a computer, perform the method of the second aspect described above.
According to the point exchange intelligent recommendation system and the method thereof based on the user consumption behavior analysis, the consumption behavior data of the user to be recommended and the text description of the object to be recommended are synthesized, and the deep learning and artificial intelligent technology is utilized to realize personalized custom services of the point exchange system, so that the interests of the user on the exchange objects are improved, the user experience is optimized, and the user viscosity is enhanced. .
Drawings
FIG. 1 is a schematic block diagram of a point redemption intelligent recommendation system based on user consumption behavior analysis in accordance with an embodiment of the present application.
FIG. 2 is a schematic block diagram of a consumption behavior semantic understanding module in the point redemption intelligent recommendation system based on user consumption behavior analysis in accordance with an embodiment of the present application.
FIG. 3 is a schematic block diagram of a redeemer description semantic understanding module in a point redemption intelligent recommendation system based on user consumption behavior analysis in accordance with an embodiment of the present application.
FIG. 4 is a schematic block diagram of a difference module in the point redemption intelligent recommendation system based on user consumption behavior analysis in accordance with an embodiment of the present application.
FIG. 5 is a schematic block diagram of a matching result generation module in the point redemption intelligent recommendation system based on user consumption behavior analysis in accordance with an embodiment of the present application.
FIG. 6 is a schematic flow chart of a point redemption intelligent recommendation method based on user consumption behavior analysis in accordance with an embodiment of the present application.
Fig. 7 is a schematic diagram of a model architecture of a point redemption intelligent recommendation method based on user consumption behavior analysis according to an embodiment of the application.
Detailed Description
The technical scheme of the application will be described below with reference to the accompanying drawings.
Because of the deep learning-based deep neural network model, related terms and concepts of the deep neural network model that may be related to embodiments of the present application are described below.
1. Deep neural network model: in the deep neural network model, the hidden layers may be convolutional layers and pooled layers. The set of weight values corresponding to the convolutional layer is referred to as a filter, also referred to as a convolutional kernel. The filter and the input eigenvalue are both represented as a multi-dimensional matrix, correspondingly, the filter represented as a multi-dimensional matrix is also called a filter matrix, the input eigenvalue represented as a multi-dimensional matrix is also called an input eigenvalue, of course, besides the input eigenvalue, the eigenvector can also be input, and the input eigenvector is only exemplified by the input eigenvector. The operation of the convolution layer is called a convolution operation, which is to perform an inner product operation on a part of eigenvalues of the input eigenvalue matrix and weight values of the filter matrix.
The operation process of each convolution layer in the deep neural network model can be programmed into software, and then the output result of each layer of network, namely the output characteristic matrix, is obtained by running the software in an operation device. For example, the software performs inner product operation by taking the upper left corner of the input feature matrix of each layer of network as a starting point and taking the size of the filter as a window in a sliding window mode, and extracting data of one window from the feature value matrix each time. After the inner product operation is completed between the data of the right lower corner window of the input feature matrix and the filter, a two-dimensional output feature matrix of each layer of network can be obtained. The software repeats the above process until the entire output feature matrix for each layer of network is generated.
The convolution layer operation process is to slide a window with a filter size across the whole input image (i.e. the input feature matrix), and at each moment, to perform inner product operation on the input feature value covered in the window and the filter, wherein the step length of window sliding is 1. Specifically, the upper left corner of the input feature matrix is used as a starting point, the size of the filter is used as a window, the sliding step length of the window is 1, the input feature value of one window is extracted from the feature value matrix each time and the filter performs inner product operation, and when the data of the lower right corner of the input feature matrix and the filter complete inner product operation, a two-dimensional output feature matrix of the input feature matrix can be obtained.
Since it is often necessary to reduce the number of training parameters, the convolutional layer often requires a periodic introduction of a pooling layer, the only purpose of which is to reduce the spatial size of the image during image processing. The pooling layer may include an average pooling operator and/or a maximum pooling operator for sampling the input image to obtain a smaller size image. The average pooling operator may calculate pixel values in the image over a particular range to produce an average as a result of the average pooling. The max pooling operator may take the pixel with the largest value in a particular range as the result of max pooling. In addition, just as the size of the weighting matrix used in the convolutional layer should be related to the image size, the operators in the pooling layer should also be related to the image size. The size of the image output after the processing by the pooling layer can be smaller than the size of the image input to the pooling layer, and each pixel point in the image output by the pooling layer represents the average value or the maximum value of the corresponding sub-region of the image input to the pooling layer.
Since the functions actually required to be simulated in the deep neural network are nonlinear, the rolling and pooling can only simulate linear functions, in order to introduce nonlinear factors in the deep neural network model to increase the characterization capacity of the whole network, an activation layer is further arranged after the pooling layer, an activation function is arranged in the activation layer, and common excitation functions comprise sigmoid, tanh, reLU functions.
2. The channel is as follows: three parameters of the spatial arrangement of the convolution layers, namely width, height and channel number, wherein the width and the height of the convolution layers are the size of the convolution kernel, and the channel number of the convolution layers corresponds to the number of the convolution kernels. Each convolution kernel can only extract a partial feature of the input data. Each convolution kernel performs a convolution operation with the original input data to obtain a feature matrix, and such feature matrices are collected together, which is called a feature map.
3. Softmax classification function: the Softmax classification function is also called soft maximum function, normalized exponential function. One K-dimensional vector containing arbitrary real numbers can be "compressed" into another K-dimensional real vector such that each element ranges between (0, 1) and the sum of all elements is 1. The Softmax classification function is commonly used to classify problems.
Having described the relevant terms and concepts of the deep neural network model to which embodiments of the present application may relate, the following description of the basic principles of the present application will be presented for ease of understanding by those skilled in the art.
Aiming at the technical problems, the technical conception of the application is as follows: and integrating consumption behavior data of the user to be recommended and text description of the exchange object to be recommended, and utilizing deep learning and artificial intelligence technology to realize personalized custom service of the point exchange system, thereby improving the interest of the user in exchanging the object, optimizing the user experience and further enhancing the user viscosity.
