CN113988951A - Commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering - Google Patents
Commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering Download PDFInfo
- Publication number
- CN113988951A CN113988951A CN202111373959.6A CN202111373959A CN113988951A CN 113988951 A CN113988951 A CN 113988951A CN 202111373959 A CN202111373959 A CN 202111373959A CN 113988951 A CN113988951 A CN 113988951A
- Authority
- CN
- China
- Prior art keywords
- user
- commodity
- behavior
- collaborative filtering
- data
- 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.)
- Pending
Links
Images
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
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0255—Targeted advertisements based on user history
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine 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
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
- G06Q30/0271—Personalized advertisement
-
- 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
- G06Q30/0241—Advertisements
- G06Q30/0277—Online advertisement
-
- 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
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Theoretical Computer Science (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering, which comprises the steps of firstly, acquiring historical behavior data of a user on a commodity, and preprocessing the historical behavior data to obtain a verification set, a training set and a test set; then embedding the user, commodity and behavior information, and initializing the feature vector randomly; calculating a similarity score according to a CP decomposition method, and setting and optimizing a square loss function; and finally, jointly learning by adopting a multi-task learning framework, and adjusting and training the model. The method can be used for uranium mine raw material recommendation, assists user decision, and can realize individual recommendation for users under the conditions of unclear user indexes, insufficient historical purchase data and large quantity of candidate raw materials. The cold start problem caused by sparse data can be effectively reduced, and the user can be helped to make a purchase decision quickly and accurately. The method reaches considerable height in recommendation precision and training speed, and has potential for practical application.
Description
Technical Field
The invention belongs to the field of natural language processing and information retrieval, and particularly relates to a commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering.
Background
Recommendation systems have been widely used in various online services, particularly e-commerce platforms. In the face of huge amount of raw materials, only relying on active search of users can cause the users to be submerged in the problem of information overload. On the platform, the recommendation system pushes raw material information and suggestions to the user according to historical interaction information of the user on the platform, helps the user determine raw materials to be purchased, and assists the user in making decisions. The recommendation system is established on massive user behavior information, the requirement cannot be met increasingly by singly utilizing historical purchase records, and a new technology is urgently needed to be used for fusing various user behavior information so as to improve the recommendation effect.
Tensor decomposition is one of the technical and research fields of machine learning, and artificial intelligence is realized in a computing system by establishing and decomposing a tensor composed of element values determined by a plurality of dimensional data. Due to the fact that the multi-dimensional tensor macroelement values are missing, factor matrixes expressing different dimensions can be obtained through a series of low-rank decomposition approximation methods, tensor decomposition has the characteristic learning capacity, and missing value prediction and filling can be achieved. The decomposition method used by tensor decomposition has various forms, and the methods of tensor decomposition include CP decomposition, Tucker decomposition, BTD decomposition and chain decomposition according to the factor form used by decomposition. Tensor decomposition uses data to update parameters in its construction to achieve a training goal, a process commonly referred to as "learning". A common method of learning is the gradient descent algorithm and its variants, some statistical learning theory being used for the optimization of the learning process.
Collaborative filtering is a common algorithm idea in a recommendation system, and raw materials are recommended for a user according to similarity scores of users or commodities. Especially, the collaborative filtering idea based on commodity similarity has become the mainstream idea of the e-commerce platform recommendation system.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering, can be used for uranium mine raw material recommendation and user decision assistance, and can realize personalized recommendation for users after training of various behavior data under the conditions that user interest indexes are unclear, historical purchase data are insufficient, and the number of candidate raw materials is large. The cold start problem caused by sparse data can be effectively reduced, and the user can be helped to make a purchase decision quickly and accurately. The method reaches considerable height in recommendation precision and training speed, and has potential for practical application.
The invention is realized by the following technical scheme:
a commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering comprises the following steps:
step S1: acquiring historical behavior data of a user on a commodity, and preprocessing the historical behavior data to obtain a verification set, a training set and a test set;
step S2: embedding user, commodity and behavior information, and initializing a feature vector at random;
step S3: calculating a similarity score according to a CP decomposition method, and setting and optimizing a square loss function;
step S4: and (4) performing joint learning by adopting a multi-task learning framework, and adjusting and training the model.