Specifically, in the technical scheme, first, consumption behavior data of a user to be recommended and text description of an exchange to be recommended are obtained. Here, the information such as the interests and the hobbies of the user, the purchasing preference and the like can be known through the consumption behavior data of the user to be recommended, and the information such as the attribute characteristics, the functions, the purposes and the like of the exchange can be understood through the text description of the exchange to be recommended, so that semantic understanding and classification can be better performed, and the recommendation accuracy of the exchange is improved. Specifically, the consumption behavior data of the user can be obtained by recording information such as behavior tracks, browsing records, purchasing records, searching histories and the like of the user in a website or an APP; the text description of the to-be-recommended exchange can be obtained through a commodity detail page, a merchant official network and the like.
And then, the consumption behavior data is subjected to word segmentation processing and then passes through a first context encoder comprising an embedded layer to obtain a plurality of consumption behavior semantic understanding feature vectors, and the text description of the to-be-recommended exchange object is subjected to word segmentation processing and then passes through a second context encoder comprising an embedded layer to obtain a plurality of exchange object semantic understanding feature vectors. It should be appreciated that text data in natural language is typically presented in the form of words or phrases, and that the computer cannot directly understand the text data, and that the text data may be divided into discrete phrases by a word segmentation process. Here, the use of the embedding layer (Embedding) may convert each word after segmentation into a vector, that is, it may map each discrete word to a continuous vector representation in a low-dimensional space, facilitating its processing by a computer. In order to better express semantic information in text, in the technical scheme, a context encoder is used for processing the consumption behavior data and the text description of the to-be-recommended exchange object to capture semantic relations of the words in the context so as to obtain the plurality of consumption behavior semantic understanding feature vectors and the plurality of exchange object semantic understanding feature vectors.
In order to measure the similarity and the matching degree between the to-be-recommended exchange object and the user consumption behavior, in the technical scheme, firstly, the plurality of consumption behavior semantic understanding feature vectors and the plurality of exchange object semantic understanding feature vectors are respectively arranged in two dimensions so as to facilitate matrix operation, and therefore a consumption behavior semantic understanding feature matrix and an exchange object semantic understanding feature matrix are obtained. Then, a differential matrix between the consumption behavior semantic understanding feature matrix and the redeemed semantic understanding feature matrix is calculated. And then, the difference matrix passes through a difference extractor based on a convolutional neural network model to further extract difference characteristic information, so that a difference characteristic matrix is obtained. Here, the difference matrix can represent the difference relationship between the consumption behavior semantic understanding feature matrix and the conversion object semantic understanding feature matrix, but contains a large amount of redundant information, and the classification accuracy can be improved by further extracting the features. The difference extractor based on the convolutional neural network model can effectively extract difference characteristic information and reject redundant information, so that a more compact and accurate difference characteristic matrix is obtained.
After the difference feature matrix is obtained, the difference feature matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the matching degree of the to-be-recommended exchange object and the current user reaches a preset threshold value. Here, the difference feature matrix already contains difference information between consumption behaviors and the exchange objects, can reflect the similarity or the matching degree between the consumption behaviors and the exchange objects, and can judge whether the matching degree between the exchange objects to be recommended and the current user reaches a preset threshold value or not through classification by the classifier. In practical application, the exchange object can be recommended or replaced to be recommended based on the classification result, that is, if the matching degree is higher than the threshold value, the exchange object is considered to be matched with the interest of the user, and the user can be recommended; otherwise, the recommendation is not performed to avoid causing interference or discomfort to the user. By the method, personalized customization service of the point exchange system is realized, the interest of the user in exchanging articles is improved, the user experience is optimized, and therefore the user viscosity is enhanced.
In the technical scheme of the application, for the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix, as the context correlation semantics of the consumption behavior data of the user to be recommended and the text context Wen Yuyi of the text description of the exchange to be recommended are respectively expressed, the heterogeneity of the source data can be amplified due to the extraction of the context correlation, so that the weak correlation distribution examples of class labels relative to the classifier exist in the overall feature distribution of each of the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix, that is, the compatibility of the overall feature distribution of the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix under the class labels of the classifier is lower, which can influence the accuracy of differential extraction between the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix, and thus influence the accuracy of classification results obtained by the classifier through the classifier.
Based on this, the consumption behavior semantic understanding feature matrix and the redeemer semantic understanding feature matrix are preferably first expanded into one-dimensional consumption behavior semantic understanding feature vectors, e.g. denoted asAnd one-dimensional redeemed object semantic understanding feature vectors, e.g., denoted/>Calculating the one-dimensional consumption behavior semantic understanding feature vector/>And the one-dimensional redeemed article semantic understanding feature vector/>The helmholtz-like free energy factor of (c) is specifically:，/> And/> Respectively representing the semantic understanding feature vector/>, of the one-dimensional consumption behaviorAnd the one-dimensional redeemed article semantic understanding feature vector/>And/>Is the length of the feature vector.
Here, based on the helmholtz free energy formula, the one-dimensional consumer behavior semantics can be understood as a feature vectorAnd the one-dimensional redeemed article semantic understanding feature vector/>The respective feature value sets describe the energy value of the predetermined class label as the class free energy of the feature vector as a whole, and the feature vector/>, by semantic understanding the one-dimensional consumption behavior by using the energy value of the predetermined class labelAnd the one-dimensional redeemed article semantic understanding feature vector/>Weighting is carried out, so that the feature vector/>, which is understood by the one-dimensional consumption behavior semantically, can be obtainedAnd the one-dimensional redeemed article semantic understanding feature vector/>Class-dependent prototype instance (prototype instance) distribution of features overlapping with truth instance (groundtruth instance) distribution in class target domain is focused to facilitate semantic understanding of feature vector/>, in the one-dimensional consumption behaviorAnd the one-dimensional redeemed article semantic understanding feature vector/>And under the condition that a class weak correlation distribution example exists in the integral feature distribution, incremental learning is realized by carrying out fuzzy labeling on the class weak correlation distribution example, so that the compatibility of the integral feature distribution under class labels is improved, the accuracy of differential extraction between the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix is improved, and the accuracy of a classification result obtained by the differential feature matrix through a classifier is improved.