In the above technical solution, in step S1, the historical behavior data of the user on the commodity includes data of behavior record of the user purchasing the commodity, data of behavior record of the user adding the commodity to a shopping cart, and data of behavior record of the user browsing the commodity, and the historical behavior data of the user on the commodity includes user information, behavior information, and commodity information. The behavior information comprises three types of behaviors of purchasing commodities, adding shopping carts and browsing commodities.
In the above technical solution, in step S1, the additional label information included in the input historical behavior data of the user on the commodity, such as the name and model of the device, and the user privacy information, is removed.
In the above technical solution, in step S1, it is further necessary to combine the history records of multiple interactions between the same user and the same commodity under a certain action into one record.
In the above technical solution, in step S2, randomly initializing user, behavior, and commodity memory component parameters during first training, and respectively embedding the input information of the user, behavior, and commodity into the memory components of the user, behavior, and commodity, where the memory component of the user is M, the memory component of the behavior is H, and the memory component of the commodity is E; wherein the user's preferences are stored in a user memory component slice muThe characteristics of behavior b are stored in a behavior memory component slice hbIn the commodity i, the characteristics of the commodity i are stored in e of the commodity memory sliceiIn the initial training, the feature vectors in each memory component are randomly initialized to obtain an embedded representation in its specific behavior for a specific commodity and a specific user.
In the above technical solution, in step S3, a dropout method is used to randomly discard part of the input users, and according to CP decomposition, a method of adding the embedded expressions of the users, behaviors and commodities after element-by-element multiplication is used to calculate similarity scores of the users and the commodities under different behaviors; a square loss function is used for model training, calculation of the loss function is divided into user, behavior, commodity triplets and all triplets with historical behavior records, and the model training process is accelerated by abandoning constant items irrelevant to gradients and changing the accumulation and calculation sequence.
In the above technical solution, in step S4, a multitask learning framework is adopted for joint learning, and a model is adjusted and trained, specifically including the following steps: the interaction relation among the various behaviors is learned and adjusted through a multi-task learning framework, and a data set, the iteration times and model parameters are adjusted according to the recommendation result; and after multiple adjustments, training the recommendation model formally.
The invention also provides a computer-readable storage medium, in which a computer program is stored which, when being executed, carries out the above-mentioned method steps.
The invention has the advantages and beneficial effects that:
according to the commodity recommendation learning model based on tensor decomposition and collaborative filtering, interactive records of various behaviors are introduced, a CP decomposition scoring calculation method is adopted, a multitask learning framework is used, and comparison experiments show that the hit rate and normalized depreciation accumulated gain of the experimental effect of the commodity recommendation learning model based on tensor decomposition and collaborative filtering are high, and the effectiveness of the model is good.
The method can be used for uranium mine raw material recommendation and assisting user decision, and can realize personalized recommendation for users after training of various behavior data under the conditions that user interest indexes are not clear, historical purchase data are insufficient, and the number of candidate raw materials is large. The cold start problem caused by sparse data can be effectively reduced, and the user can be helped to make a purchase decision quickly and accurately. The method reaches considerable height in recommendation precision and training speed, and has potential for practical application.
Drawings
Fig. 1 is a flowchart of a commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering according to the present invention.
Fig. 2 is a schematic network architecture diagram of the commodity recommendation learning model based on tensor decomposition and collaborative filtering according to the present invention.
For a person skilled in the art, other relevant figures can be obtained from the above figures without inventive effort.
Detailed Description
In order to make the technical solution of the present invention better understood, the technical solution of the present invention is further described below with reference to specific examples.
Example one
Referring to fig. 1 and 2, a method for constructing a commodity recommendation learning model based on tensor decomposition and collaborative filtering includes the following steps:
step S1: reading the historical behavior data of a user on a commodity by using a python writing program, and preprocessing to obtain a verification set, a training set and a test set, specifically comprising the following steps:
step S101: inputting historical behavior records of a user on commodities for model training and recommendation, wherein the historical behavior records of the user on the commodities comprise behavior records of the user on purchasing the commodities, behavior records of the user on adding the commodities to a shopping cart and behavior records of the user on browsing the commodities, and the behavior records of the user comprise user information, behavior information (comprising purchasing behavior and behavior browsing behavior of adding the shopping cart) and commodity information; in addition, the input historical behavior data of the user on the commodity contains some additional marks, such as the name and the model of equipment, user privacy information and the like, which need to be cut off, and the historical records of multiple interactions of the same user and the same commodity under a certain behavior need to be combined into one record.
Step S102: and storing the records into a text file according to the behavior types and naming the records in a unified format.
Step S103: and dividing a data set to generate txt files under the Main folder, wherein the txt files comprise a verification set, a training set and a test set, and the training sets are divided according to behavior categories and comprise a purchasing training set, an adding shopping cart training set and a browsing training set.
Step S2: embedding the user information, the behavior information and the commodity information respectively, embedding the input user information, behavior information and commodity information into memory components of the user information, behavior information and commodity information respectively, wherein the memory component of the user is M, the memory component of the behavior is H, the memory component of the commodity is E, and the preference u of the user is stored in a user memory component slice MuThe characteristics of behavior b are stored in a behavior memory component slice hbWherein the characteristics of the item i are stored in the item memory slice eiIn (1). During initial training, the parameters of each memory component need to be initialized randomly (i.e. randomly initializing the feature vectors) to obtain an embedded representation of the specific behavior of the specific commodity and the specific user.
Step S3: calculating a similarity score according to a CP decomposition method, and setting and optimizing a square loss function, specifically, the method comprises the following steps:
step S301: and (4) embedding layer node information by adopting a dropout layer random abandon part, wherein the dropout proportion is adjusted according to data.
Step S302, calculating similarity scores by using a CP decomposition method, and calculating the similarity scores of the user and the commodity under different behaviors by adopting a method of multiplying the embedded expressions of the user, the behaviors and the commodity element by element and then adding the multiplied expressionsAs shown in equation (1):
in formula (1), k is an index for the user, commodity, behavior embedding representation.
Step S303: setting and optimizing a square loss function, which comprises the following specific steps:
a square loss function is used for model training, calculation of the loss function is divided into a user, behavior and commodity triple with historical behavior records and all triples, and a model training process is accelerated by abandoning constant terms irrelevant to gradient and changing the accumulation and calculation sequence, as shown in a formula (2), a formula (3) and a formula (4).
In the formulas (2), (3) and (4), B refers to the user set which is fed into the model in batches in the training process, V refers to all raw material sets,is a constant value set according to the data, + represents the presence of a history triplet, LbIn the form of an initial square-loss function,in order to optimize the squared loss function,the way of changing the order of accumulation and calculation for similarity scoring, a gradient descent method is used to minimize the loss function. The Adam optimizer is adopted in the method, the gradient descending rate is adaptively adjusted, and whether a regularization term is added or not is selected according to needs to limit the model parameters.
Step S4: and (3) jointly learning by adopting a multi-task learning framework, adjusting and training the model, as shown in formula (5):
in the formula (5), λbThe task weight parameters for different user behavior data, the weight parameter sum is 1.
Example two
In this embodiment, validity experimental verification is performed on the commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering described in the first embodiment.
In the embodiment, a comparison test is performed through six baseline experiments, all unrecorded user, behavior and commodity triples are used as negative samples in the experiments, an Adam optimizer adopts default parameters in training, the dropout ratio is set to be 0.5, the embedding size d of a memory is 64, and a commodity information embedding matrix, a behavior information embedding matrix and a user information embedding matrix are all generated randomly and automatically. The number of training sets, the number of verification sets, the number of test sets and the number of iterations in the experiment all have important influence on the model table.
In the embodiment, two evaluation indexes, namely Hit Rate (HR) and normalized breaking cumulative gain (NDCG) values, are used to evaluate the effect of the commodity recommendation learning model based on tensor decomposition and collaborative filtering, the hit rate calculation mainly aims at evaluating the proportion of the number of items in a top-N recommendation list obtained by the model in a test set, and the normalized breaking cumulative gain calculation mainly aims at comparing the effect of the recommendation list generated by the evaluation model with the effect of the recommendation list generated in an ideal state. The formula for calculating the Hit Rate (HR) is shown in formula (6), and the formula for calculating the normalized break-up cumulative gain (NDCG) is shown in formula (7).