Having described the basic principles of the present application, various non-limiting embodiments of the present application will now be described in detail with reference to the accompanying drawings.
FIG. 1 is a schematic block diagram of a point redemption intelligent recommendation system based on user consumption behavior analysis in accordance with an embodiment of the present application. As shown in fig. 1, the intelligent recommendation system 100 for point redemption based on consumer behavior analysis of users includes: the data acquisition module 110 is configured to acquire consumption behavior data of a user to be recommended, and a text description of an exchange to be recommended. It is understood that the information such as interests and preferences of the user, such as purchasing preferences, can be known through the consumption behavior data of the user to be recommended, and the information such as attribute characteristics, functions and purposes of the exchange can be understood through the text description of the exchange to be recommended, so that semantic understanding and classification can be better performed, and the recommendation accuracy of the exchange can be improved. Specifically, the consumption behavior data of the user can be obtained by recording information such as behavior tracks, browsing records, purchasing records, searching histories and the like of the user in a website or an APP; the text description of the to-be-recommended exchange can be obtained through a commodity detail page, a merchant official network and the like.
The consumption behavior semantic understanding module 120 is configured to obtain a plurality of consumption behavior semantic understanding feature vectors by using a first context encoder including an embedded layer after the consumption behavior data is subjected to word segmentation. It should be appreciated that text data in natural language is typically presented in the form of words or words, and that the computer cannot directly understand the text data, and that the text data may be divided into discrete words by a word segmentation process. Here, the use of the embedding layer (Embedding) may convert each word after segmentation into a vector, that is, it may map each discrete word to a continuous vector representation in a low-dimensional space, facilitating its processing by a computer. In order to better express semantic information in text, in the technical scheme, a context encoder is used for processing the text description of the consumption behavior data to capture semantic relations of the words in the context, so that the plurality of consumption behavior semantic understanding feature vectors are obtained.
FIG. 2 is a schematic block diagram of a consumption behavior semantic understanding module in the point redemption intelligent recommendation system based on user consumption behavior analysis in accordance with an embodiment of the present application. Optionally, in an embodiment of the present application, the consumption behavior semantic understanding module 120 includes: a first word segmentation unit 121, configured to perform word segmentation on the consumption behavior data to obtain a first word sequence; a first embedding vectorization unit 122, configured to map each word in the first word sequence into a first word embedding vector by using an embedding layer of the first context encoder to obtain a sequence of first word embedding vectors; and a first context encoding unit 123, configured to perform global-based context semantic encoding on the sequence of the first word embedded vectors using a Bert model based on a converter of the first context encoder to obtain a plurality of consumption behavior semantic understanding feature vectors.
The redeem description semantic understanding module 130 is configured to obtain a plurality of redeem semantic understanding feature vectors through a second context encoder including an embedded layer after the text description of the redeem to be recommended is subjected to word segmentation processing. Similarly, considering that each word in the text description of the to-be-recommended exchange may vary according to different context environments, in order to better express semantic information in the text, in the technical scheme, a context encoder is used for processing the words so as to capture semantic relations of each word in the context, so that the semantic understanding feature vectors of the multiple exchanges are obtained.
FIG. 3 is a schematic block diagram of a redeemer description semantic understanding module in a point redemption intelligent recommendation system based on user consumption behavior analysis in accordance with an embodiment of the present application. Optionally, in an embodiment of the present application, the redeem description semantic understanding module 130 includes: the second word segmentation processing unit 131 is configured to perform word segmentation processing on the text description of the to-be-recommended exchange object to obtain a second word sequence; a second embedding vectorization unit 132, configured to map each word in the second word sequence into a second word embedding vector by using an embedding layer of the second context encoder to obtain a sequence of second word embedding vectors; and a second context encoding unit 133, configured to perform global-based context semantic encoding on the sequence of second word embedded vectors using a Bert model based on a converter of the second context encoder to obtain a plurality of redeemed semantic understanding feature vectors.
The data structure adjustment module 140 is configured to two-dimensionally arrange the plurality of consumption behavior semantic understanding feature vectors and the plurality of exchangeable semantic understanding feature vectors to obtain a consumption behavior semantic understanding feature matrix and an exchangeable semantic understanding feature matrix, respectively. It should be understood that, in order to facilitate matrix operation, the plurality of consumption behavior semantic understanding feature vectors and the plurality of redeemed semantic understanding feature vectors are respectively arranged in two dimensions for integration, so as to obtain a consumption behavior semantic understanding feature matrix and a redeemed semantic understanding feature matrix.
And the difference module 150 is used for calculating a difference matrix between the consumption behavior semantic understanding characteristic matrix and the conversion object semantic understanding characteristic matrix. It should be appreciated that in order to be able to measure the degree of similarity and matching between the redemption to be recommended and the user's consumption behaviour, a differential matrix between the consumption behaviour semantic understanding feature matrix and the redemption semantic understanding feature matrix is calculated.
In particular, in the technical solution of the present application, for the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix, as the context correlation semantics of the consumption behavior data of the user to be recommended and the text context Wen Yuyi of the text description of the exchange to be recommended are expressed respectively, heterogeneity of source data may be amplified due to extraction of context correlation, so that weak correlation distribution examples of class of the classification labels relative to the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix exist in the overall feature distribution of each of the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix, that is, compatibility of the overall feature distribution of the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix under the class labels of the classifier is low, which may affect accuracy of differential extraction between the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix, thereby affecting accuracy of classification results obtained by the classifier. Based on this, the consumption behavior semantic understanding feature matrix and the redeemer semantic understanding feature matrix are preferably first expanded into one-dimensional consumption behavior semantic understanding feature vectors, e.g. denoted asAnd one-dimensional redeemed object semantic understanding feature vectors, e.g., denoted/>Calculating the one-dimensional consumption behavior semantic understanding feature vector/>And the one-dimensional redeemed article semantic understanding feature vector/>Is a helmholtz-like free energy factor.
FIG. 4 is a schematic block diagram of a difference module in the point redemption intelligent recommendation system based on user consumption behavior analysis in accordance with an embodiment of the present application. Optionally, in an embodiment of the present application, the differential module 150 includes: a developing unit 151, configured to develop the feature matrix of the consumption behavior semantic understanding feature matrix and the converted object semantic understanding feature matrix to obtain a one-dimensional consumption behavior semantic understanding feature vector and a one-dimensional converted object semantic understanding feature vector; a helmholtz free energy factor calculating unit 152, configured to calculate helmholtz free energy factors of the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector, respectively, so as to obtain a first helmholtz free energy factor and a second helmholtz free energy factor; the weighting unit 153 is configured to weight the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector based on the first helmholtz class free energy factor and the second helmholtz class free energy factor to obtain a weighted one-dimensional consumption behavior semantic understanding feature vector and a weighted one-dimensional conversion object semantic understanding feature vector; the dimension reconstruction unit 154 is configured to reconstruct the weighted one-dimensional consumption behavior semantic understanding feature vector and the weighted one-dimensional conversion object semantic understanding feature vector into a weighted consumption behavior semantic understanding feature matrix and a weighted conversion object semantic understanding feature matrix corresponding to the consumption behavior semantic understanding feature matrix and the conversion object semantic understanding feature matrix, respectively; and a difference operation unit 155 for calculating the difference matrix between the weighted consumption behavior semantic understanding feature matrix and the weighted conversion object semantic understanding feature matrix.
Optionally, in an embodiment of the present application, the helmholtz free energy factor calculating unit is configured to: calculating the Helmholtz type free energy factors of the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector by the following factor calculation formula to obtain a first Helmholtz type free energy factor and a second Helmholtz type free energy factor; wherein, the factor calculation formula is: Wherein/> Representing the feature value of each position in the one-dimensional consumption behavior semantic understanding feature vector,/>Characteristic values representing various positions in the one-dimensional exchange object semantic understanding characteristic vector,/>And/>Classification probability values respectively representing the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional exchange semantic understanding feature vector, and/>Is the length of the feature vector,/>Represents a logarithmic function value based on 2,/>Representing an exponential operation,/>And/>Representing the first helmholtz-like free energy factor and the second helmholtz-like free energy factor, respectively.
Optionally, in an embodiment of the present application, the differential operation unit is configured to: calculating the differential matrix between the weighted consumption behavior semantic understanding feature matrix and the weighted conversion object semantic understanding feature matrix according to the following differential formula; wherein, the difference formula is: Wherein/> Representing the weighted consumption behavior semantic understanding feature matrix,/>Representing difference by location,/>Representing the weighted redemption semantic understanding feature matrix, and/>Representing the differential matrix.
Here, based on the helmholtz free energy formula, the one-dimensional consumer behavior semantics can be understood as a feature vectorAnd the one-dimensional redeemed article semantic understanding feature vector/>The respective feature value sets describe the energy value of the predetermined class label as the class free energy of the feature vector as a whole, and the feature vector/>, by semantic understanding the one-dimensional consumption behavior by using the energy value of the predetermined class labelAnd the one-dimensional redeemed article semantic understanding feature vector/>Weighting is carried out, so that the feature vector/>, which is understood by the one-dimensional consumption behavior semantically, can be obtainedAnd the one-dimensional redeemed article semantic understanding feature vector/>Class-dependent prototype instance (prototype instance) distribution of features overlapping with truth instance (groundtruth instance) distribution in class target domain is focused to facilitate semantic understanding of feature vector/>, in the one-dimensional consumption behaviorAnd the one-dimensional redeemed article semantic understanding feature vector/>And under the condition that a class weak correlation distribution example exists in the integral feature distribution, incremental learning is realized by carrying out fuzzy labeling on the class weak correlation distribution example, so that the compatibility of the integral feature distribution under class labels is improved, the accuracy of differential extraction between the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix is improved, and the accuracy of a classification result obtained by the differential feature matrix through a classifier is improved.
The differential feature capturing module 160 is configured to pass the differential matrix through a differential extractor based on a convolutional neural network model to obtain a differential feature matrix. It should be understood that, although the differential matrix may represent the differential relationship between the consumption behavior semantic understanding feature matrix and the conversion object semantic understanding feature matrix, a great amount of redundant information is included therein, and further feature extraction is performed on the differential matrix, so that the classification accuracy can be improved. The difference extractor based on the convolutional neural network model can effectively extract difference characteristic information and reject redundant information, so that a more compact and accurate difference characteristic matrix is obtained.
Optionally, in an embodiment of the present application, the differential feature capturing module is configured to: each layer of the difference extractor based on the convolutional neural network model is used for respectively carrying out input data in forward transfer of the layer: performing convolution processing on the input data based on the convolution check to generate a convolution feature map; pooling the convolution feature map along a channel dimension to generate a pooled feature map; performing nonlinear activation on the feature values of all positions in the pooled feature map to generate an activated feature map; the input of the difference extractor based on the convolutional neural network model is the difference matrix, and the output of the last layer of the difference extractor based on the convolutional neural network model is the difference feature matrix.
And the matching result generating module 170 is configured to pass the difference feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate whether the matching degree between the to-be-recommended exchange object and the current user reaches a predetermined threshold. It should be appreciated that after the difference feature matrix is obtained, the difference feature matrix is passed through a classifier to obtain a classification result, where the classification result is used to indicate whether the matching degree of the to-be-recommended exchange object and the current user reaches a predetermined threshold. Here, the difference feature matrix already contains difference information between consumption behaviors and the exchange objects, can reflect the similarity or the matching degree between the consumption behaviors and the exchange objects, and can judge whether the matching degree between the exchange objects to be recommended and the current user reaches a preset threshold value or not through classification by the classifier. In practical application, the exchange object can be recommended or replaced to be recommended based on the classification result, that is, if the matching degree is higher than the threshold value, the exchange object is considered to be matched with the interest of the user, and the user can be recommended; otherwise, the recommendation is not performed to avoid causing interference or discomfort to the user. By the method, personalized customization service of the point exchange system is realized, the interest of the user in exchanging articles is improved, the user experience is optimized, and therefore the user viscosity is enhanced.