The data value | GT | in equation (6) refers to all test sets, and numberfhits @ K is the sum of the number of test sets belonging to the top K recommendation list for each user. DCG in equation (7) is the average discounted cumulative gain, and the ideal DCG @ K is the maximum DCG value under ideal conditions. The hit rate and the normalized breaking accumulated gain are both larger, and the effect is better.
The experimental effects of the comparative experiments are shown in table 1, and the three methods of Neural Collaborative Filtering (NCF), bayesian probabilistic model (BPR) and Neural Matrix Factorization (NMF) are methods that only adopt purchasing behavior, joint matrix factorization (CMF), multi-channel bayesian probabilistic model (MC-BPR) and neural multi-behavior recommendation (NMTR) are methods that use multiple behavior records. The results show that the model recorded by adopting various behavior data is better than the model of the recommendation system only using the purchasing behavior, the hit rate and the normalized depreciation accumulated gain of the experimental effect of the commodity recommendation learning model based on tensor decomposition and collaborative filtering are the highest, and the effectiveness of the model is good.
Table 1: experimental evaluation index table
The invention has been described in an illustrative manner, and it is to be understood that any simple variations, modifications or other equivalent changes which can be made by one skilled in the art without departing from the spirit of the invention fall within the scope of the invention.
Claims (10)
1. A commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering is characterized by comprising the following steps:
step S1: acquiring historical behavior data of a user on a commodity, and preprocessing the historical behavior data to obtain a verification set, a training set and a test set;
step S2: embedding user, commodity and behavior information, and initializing a feature vector at random;
step S3: calculating a similarity score according to a CP decomposition method, and setting and optimizing a square loss function;
step S4: and (4) performing joint learning by adopting a multi-task learning framework, and adjusting and training the model.
2. The method for constructing the commodity recommendation learning model based on tensor decomposition and collaborative filtering as claimed in claim 1, wherein: in step S1, the historical behavior data of the user on the goods includes the data of the behavior record of the user purchasing the goods, the data of the behavior record of the user adding the goods to the shopping cart, and the data of the behavior record of the user browsing the goods, and the historical behavior data of the user on the goods includes the user information, the behavior information and the goods information.
3. The method for constructing the commodity recommendation learning model based on tensor decomposition and collaborative filtering as claimed in claim 2, wherein: in step S1, the behavior information includes three types, namely a behavior of purchasing goods, a behavior of adding a shopping cart, and a behavior of browsing goods.
4. The method for constructing the commodity recommendation learning model based on tensor decomposition and collaborative filtering as claimed in claim 2, wherein: in step S1, the additional label information contained in the input historical behavior data of the user on the commodity, such as the name and model of the device, and the user privacy information, is removed.
5. The method for constructing the commodity recommendation learning model based on tensor decomposition and collaborative filtering as claimed in claim 2, wherein: in step S1, the history records of multiple interactions between the same user and the same product under a certain action are combined into one record.
6. The method for constructing the commodity recommendation learning model based on tensor decomposition and collaborative filtering as claimed in claim 1, wherein: in step S2, randomly initializing user, behavior, and commodity storage component parameters during first training, and embedding the input user, behavior, and commodity information into the user, behavior, and commodity storage components, respectively, where the user storage component is M, the behavior storage component is H, and the commodity storage component is E; wherein the user's preferences are stored in a user memory component slice muThe characteristics of behavior b are stored in a behavior memory component slice hbIn the commodity i, the characteristics of the commodity i are stored in e of the commodity memory sliceiIn the initial training, the feature vectors in each memory component are randomly initialized to obtain an embedded representation in its specific behavior for a specific commodity and a specific user.
7. The method for constructing the commodity recommendation learning model based on tensor decomposition and collaborative filtering as claimed in claim 6, wherein: in step S3, a dropout method is used to randomly discard part of the input users, and according to CP decomposition, a method of adding the user, behavior, and commodity embedded expressions after element-by-element multiplication is used to calculate similarity scores of the user and the commodity under different behaviors; a square loss function is used for model training.
8. The method for constructing the commodity recommendation learning model based on tensor decomposition and collaborative filtering as claimed in claim 7, wherein: in step S3, the calculation of the loss function is split into a user, behavior, commodity triple and all triples with historical behavior records, and the model training process is accelerated by discarding constant terms that are independent of gradient and changing the accumulation and calculation order.
9. The method for constructing the commodity recommendation learning model based on tensor decomposition and collaborative filtering as claimed in claim 1, wherein: in step S4, a multitask learning framework is used for joint learning, and a model is adjusted and trained, specifically including the following steps: the interaction relation among the various behaviors is learned and adjusted through a multi-task learning framework, and a data set, the iteration times and model parameters are adjusted according to the recommendation result; after adjustment, the model is formally trained.
10. A computer-readable storage medium, characterized in that a computer program is stored which, when executed, realizes the steps of the method according to any one of claims 1 to 9.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111373959.6A CN113988951A (en) | 2021-11-19 | 2021-11-19 | Commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111373959.6A CN113988951A (en) | 2021-11-19 | 2021-11-19 | Commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113988951A true CN113988951A (en) | 2022-01-28 |
Family
ID=79749510
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111373959.6A Pending CN113988951A (en) | 2021-11-19 | 2021-11-19 | Commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113988951A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114385619A (en) * | 2022-03-23 | 2022-04-22 | 山东省计算中心(国家超级计算济南中心) | Multi-channel ocean observation time sequence scalar data missing value prediction method and system |
CN114723591A (en) * | 2022-04-13 | 2022-07-08 | 北京邮电大学 | Education recommendation method and system based on incremental tensor Tucker decomposition |
-
2021
- 2021-11-19 CN CN202111373959.6A patent/CN113988951A/en active Pending
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114385619A (en) * | 2022-03-23 | 2022-04-22 | 山东省计算中心(国家超级计算济南中心) | Multi-channel ocean observation time sequence scalar data missing value prediction method and system |
CN114723591A (en) * | 2022-04-13 | 2022-07-08 | 北京邮电大学 | Education recommendation method and system based on incremental tensor Tucker decomposition |
CN114723591B (en) * | 2022-04-13 | 2023-10-20 | 北京邮电大学 | Education recommendation method and system based on incremental tensor Tucker decomposition |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109299396B (en) | Convolutional neural network collaborative filtering recommendation method and system fusing attention model | |
CN108648049B (en) | Sequence recommendation method based on user behavior difference modeling | |
CN107563841B (en) | Recommendation system based on user score decomposition | |
CN108647996B (en) | Spark-based personalized recommendation method and system | |
CN109087178B (en) | Commodity recommendation method and device | |
CN110781409B (en) | Article recommendation method based on collaborative filtering | |
CN111797321A (en) | Personalized knowledge recommendation method and system for different scenes | |
CN106447463A (en) | Commodity recommendation method based on Markov decision-making process model | |
CN113988951A (en) | Commodity recommendation learning model construction method based on tensor decomposition and collaborative filtering | |
CN113158024B (en) | Causal reasoning method for correcting popularity deviation of recommendation system | |
CN111949887A (en) | Item recommendation method and device and computer-readable storage medium | |
CN113159892B (en) | Commodity recommendation method based on multi-mode commodity feature fusion | |
CN112085525A (en) | User network purchasing behavior prediction research method based on hybrid model | |
CN109214882B (en) | Commodity recommendation method | |
CN110781401A (en) | Top-n project recommendation method based on collaborative autoregressive flow | |
CN112613953A (en) | Commodity selection method, system and computer readable storage medium | |
CN110348906A (en) | A kind of improvement Method of Commodity Recommendation based on polymorphic type implicit feedback | |
CN115204985A (en) | Shopping behavior prediction method, device, equipment and storage medium | |
CN114861050A (en) | Feature fusion recommendation method and system based on neural network | |
CN112800207A (en) | Commodity information recommendation method and device and storage medium | |
CN111177581A (en) | Multi-platform-based social e-commerce website commodity recommendation method and device | |
CN114840745A (en) | Personalized recommendation method and system based on graph feature learning and deep semantic matching model | |
Purkaystha et al. | Rating prediction for recommendation: Constructing user profiles and item characteristics using backpropagation | |
CN111274791B (en) | Modeling method of user loss early warning model in online home decoration scene | |
CN110717103B (en) | Improved collaborative filtering method based on stack noise reduction encoder |
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 |