FIG. 5 is a schematic block diagram of a matching result generation module in the point redemption intelligent recommendation system based on user consumption behavior analysis in accordance with an embodiment of the present application. Optionally, in an embodiment of the present application, the matching result generating module 170 includes: a matrix expansion unit 171 for expanding the difference feature matrix into classification feature vectors according to row vectors or column vectors; a full-connection unit 172, configured to perform full-connection encoding on the classification feature vector by using a full-connection layer of the classifier to obtain a full-connection encoded feature vector; a probabilizing unit 173, configured to input the fully-connected encoding feature vector into a Softmax classification function of the classifier to obtain probability values of the difference feature matrix belonging to respective classification labels, where the classification labels include a first label for indicating that the matching degree of the to-be-recommended redeeming object with the current user reaches a predetermined threshold value and a second label for indicating that the matching degree of the to-be-recommended redeeming object with the current user does not reach the predetermined threshold value; and a classification unit 174 configured to determine a classification label corresponding to the largest one of the probability values as the classification result. It should be noted that the first tag p1 and the second tag p2 do not include a manually set concept, and in fact, during the training process, the computer model does not have a concept of "whether the matching degree of the to-be-recommended exchange object and the current user reaches a predetermined threshold", which is just two kinds of classification tags, and the probability that the output feature is the two classification tags sign, that is, the sum of p1 and p2 is one. Therefore, the classification result of whether the matching degree of the to-be-recommended exchange object and the current user reaches the preset threshold value is actually converted into the classification probability distribution conforming to the natural rule through classifying the labels, and the physical meaning of the natural probability distribution of the labels is essentially used instead of the language text meaning of whether the matching degree of the to-be-recommended exchange object and the current user reaches the preset threshold value.
In summary, the point exchange intelligent recommendation system based on user consumption behavior analysis synthesizes consumption behavior data of users to be recommended and text description of the objects to be recommended, and utilizes deep learning and artificial intelligence technology to realize personalized custom services of the point exchange system, promote interests of users for exchanging objects, optimize user experience and further enhance user viscosity.
FIG. 6 is a schematic flow chart of a point redemption intelligent recommendation method based on user consumption behavior analysis in accordance with an embodiment of the present application. As shown in fig. 6, the method includes: s110, acquiring consumption behavior data of a user to be recommended and text description of an exchange object to be recommended; s120, processing the consumption behavior data through word segmentation, and then obtaining a plurality of consumption behavior semantic understanding feature vectors through a first context encoder comprising an embedded layer; s130, word segmentation processing is carried out on the text description of the to-be-recommended exchange object, and a plurality of exchange object semantic understanding feature vectors are obtained through a second context encoder comprising an embedded layer; s140, respectively carrying out two-dimensional arrangement on the plurality of consumption behavior semantic understanding feature vectors and the plurality of exchange object semantic understanding feature vectors to obtain a consumption behavior semantic understanding feature matrix and an exchange object semantic understanding feature matrix; s150, calculating a difference matrix between the consumption behavior semantic understanding feature matrix and the exchange object semantic understanding feature matrix; s160, the differential matrix is processed through a differential extractor based on a convolutional neural network model to obtain a differential feature matrix; and S170, passing the difference feature matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the matching degree of the to-be-recommended exchange object and the current user reaches a preset threshold value.
Fig. 7 is a schematic diagram of a model architecture of a point redemption intelligent recommendation method based on user consumption behavior analysis according to an embodiment of the application. As shown in fig. 7, the input of the model architecture of the point redemption intelligent recommendation method based on the user consumption behavior analysis is consumption behavior data of the user to be recommended, and a text description of the redeemer to be recommended. And then, carrying out operation of two branches, wherein the first branch carries out word segmentation processing on the consumption behavior data, then obtains a plurality of consumption behavior semantic understanding feature vectors through a first context encoder comprising an embedded layer, and carries out two-dimensional arrangement on the plurality of consumption behavior semantic understanding feature vectors to obtain a consumption behavior semantic understanding feature matrix. And the second branch is used for processing the text description of the to-be-recommended exchange through word segmentation, obtaining a plurality of exchange semantic understanding feature vectors through a second context encoder comprising an embedded layer, and carrying out two-dimensional arrangement on the plurality of exchange semantic understanding feature vectors to obtain an exchange semantic understanding feature matrix. And then, calculating a differential matrix between the consumption behavior semantic understanding feature matrix and the conversion object semantic understanding feature matrix, and passing the differential matrix through a differential extractor based on a convolutional neural network model to obtain a differential feature matrix. And finally, the difference feature matrix passes through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the matching degree of the to-be-recommended exchange object and the current user reaches a preset threshold value.
Here, it will be understood by those skilled in the art that the specific operations of the respective steps in the above-described point redemption intelligent recommendation method based on the user's consumption behavior analysis have been described in detail in the above description of the point redemption intelligent recommendation system based on the user's consumption behavior analysis with reference to fig. 1 to 5, and thus, repeated descriptions thereof will be omitted.
The embodiment of the application also provides a chip system, which comprises at least one processor, and when the program instructions are executed in the at least one processor, the method provided by the embodiment of the application is realized.
The embodiment of the invention also provides a computer storage medium, on which a computer program is stored, which when executed by a computer causes the computer to perform the method of the above-described method embodiment.
The present invention also provides a computer program product comprising instructions which, when executed by a computer, cause the computer to perform the method of the method embodiment described above.
In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, produces a flow or function in accordance with embodiments of the present invention, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by a wired (e.g., coaxial cable, fiber optic, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains an integration of one or more available media. The usable medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a digital video disc (digital video disc, DVD)), or a semiconductor medium (e.g., a Solid State Drive (SSD)), or the like.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.

Claims (7)

1. An intelligent recommendation system for point redemption based on consumer behavior analysis, comprising:
the data acquisition module is used for acquiring consumption behavior data of a user to be recommended and text description of an exchange object to be recommended;
the consumption behavior semantic understanding module is used for obtaining a plurality of consumption behavior semantic understanding feature vectors through a first context encoder comprising an embedded layer after the consumption behavior data are subjected to word segmentation;
The exchange object description semantic understanding module is used for obtaining a plurality of exchange object semantic understanding feature vectors through a second context encoder comprising an embedded layer after word segmentation processing of the text description of the exchange object to be recommended;
The data structure adjustment module is used for respectively carrying out two-dimensional arrangement on the plurality of consumption behavior semantic understanding feature vectors and the plurality of exchange object semantic understanding feature vectors to obtain a consumption behavior semantic understanding feature matrix and an exchange object semantic understanding feature matrix;
The difference module is used for calculating a difference matrix between the consumption behavior semantic understanding characteristic matrix and the conversion object semantic understanding characteristic matrix;
The differential feature capturing module is used for enabling the differential matrix to pass through a differential extractor based on a convolutional neural network model to obtain a differential feature matrix; and
The matching result generation module is used for enabling the difference feature matrix to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the matching degree of the to-be-recommended exchange object and the current user reaches a preset threshold value or not;
Wherein, the difference module includes:
The unfolding unit is used for respectively conducting feature matrix unfolding on the consumption behavior semantic understanding feature matrix and the exchange object semantic understanding feature matrix to obtain a one-dimensional consumption behavior semantic understanding feature vector and a one-dimensional exchange object semantic understanding feature vector;
The Helmholtz free energy factor calculation unit is used for calculating the Helmholtz free energy factors of the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector respectively to obtain a first Helmholtz free energy factor and a second Helmholtz free energy factor;
the weighting unit is used for weighting the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector based on the first Helmholtz class free energy factor and the second Helmholtz class free energy factor to obtain a weighted one-dimensional consumption behavior semantic understanding feature vector and a weighted one-dimensional conversion object semantic understanding feature vector;
The dimension reconstruction unit is used for respectively carrying out feature dimension reconstruction on the weighted one-dimensional consumption behavior semantic understanding feature vector and the weighted one-dimensional conversion object semantic understanding feature vector so as to restore the weighted consumption behavior semantic understanding feature matrix and the weighted conversion object semantic understanding feature matrix which correspond to the consumption behavior semantic understanding feature matrix and the conversion object semantic understanding feature matrix; and
The difference operation unit is used for calculating the difference matrix between the weighted consumption behavior semantic understanding characteristic matrix and the weighted conversion object semantic understanding characteristic matrix;
Wherein, the helmholtz class free energy factor calculating unit is used for: calculating the Helmholtz type free energy factors of the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector by the following factor calculation formula to obtain a first Helmholtz type free energy factor and a second Helmholtz type free energy factor;
Wherein, the factor calculation formula is:
Wherein v 1i represents the feature value of each position in the one-dimensional consumption behavior semantic understanding feature vector, v 2o represents the feature value of each position in the one-dimensional conversion object semantic understanding feature vector, p 1 and p 2 represent the classification probability values of the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector respectively, L is the length of the feature vector, log represents a logarithmic function value based on 2, exp (·) represents an exponential operation, and w 1 and w 2 represent the first helmholtz free energy factor and the second helmholtz free energy factor respectively.
2. The intelligent recommendation system for redemption of points based on analysis of consumer behavior of claim 1, wherein the semantic understanding module for consumer behavior comprises:
the first word segmentation processing unit is used for carrying out word segmentation processing on the consumption behavior data to obtain a first word sequence;
The first embedding vectorization unit is used for mapping each word in the first word sequence into a first word embedding vector by using an embedding layer of the first context encoder so as to obtain a sequence of the first word embedding vector; and
And the first context coding unit is used for performing global context semantic coding on the sequence of the first word embedded vectors by using a converter-based Bert model of the first context coder so as to obtain a plurality of consumption behavior semantic understanding feature vectors.
3. The intelligent recommendation system for redemption of points based on analysis of consumer behavior of claim 2, wherein the redemption object description semantic understanding module comprises:
the second word segmentation processing unit is used for carrying out word segmentation processing on the text description of the to-be-recommended exchange object to obtain a second word sequence;
A second embedding vectorization unit, configured to map each word in the second word sequence into a second word embedding vector by using an embedding layer of the second context encoder to obtain a sequence of second word embedding vectors; and
And the second context coding unit is used for performing global context semantic coding on the sequence of the second word embedded vectors by using a converter-based Bert model of the second context coder so as to obtain a plurality of converted object semantic understanding feature vectors.
4. The intelligent recommendation system for point redemption based on consumer behavior analysis of claim 3, wherein the differential computing unit is configured to: calculating the differential matrix between the weighted consumption behavior semantic understanding feature matrix and the weighted conversion object semantic understanding feature matrix according to the following differential formula;
wherein, the difference formula is:
Wherein M a represents the weighted consumption behavior semantic understanding feature matrix, Representing difference by location, M b represents the weighted redemption semantic understanding feature matrix, and M n represents the difference matrix.
5. The intelligent recommendation system for redemption of points based on consumer behavior analysis of claim 4, wherein the differential feature capture module is configured to: each layer of the difference extractor based on the convolutional neural network model is used for respectively carrying out input data in forward transfer of the layer:
Performing convolution processing on the input data based on the convolution check to generate a convolution feature map;
pooling the convolution feature map along a channel dimension to generate a pooled feature map; and
Non-linear activation is carried out on the characteristic values of all positions in the pooled characteristic map so as to generate an activated characteristic incremental map;
the input of the difference extractor based on the convolutional neural network model is the difference matrix, and the output of the last layer of the difference extractor based on the convolutional neural network model is the difference feature matrix.
6. The intelligent recommendation system for redemption of points based on analysis of consumer behavior of claim 5, wherein the matching result generation module comprises:
the matrix unfolding unit is used for unfolding the difference feature matrix into a classification feature vector according to a row vector or a column vector;
the full-connection unit is used for carrying out full-connection coding on the classification feature vectors by using a full-connection layer of the classifier so as to obtain full-connection coding feature vectors;
The probability unit is used for inputting the fully-connected coding feature vector into a Softmax classification function of the classifier to obtain probability values of the difference feature matrix belonging to various classification labels, wherein the classification labels comprise a preset threshold value for the matching degree of the to-be-recommended exchange object and the current user and a preset threshold value for indicating that the matching degree of the to-be-recommended exchange object and the current user does not reach the preset threshold value; and
And the classification unit is used for determining the classification label corresponding to the maximum probability value as the classification result.
7. The intelligent recommendation method for point redemption based on user consumption behavior analysis is characterized by comprising the following steps:
Acquiring consumption behavior data of a user to be recommended and text description of an exchange object to be recommended;
the consumption behavior data is subjected to word segmentation processing and then passes through a first context encoder comprising an embedded layer so as to obtain a plurality of consumption behavior semantic understanding feature vectors;
Word segmentation processing is carried out on the text description of the to-be-recommended exchange object, and then a plurality of exchange object semantic understanding feature vectors are obtained through a second context encoder comprising an embedded layer;
Respectively carrying out two-dimensional arrangement on the plurality of consumption behavior semantic understanding feature vectors and the plurality of exchange object semantic understanding feature vectors to obtain a consumption behavior semantic understanding feature matrix and an exchange object semantic understanding feature matrix;
Calculating a differential matrix between the consumption behavior semantic understanding feature matrix and the exchange object semantic understanding feature matrix;
the difference matrix passes through a difference extractor based on a convolutional neural network model to obtain a difference feature matrix; and
The difference feature matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the matching degree of the to-be-recommended exchange object and the current user reaches a preset threshold value or not;
wherein calculating a differential matrix between the consumption behavior semantic understanding feature matrix and the conversion object semantic understanding feature matrix comprises:
Respectively expanding the consumption behavior semantic understanding feature matrix and the exchange object semantic understanding feature matrix to obtain a one-dimensional consumption behavior semantic understanding feature vector and a one-dimensional exchange object semantic understanding feature vector;
Respectively calculating the Helmholtz type free energy factors of the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector to obtain a first Helmholtz type free energy factor and a second Helmholtz type free energy factor;
weighting the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector based on the first Helmholtz class free energy factor and the second Helmholtz class free energy factor to obtain a weighted one-dimensional consumption behavior semantic understanding feature vector and a weighted one-dimensional conversion object semantic understanding feature vector;
respectively carrying out feature dimension reconstruction on the weighted one-dimensional consumption behavior semantic understanding feature vector and the weighted one-dimensional exchange semantic understanding feature vector to restore to a weighted consumption behavior semantic understanding feature matrix and a weighted exchange semantic understanding feature matrix corresponding to the consumption behavior semantic understanding feature matrix and the exchange semantic understanding feature matrix; and
Calculating the difference matrix between the weighted consumption behavior semantic understanding feature matrix and the weighted conversion object semantic understanding feature matrix;
The method for calculating the Helmholtz type free energy factors of the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector to obtain a first Helmholtz type free energy factor and a second Helmholtz type free energy factor comprises the following steps of: calculating the Helmholtz type free energy factors of the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector by the following factor calculation formula to obtain a first Helmholtz type free energy factor and a second Helmholtz type free energy factor;
Wherein, the factor calculation formula is:
Wherein v 1i represents the feature value of each position in the one-dimensional consumption behavior semantic understanding feature vector, v 2i represents the feature value of each position in the one-dimensional conversion object semantic understanding feature vector, p 1 and p 2 represent the classification probability values of the one-dimensional consumption behavior semantic understanding feature vector and the one-dimensional conversion object semantic understanding feature vector respectively, L is the length of the feature vector, log represents a logarithmic function value based on 2, exp (·) represents an exponential operation, and w 1 and w 2 represent the first helmholtz free energy factor and the second helmholtz free energy factor respectively.
CN202310488076.2A 2023-05-04 2023-05-04 Intelligent recommendation system and method for point exchange based on user consumption behavior analysis Active CN117114760B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310488076.2A CN117114760B (en) 2023-05-04 2023-05-04 Intelligent recommendation system and method for point exchange based on user consumption behavior analysis

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310488076.2A CN117114760B (en) 2023-05-04 2023-05-04 Intelligent recommendation system and method for point exchange based on user consumption behavior analysis

Publications (2)

Publication Number Publication Date
CN117114760A CN117114760A (en) 2023-11-24
CN117114760B true CN117114760B (en) 2024-05-28

Family

ID=88806200

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310488076.2A Active CN117114760B (en) 2023-05-04 2023-05-04 Intelligent recommendation system and method for point exchange based on user consumption behavior analysis

Country Status (1)

Country Link
CN (1) CN117114760B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110148023A (en) * 2019-05-15 2019-08-20 山大地纬软件股份有限公司 The electric power integral Method of Commodity Recommendation and system that logic-based returns
CN113590970A (en) * 2021-08-11 2021-11-02 东北林业大学 Personalized digital book recommendation system and method based on reader preference, computer and storage medium
CN114565041A (en) * 2022-02-28 2022-05-31 上海嘉甲茂技术有限公司 Payment big data analysis system based on internet finance and analysis method thereof

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11176330B2 (en) * 2019-07-22 2021-11-16 Advanced New Technologies Co., Ltd. Generating recommendation information

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110148023A (en) * 2019-05-15 2019-08-20 山大地纬软件股份有限公司 The electric power integral Method of Commodity Recommendation and system that logic-based returns
CN113590970A (en) * 2021-08-11 2021-11-02 东北林业大学 Personalized digital book recommendation system and method based on reader preference, computer and storage medium
CN114565041A (en) * 2022-02-28 2022-05-31 上海嘉甲茂技术有限公司 Payment big data analysis system based on internet finance and analysis method thereof

Also Published As

Publication number Publication date
CN117114760A (en) 2023-11-24

Similar Documents

Publication Publication Date Title
Verdonck et al. Special issue on feature engineering editorial
CN113508378B (en) Training method, recommendation method, device and computer readable medium for recommendation model
CN104298682B (en) A kind of evaluation method and mobile phone of the information recommendation effect based on Facial Expression Image
Peng et al. Research on image feature extraction and retrieval algorithms based on convolutional neural network
CN112487199B (en) User characteristic prediction method based on user purchasing behavior
CN116911929B (en) Advertisement service terminal and method based on big data
CN108427754B (en) Information pushing method, analysis device, computer storage medium and terminal
Zhao et al. PCA dimensionality reduction method for image classification
CN116739695A (en) Electronic commerce management system and method based on big data
CN117036894B (en) Multi-mode data classification method and device based on deep learning and computer equipment
CN115730125A (en) Object identification method and device, computer equipment and storage medium
CN117114760B (en) Intelligent recommendation system and method for point exchange based on user consumption behavior analysis
CN112463966A (en) False comment detection model training method, detection method and device
CN116796248A (en) Forest health environment assessment system and method thereof
CN116089605A (en) Text emotion analysis method based on transfer learning and improved word bag model
CN112328899B (en) Information processing method, information processing apparatus, storage medium, and electronic device
McAllister et al. Towards personalised training of machine learning algorithms for food image classification using a smartphone camera
CN114692715A (en) Sample labeling method and device
Chen et al. Probabilistic urban structural damage classification using bitemporal satellite images
Shanqing et al. A multi-level feature weight fusion model for salient object detection
Wu et al. DDFPN: Context enhanced network for object detection
Rahman et al. Machine Learning Based Parkinson’s Disease Diagnosis using Hand Writing Related Variables
CN115689648B (en) User information processing method and system applied to directional delivery
CN117807270B (en) Video recommendation method, device, equipment and storage medium based on news content
CN117938951B (en) Information pushing method, device, computer equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant