CN115222486B - Article recommendation model training method, article recommendation method, device and storage medium - Google Patents
Article recommendation model training method, article recommendation method, device and storage medium Download PDFInfo
- Publication number
- CN115222486B CN115222486B CN202210906295.3A CN202210906295A CN115222486B CN 115222486 B CN115222486 B CN 115222486B CN 202210906295 A CN202210906295 A CN 202210906295A CN 115222486 B CN115222486 B CN 115222486B
- Authority
- CN
- China
- Prior art keywords
- historical
- information
- article
- comment
- embedded vector
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 83
- 238000012549 training Methods 0.000 title claims abstract description 77
- 239000013598 vector Substances 0.000 claims abstract description 169
- 230000006870 function Effects 0.000 claims description 122
- 238000004891 communication Methods 0.000 claims description 13
- 238000011156 evaluation Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 230000010365 information processing Effects 0.000 claims description 2
- 238000013473 artificial intelligence Methods 0.000 abstract description 8
- 230000008569 process Effects 0.000 description 9
- 238000002372 labelling Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000036651 mood Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000008451 emotion Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012552 review Methods 0.000 description 2
- 238000000889 atomisation Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
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
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the application provides an article recommendation model training method, an article recommendation device and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring a plurality of historical user information, historical article information, historical comment information of the historical article information, a real article label and a real scoring label, inputting a preset prediction model to obtain a historical user embedded vector, a historical article embedded vector and a historical comment embedded vector, and further determining a point-by-point loss function; determining predictive score information according to the historical user embedded vector and the historical comment embedded vector, and determining a pair-wise loss function according to the predictive score information and the real score label; and training the prediction model based on the point-by-point loss function and the paired loss function to obtain an article recommendation model. According to the method and the device for recommending the goods, the goods recommending model can be prevented from being influenced by semantic deviation of comment information, accuracy of prediction scoring information is guaranteed, and accuracy of goods recommending is improved.
Description
Technical Field
The present disclosure relates to the field of artificial intelligence, and in particular, to an article recommendation model training method, an article recommendation method, an apparatus, and a storage medium.
Background
After purchasing the articles, the user usually reviews the articles; at present, in an article recommendation system, integrating each piece of comment information of all users into the same document, determining a scoring standard through the document, scoring each piece of comment information according to the scoring standard, further determining the overall score of each article, and recommending articles with high scores to the users; however, when different users have different word habits and the mood states of the same user are different, the obtained comment information has semantic deviation, so that the scoring standard of the comment information is influenced, and the accuracy of article recommendation is low.
Disclosure of Invention
The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.
The embodiment of the application provides an article recommendation model training method, an article recommendation device and a storage medium, which can improve the recommendation accuracy of an article recommendation system.
To achieve the above object, a first aspect of an embodiment of the present application provides an item recommendation model training method, including: acquiring a plurality of historical user information, historical article information, historical comment information of the historical article information, a real article label and a real score label, wherein the historical comment information is matched with the historical user information, the real article label is matched with the historical article information, and the real score label is matched with the historical comment information; inputting the historical user information, any two pieces of historical article information and any two pieces of historical comment information of the historical article information into a preset prediction model to obtain a historical user embedded vector, a historical article embedded vector and a historical comment embedded vector; determining a point-by-point loss function according to the historical user embedded vector, the historical article embedded vector and the real article tag; determining prediction score information according to the historical user embedded vector and the historical comment embedded vector, and determining a pair-wise loss function according to the prediction score information and the real score label; and training the prediction model based on the point-by-point loss function and the paired loss function to obtain an article recommendation model.
In some embodiments, the predictive model includes a user network, an item network, a comment origin network, and a comment momentum network, wherein the comment momentum network is the same network structure as the comment origin network.
In some embodiments, the historical comment embedded vector includes a first historical comment pair embedded vector and a second historical comment pair embedded vector; inputting the historical user information, any two of the historical item information and any two of the historical comment information of the historical item information into a preset prediction model to obtain a historical user embedded vector, a historical item embedded vector and a historical comment embedded vector, wherein the method comprises the following steps of: inputting the historical user information into the user network to obtain a historical user embedded vector; inputting any two pieces of historical article information into the article network to obtain two historical article embedded vectors; inputting the history comment information of any two pieces of history article information into the evaluation original network to obtain the first history comment pair embedded vector; and inputting the history comment information of any two pieces of history article information into the comment quantity network to obtain the second history comment pair embedded vector.
In some embodiments, the training the prediction model based on the point-by-point loss function and the pair-wise loss function to obtain an item recommendation model includes: determining a model total loss function according to the point-by-point loss function and the paired loss functions; updating model parameters of the historical user network, the item network and the evaluation original network according to the model total loss function; determining a momentum update function according to the updated evaluation original network and a preset momentum update coefficient; and updating model parameters of the assessment momentum network according to the momentum updating function to obtain an article recommendation model.
In some embodiments, the point-wise loss function is formulated as: wherein L is 1 For the point-wise loss function, y ij For the real article label corresponding to the ith historical user information and the jth historical article information, p ij The predicted article information corresponding to the i-th historical user information and the j-th historical article information is obtained, and n is the number of all the historical article information; p is p ij =u i ·v j Wherein u is i Embedding vectors for the i-th historical user corresponding to the historical user information, v j Embedding vectors for the history articles corresponding to the j-th history article information; the formula of the pair-wise loss function is:
wherein L is 2 For the pair-wise loss function, y is For the real scoring tag corresponding to the ith historical user information and the s th historical comment information, y it R 'for the real scoring tag corresponding to the ith historical user information and the t historical comment information' s Embedding vectors for the second historical comment pair corresponding to the s-th historical comment information, and r '' t Embedding vectors for the second historical comment pairs corresponding to the t-th historical comment information, wherein m is the number of all the historical user information, and n is the number of all the historical comment information; when y is is <y it ,I(y is <y it ) =1, otherwise I (y is <y it 0=0;max(0,u i ·r′ s -u i ·r′ t ) For determining 0 and u i ·r′ s -u i ·r′ t Maximum value of (2); the formula of the model total loss function is as follows: l (L) total =L 1 +λ 1 L 2 +λ 2 L reg Wherein L is total Lambda is the total loss function of the model 1 And lambda (lambda) 2 Is a preset super parameter L reg Is a regularization term; the formula of the regularization term is: />Wherein Θ is k K is the number of all model parameters in the predictive model, which is the kth model parameter in the predictive model.
In some embodiments, the momentum update function is formulated as: w (w) m =δw m ′+(1-δ)w v Wherein w is m Is saidCommenting on model parameters of a momentum network, wherein delta is the momentum update coefficient, and w m ' model parameters for the theoretical momentum network before update, w v And evaluating the model parameters of the original network for updating.
To achieve the above object, a second aspect of the embodiments of the present application proposes an item recommendation method, including: acquiring target user information, a plurality of target article information and target comment information of the target article information, and inputting the target user information, the target article information and the target comment information into an article recommendation model to obtain prediction score information of each target article information, wherein the article recommendation model is trained by the article recommendation model training method according to the first aspect; item recommendation information is determined among the plurality of target item information based on the predictive scoring information.
To achieve the above object, a third aspect of the embodiments of the present application proposes an article recommendation model training apparatus, the apparatus including: the information processing device comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring a plurality of pieces of historical user information, historical article information, historical comment information of the historical article information, real article labels and real grading labels, wherein the historical comment information is matched with the historical user information, the real article labels are matched with the historical article information, and the real grading labels are matched with the historical comment information; the input unit is used for inputting the historical user information, any two pieces of historical article information and any two pieces of historical comment information of the historical article information into a preset prediction model to obtain a historical user embedded vector, a historical article embedded vector and a historical comment embedded vector; a first determining unit configured to determine a point-by-point loss function according to the historical user embedded vector, the historical article embedded vector, and the real article tag; the second determining unit is used for determining prediction score information according to the historical user embedded vector and the historical comment embedded vector and determining a pair-wise loss function according to the prediction score information and the real score label; and the training unit is used for training the prediction model based on the point-by-point loss function and the paired loss function to obtain an article recommendation model.
To achieve the above object, a fourth aspect of the embodiments of the present application proposes an electronic device, the electronic device including a memory, a processor, a program stored on the memory and executable on the processor, and a data bus for implementing connection communication between the processor and the memory, the program implementing the method for training an article recommendation model according to the first aspect when executed by the processor.
To achieve the above object, a fifth aspect of the embodiments of the present application proposes a storage medium, which is a computer-readable storage medium, for computer-readable storage, the storage medium storing one or more programs, the one or more programs being executable by one or more processors to implement the method for training an item recommendation model according to the first aspect, or the method for recommending an item according to the second aspect.
The method, the device and the storage medium for training the article recommendation model provided by the application comprise the following steps: acquiring a plurality of historical user information, historical article information, historical comment information of the historical article information, a real article label and a real score label, wherein the historical comment information is matched with the historical user information, the real article label is matched with the historical article information, and the real score label is matched with the historical comment information; inputting the historical user information, any two pieces of historical article information and any two pieces of historical comment information of the historical article information into a preset prediction model to obtain a historical user embedded vector, a historical article embedded vector and a historical comment embedded vector; determining a point-by-point loss function according to the historical user embedded vector, the historical article embedded vector and the real article tag; determining prediction score information according to the historical user embedded vector and the historical comment embedded vector, and determining a pair-wise loss function according to the prediction score information and the real score label; and training the prediction model based on the point-by-point loss function and the paired loss function to obtain an article recommendation model. According to the scheme provided by the embodiment of the application, the historical user information, the historical article information and the historical comment information are used as training data, the training data are input into the prediction model, then the historical user embedded vector, the historical article embedded vector and the historical comment embedded vector are obtained, the real article label is combined to determine the point-by-point loss function, then the prediction score information is determined through the historical user embedded vector and the historical comment embedded vector, and the real score label is combined to determine the paired loss function, so that the prediction model is trained by using the point-by-point loss function, the article recommendation model can be used for accurately predicting user preference, the paired loss function is used for training the prediction model, the article recommendation model can be prevented from being influenced by semantic deviation of comment information, accuracy of the prediction score information is guaranteed, and further article recommendation accuracy is improved.
Additional features and advantages of the application will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the application. The objectives and other advantages of the application will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The accompanying drawings are included to provide a further understanding of the technical aspects of the present application, and are incorporated in and constitute a part of this specification, illustrate the technical aspects of the present application and together with the examples of the present application, and not constitute a limitation of the technical aspects of the present application.
FIG. 1 is a flow chart of an item recommendation model training method provided in one embodiment of the present application;
FIG. 2 is a flow chart of a predictive model information entry process provided in accordance with another embodiment of the present application;
FIG. 3 is a flow chart of obtaining a recommendation model for an item according to another embodiment of the present application;
FIG. 4 is a flow chart of an item recommendation method provided in another embodiment of the present application;
FIG. 5 is a system block diagram of an item recommendation model provided in accordance with another embodiment of the present application;
FIG. 6 is a schematic diagram of an article recommendation model training apparatus according to another embodiment of the present application;
Fig. 7 is a schematic hardware structure of an electronic device according to another embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
In the description of the present application, the meaning of a number is one or more, the meaning of a number is two or more, and greater than, less than, exceeding, etc. are understood to exclude the present number, and the meaning of above, below, within, etc. are understood to include the present number.
It should be noted that although functional block division is performed in a device diagram and a logic sequence is shown in a flowchart, in some cases, the steps shown or described may be performed in a different order than the block division in the device, or in the flowchart. The terms first, second and the like in the description, in the claims and in the above-described figures, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order.
First, several nouns referred to in this application are parsed:
Artificial intelligence (Artificial Intelligence, AI): is a new technical science for researching and developing theories, methods, technologies and application systems for simulating, extending and expanding the intelligence of people; artificial intelligence is a branch of computer science that attempts to understand the nature of intelligence and to produce a new intelligent machine that can react in a manner similar to human intelligence, research in this field including robotics, language recognition, image recognition, natural language processing, and expert systems. Artificial intelligence can simulate the information process of consciousness and thinking of people. Artificial intelligence is also a theory, method, technique, and application system that utilizes a digital computer or digital computer-controlled machine to simulate, extend, and expand human intelligence, sense the environment, acquire knowledge, and use knowledge to obtain optimal results.
After purchasing the articles, the user usually reviews the articles; at present, in an article recommendation system, integrating each piece of comment information of all users into the same document, determining a scoring standard through the document, scoring each piece of comment information according to the scoring standard, further determining the overall score of each article, and recommending articles with high scores to the users; however, when different users have different word habits and the mood states of the same user are different, the obtained comment information has semantic deviation, so that the scoring standard of the comment information is influenced, and the accuracy of article recommendation is low.
Aiming at the problem that comment information has semantic deviation and low accuracy of article recommendation, the application provides an article recommendation model training method, an article recommendation device and a storage medium, wherein the method comprises the following steps: acquiring a plurality of historical user information, historical article information, historical comment information of the historical article information, a real article label and a real grading label, wherein the historical comment information is matched with the historical user information, the real article label is matched with the historical article information, and the real grading label is matched with the historical comment information; inputting the historical user information, any two pieces of historical article information and any two pieces of historical comment information of the historical article information into a preset prediction model to obtain a historical user embedded vector, a historical article embedded vector and a historical comment embedded vector; determining a point-by-point loss function according to the historical user embedded vector, the historical article embedded vector and the real article label; determining predictive score information according to the historical user embedded vector and the historical comment embedded vector, and determining a pair-wise loss function according to the predictive score information and the real score label; and training the prediction model based on the point-by-point loss function and the paired loss function to obtain an article recommendation model. According to the scheme provided by the embodiment of the application, the historical user information, the historical article information and the historical comment information are used as training data, the training data are input into the prediction model, then the historical user embedded vector, the historical article embedded vector and the historical comment embedded vector are obtained, the real article label is combined to determine the point-by-point loss function, then the prediction score information is determined through the historical user embedded vector and the historical comment embedded vector, and the real score label is combined to determine the paired loss function, so that the prediction model is trained by using the point-by-point loss function, the article recommendation model can be used for accurately predicting user preference, the paired loss function is used for training the prediction model, the article recommendation model can be prevented from being influenced by semantic deviation of comment information, accuracy of the prediction score information is guaranteed, and further article recommendation accuracy is improved.
The method, the device and the storage medium for training the article recommendation model provided by the embodiment of the application are specifically described through the following embodiments, and the method for training the article recommendation model in the embodiment of the application is described first.
The embodiment of the application provides an article recommendation model training method, and relates to the technical field of artificial intelligence. The method for training the article recommendation model can be applied to a terminal, a server side and software running in the terminal or the server side. In some embodiments, the terminal may be a smart phone, tablet, notebook, desktop, etc.; the server side can be configured as an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, and a cloud server for providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, basic cloud computing services such as big data and artificial intelligent platforms and the like; the software may be an application or the like that implements the item recommendation model training method, but is not limited to the above form.
The subject application is operational with numerous general purpose or special purpose computer system environments or configurations. For example: personal computers, server computers, hand-held or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments that include any of the above systems or devices, and the like. The application may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
In the embodiments of the present application, when related processing is required according to user information, user behavior data, user history data, user location information, and other data related to user identity or characteristics, permission or consent of the user is obtained first, and the collection, use, processing, and the like of the data comply with related laws and regulations and standards of related countries and regions. In addition, when the embodiment of the application needs to acquire the sensitive personal information of the user, the independent permission or independent consent of the user is acquired through a popup window or a jump to a confirmation page or the like, and after the independent permission or independent consent of the user is explicitly acquired, necessary user related data for enabling the embodiment of the application to normally operate is acquired.
Embodiments of the present application are further described below with reference to the accompanying drawings.
Referring to fig. 1, fig. 1 is a flowchart of a method for training an item recommendation model according to an embodiment of the present application. The method for training the article recommendation model comprises the following steps:
step S110, a plurality of historical user information, historical article information, historical comment information of the historical article information, a real article label and a real grading label are obtained, wherein the historical comment information is matched with the historical user information, the real article label is matched with the historical article information, and the real grading label is matched with the historical comment information;
step S120, inputting the historical user information, any two pieces of historical article information and any two pieces of historical comment information into a preset prediction model to obtain a historical user embedded vector, a historical article embedded vector and a historical comment embedded vector;
step S130, determining a point-by-point loss function according to the historical user embedded vector, the historical article embedded vector and the real article label;
step S140, predictive score information is determined according to the historical user embedded vector and the historical comment embedded vector, and a pair-wise loss function is determined according to the predictive score information and the real score label;
And step S150, training the prediction model based on the point-by-point loss function and the paired loss function to obtain an article recommendation model.
It can be understood that in the training process, the training data set is acquired first, the training data set is determined by the comment record of the user, when the user purchases an article and comments on the article, the user information, the article information purchased by the user and the comment information posted by the user on the article are recorded in the background database, so that the history user information, the history article information and the history comment information can be acquired through the background database; through the historical comment information, the emotion of the user on the object can be analyzed, for example, when a television is purchased, the user can make comments of clear images, intelligent system and no atomization phenomenon, the user can analyze the emotion of the user on the television, and the attribute of the television satisfied by the user can be known, so that the historical comment information can be used as auxiliary information, and the accuracy of object recommendation is improved; therefore, training is carried out aiming at each historical user information, and a real article label is combined to determine a point-by-point loss function, so that article preference of each user can be predicted better; when different users have different word habits and when the same user has different mood states, the posted comment information has semantic deviation, so that the relation between the comment words and the score is not absolute, the same comment possibly corresponds to different scores, and the influence of the semantic deviation can be avoided by combining a real score label to determine a pair loss function, so that the accuracy of article recommendation is improved; based on the method, historical user information, historical article information and historical comment information are used as training data, the training data are input into a prediction model, then a historical user embedded vector, a historical article embedded vector and a historical comment embedded vector are obtained, a point-by-point loss function is determined by combining a real article label, then prediction score information is determined by combining the historical user embedded vector and the historical comment embedded vector, and a pair-by-point loss function is determined by combining the real score label, so that the prediction model is trained by utilizing the point-by-point loss function, more accurate prediction user preference of the article recommendation model can be achieved, the prediction model is trained by utilizing the pair-by-point loss function, the article recommendation model is prevented from being influenced by semantic deviation of comment information, accuracy of the prediction score information is guaranteed, and article recommendation accuracy is improved.
It should be noted that, in the training process, the value of the real article label is 0 or 1, shopping records of all users are mixed together, namely, all historical user information is taken as one data set, all historical article information is taken as another data set, any one historical user information and one historical article information are taken, when the user purchases the article, the value of the real article label is 1, and when the user is satisfied with purchasing the article, the value of the real article label is 0; the real scoring tag refers to the real scoring value of the historical comment information and refers to the scoring value which is matched with the target comment published by the target user on the target object; the real object labels and the real scoring labels are preset manually according to actual conditions, and the accuracy of model training can be guaranteed.
It should be noted that, because comment information is introduced, user data received by each application program can be utilized to the greatest extent, and meanwhile, the comment information can also be used for cross-domain recommendation.
Notably, through combining the article embedded vector and the comment embedded vector with the user information for training, the advantages of the article embedded vector and the comment embedded vector can be integrated, so that the generated comment embedded vector can more accurately reflect the true mood of the user when the user publishes the corresponding comment.
In addition, in an embodiment, the predictive model includes a user network, an item network, a comment origin network, and a comment momentum network, wherein the comment momentum network is the same network structure as the comment origin network.
It can be understood that the user network, the article network, the comment original network and the comment momentum network can be BERT network models, the BERT network models are encoders based on a transducer model, the user network can convert input historical user information into historical user embedded vectors with specific dimensions, the article network can convert the input historical article information into the historical article embedded vectors with specific dimensions, and the comment original network and the comment momentum network can convert the input historical comment information into the historical comment embedded vectors with specific dimensions.
It should be noted that, the network structure of the article recommendation model is limited, and the article recommendation model has a network structure of three layers and four towers, so that accuracy of prediction scoring information can be ensured, and further accuracy of article recommendation is improved.
Additionally, referring to FIG. 2, in one embodiment, the historical comment embedded vector includes a first historical comment pair embedded vector and a second historical comment pair embedded vector; step S120 in the embodiment shown in fig. 1 includes, but is not limited to, the following steps:
Step S210, inputting the history user information into a user network to obtain a history user embedded vector;
step S220, inputting any two pieces of historical article information into an article network to obtain two historical article embedded vectors;
step S230, inputting the history comment information of any two pieces of history article information into a comment original network to obtain a first history comment pair embedded vector;
and step S240, inputting the history comment information of any two pieces of history article information into a comment momentum network to obtain a second history comment pair embedded vector.
It can be understood that in the training process of the model, repeated iterative updating is required to be performed on the prediction model until the prediction model meets the preset model requirement or reaches the preset iterative times; in an iterative process, when training data is required to be determined, any one of a plurality of pieces of historical user information is taken, the historical user information is input into a user network, and an embedded vector of a historical user can be obtained; then, taking two pieces of historical article information, and inputting the two pieces of historical article information into an article network to obtain two pieces of historical article embedded vectors; inputting the corresponding two pieces of history comment information into a comment original network to obtain a first history comment pair embedded vector, wherein the first history comment pair embedded vector comprises two embedded vectors; finally, inputting the corresponding two pieces of history comment information into a comment momentum network to obtain a second history comment pair embedded vector, wherein the second history comment pair embedded vector comprises two embedded vectors; in the current iteration process, a user embedded vector, two historical item embedded vectors, a first historical comment pair embedded vector and a second historical comment pair embedded vector are used for updating model parameters.
In addition, referring to fig. 3, in an embodiment, step S150 in the embodiment shown in fig. 1 includes, but is not limited to, the following steps:
step S310, determining a model total loss function according to the point-by-point loss function and the paired loss function;
step S320, updating model parameters of the historical user network, the article network and the comment original network according to the model total loss function;
step S330, determining a momentum update function according to the updated comment original network and a preset momentum update coefficient;
and step S340, updating model parameters of the comment momentum network according to the momentum update function to obtain an article recommendation model.
It will be appreciated that when model training is performed using only point-by-point loss functions, there will be labeling bias, which means that different judgment results are obtained for the same comment, for example, for the same user, in the items compared in pairs, there is a possibility that the comment of the user to item a is of interest to 0.9 for the first set of records, 0.6 for the comment of the user to item B is of interest to 0.6 for the second set of records, and 0.2 for the comment of the user to item C is of interest; in the first set of records, item A is a positive sample and item B is a negative sample; in the second set of records, item B is a positive sample and item C is a negative sample; for the article B, the two judging results are obviously different, and labeling deviation exists, so that the two iterative updating directions are opposite, and the updating speed and the updating precision of the model are influenced; the model training is carried out by setting the pair samples and utilizing the pair loss function, so that the labeling deviation can be reduced, and compared with optimizing the comment network through gradient back propagation, the comment original network and the comment momentum network are set, and the comment momentum network is optimized through momentum update, so that the update of the comment momentum network and the comment original network which are consistent and slowly evolve is realized, the influence of the sample with the labeling deviation on the whole model is reduced, the severe fluctuation and even abrupt change of the model parameters are prevented, and the problem of the labeling deviation can be effectively prevented.
In addition, in one embodiment, the point-wise loss function is formulated as:
wherein L is 1 To be a point-by-point loss function, y ij For the real article label corresponding to the ith historical user information and the jth historical article information, p ij The method comprises the steps that predicted article information corresponding to the ith historical user information and the jth historical article information is obtained, and n is the number of all the historical article information;
p ij =u i ·v j ,
wherein u is i Embedding vectors for the history user corresponding to the ith history user information, v j Embedding vectors for the history articles corresponding to the jth history article information;
the formula for the pair-wise loss function is:
wherein L is 2 As a pair-wise loss function, y is True scoring tags corresponding to the ith historical user information and the ith historical comment information, y it The true scoring label corresponding to the ith historical user information and the ith historical comment information, and r '' s Embedding vectors for a second historical comment pair corresponding to the s-th historical comment information, and r' t Embedding vectors for second historical comment pairs corresponding to the t-th historical comment information, wherein m is the number of all historical user information, and n is the number of all historical comment information; when y is is <y it ,I(y is <y it ) =1, otherwise I (y is <y it )=0;max(0,u i ·r′ s -u i ·r′ t ) For determining 0 and u i ·r′ s -u i ·r′ t Maximum value of (2);
the formula of the model total loss function is:
L total =L 1 +λ 1 L 2 +λ 2 L reg ,
Wherein L is total Lambda is the model total loss function 1 And lambda (lambda) 2 Is a preset super parameter L reg Is a regularization term;
the formula of the regularization term is:
wherein Θ is k For the kth model parameter in the predictive model, K is the number of all model parameters in the predictive model.
It will be appreciated that p, relative to the genuine article tag ij A probability value of 1 for predicting an item tag of the item; the s-th historical comment information corresponds to the article s, and the t-th historical comment information corresponds to the article t; u (u) i ·r′ s Refers to the product of the embedded vector of the historical user corresponding to the ith historical user information and the embedded vector of the second historical comment corresponding to the ith historical comment information,i i ·r′ s Predictive scoring for characterizing items s, additionally, u i ·r′ t Refers to the product of the embedded vector of the history user corresponding to the ith history user information and the embedded vector of the second history comment corresponding to the ith history comment information, u i ·r′ t A predictive score for characterizing item t.
The point-by-point loss function and the paired loss function are determined first, and then the total loss function of the model is determined, so that the reliability of model training is ensured.
It should be noted that, the regularization term refers to L2 regularization, and by setting the regularization term, the overfitting can be reduced, because it can attenuate the weight, and the reason of overfitting is generally because the assumed function considers each point in the sample, the function formed finally fluctuates greatly, the function value will change drastically due to slight change of the independent variable, and the result generally shows that the accuracy rate on the training set is high, and the accuracy rate on the test set is low. The reason why the fluctuation is large from the structural point of view of the function is that the weight (constant) in the function is too large, and if the weight can be reduced, the fluctuation can be reduced, and the over-fitting condition can be reduced to some extent. It is therefore necessary to attach to the model a canonical penalty (limiting the two norms of the model population parameters below a certain value) that is used to constrain the model population parameters to prevent overfitting.
Lambda is the sum of the values of lambda 1 And lambda (lambda) 2 The super-parameters for controlling the duty ratio of the loss function of each part are empirical constants obtained by repeated experiment tuning based on grid search.
In addition, in one embodiment, the formula of the momentum update function is:
w m =δw m ′+(1-δ)w v ,
wherein w is m For commenting on model parameters of a momentum network, delta is a momentum update coefficient, w m ' model parameters for commenting on momentum network before update, w v And (5) the model parameters of the original network are reviewed after updating.
It will be appreciated that the model has been updated for multiple roundsJourney, w m The model parameters of the comment momentum network are the comment momentum network E updated in the round m Model parameters, w m ' is the model parameter of the comment momentum network before updating, namely E after the last round of updating m Parameters of (2); w (w) v The model parameters of the comment original network after updating are the comment original network E after the current round of updating v Is a parameter of (a).
In specific practice, δ may be set to 0.99 in order to comment on the efficient update of the momentum network.
As shown in fig. 4, fig. 4 is a flowchart of an item recommendation method according to an embodiment of the present application. The item recommendation method includes, but is not limited to, the steps of:
step S410, obtaining target user information, a plurality of target item information and target comment information of the target item information, and inputting the target user information, the target item information and the target comment information into an item recommendation model to obtain prediction score information of each target item information, wherein the item recommendation model is trained by the item recommendation model training method;
Step S420, determining item recommendation information among the plurality of target item information based on the prediction score information.
It can be understood that the article recommendation model mentioned in the article recommendation method is trained by the article recommendation model training method, so that accuracy of the prediction scoring information is high, and accuracy of article recommendation can be guaranteed.
It should be noted that, for the target user information, the prediction score information corresponding to each target item information can be obtained through the item recommendation model, and then the item with high prediction score is determined through comparison, so as to obtain the item recommendation information.
In addition, referring to fig. 5, fig. 5 is a system block diagram of an item recommendation model according to an embodiment of the present application.
It can be understood that the article recommendation model has a three-layer four-tower network structure, comment information and article information are introduced to be trained together, and the comment information and the article information are used as auxiliary information to improve the recommendation effect; training in a modelIn the training process, aiming at certain historical user information, the historical user information is input into a user network E u The historical user embedded vector u can be obtained; then two pieces of historical article information are taken and input into an article network E i Obtaining two historical object embedded vectors v 1 And v 2 The method comprises the steps of carrying out a first treatment on the surface of the Then inputting the corresponding two pieces of history comment information into a comment original network E v Obtaining a first historical comment pair embedded vector r 1 And r 2 The method comprises the steps of carrying out a first treatment on the surface of the Finally, inputting the corresponding two pieces of history comment information into a comment momentum network E m Obtaining a second historical comment pair embedded vector r' 1 And r' 2 Updating comment momentum network E by momentum update m Is used for the model parameters of the model.
E is also described as m And E is v The network structures of (a) are the same and include but are not limited to: attention layer, convolution layer and max-pooling layer.
In addition, referring to fig. 6, the present application further provides an article recommendation model training apparatus 600, including:
an obtaining unit 610, configured to obtain a plurality of historical user information, historical item information, historical comment information of the historical item information, a real item tag, and a real score tag, where the historical comment information is matched with the historical user information, the real item tag is matched with the historical item information, and the real score tag is matched with the historical comment information;
an input unit 620, configured to input historical user information, any two pieces of historical item information, and any two pieces of historical comment information into a preset prediction model, to obtain a historical user embedded vector, a historical item embedded vector, and a historical comment embedded vector;
A first determining unit 630, configured to determine a point-by-point loss function according to the historical user embedded vector, the historical article embedded vector, and the real article tag;
a second determining unit 640 for determining prediction score information according to the historical user embedding vector and the historical comment embedding vector, and determining a pair-wise loss function according to the prediction score information and the true score label;
the training unit 650 is configured to train the prediction model based on the point-by-point loss function and the pair-wise loss function, and obtain an item recommendation model.
It can be appreciated that the specific implementation of the article recommendation model training apparatus 600 is substantially the same as the specific example of the article recommendation model training method described above, and will not be repeated here; based on the method, historical user information, historical article information and historical comment information are used as training data, the training data are input into a prediction model, then a historical user embedded vector, a historical article embedded vector and a historical comment embedded vector are obtained, a point-by-point loss function is determined by combining a real article label, then prediction score information is determined by combining the historical user embedded vector and the historical comment embedded vector, and a pair-by-point loss function is determined by combining the real score label, so that the prediction model is trained by utilizing the point-by-point loss function, more accurate prediction user preference of the article recommendation model can be achieved, the prediction model is trained by utilizing the pair-by-point loss function, the article recommendation model is prevented from being influenced by semantic deviation of comment information, accuracy of the prediction score information is guaranteed, and article recommendation accuracy is improved.
In addition, the application also provides an article recommending device, which comprises:
the prediction unit is used for acquiring target user information, a plurality of target article information and target comment information of the target article information, inputting the target user information, the target article information and the target comment information into the article recommendation model, and obtaining prediction score information of each target article information;
a recommendation unit configured to determine item recommendation information among the plurality of target item information based on the prediction score information;
the article recommendation model is trained by the article recommendation model training method.
It is understood that the specific embodiment of the article recommendation device is substantially the same as the specific embodiment of the article recommendation method described above, and will not be described herein.
In addition, referring to fig. 7, fig. 7 illustrates a hardware structure of an electronic device of another embodiment, the electronic device including:
the processor 701 may be implemented by a general-purpose CPU (Central Processing Unit ), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, etc. for executing relevant programs to implement the technical solutions provided by the embodiments of the present application;
The Memory 702 may be implemented in the form of a Read Only Memory (ROM), a static storage device, a dynamic storage device, or a random access Memory (Random Access Memory, RAM). The memory 702 may store an operating system and other application programs, and when the technical solution provided in the embodiments of the present disclosure is implemented by software or firmware, relevant program codes are stored in the memory 702, and the processor 701 invokes the method for performing the article recommendation model training method of the embodiments of the present disclosure, for example, performing the method steps S110 to S150 in fig. 1, the method steps S210 to S240 in fig. 2, and the method steps S310 to S340 in fig. 3 described above;
an input/output interface 703 for implementing information input and output;
the communication interface 704 is configured to implement communication interaction between the device and other devices, and may implement communication in a wired manner (e.g. USB, network cable, etc.), or may implement communication in a wireless manner (e.g. mobile network, WIFI, bluetooth, etc.);
a bus 705 for transferring information between various components of the device (e.g., the processor 701, memory 702, input/output interfaces 703, and communication interfaces 704);
Wherein the processor 701, the memory 702, the input/output interface 703 and the communication interface 704 are in communication connection with each other inside the device via a bus 705.
The embodiment of the present application further provides a storage medium, which is a computer readable storage medium, for computer readable storage, where the storage medium stores one or more programs, and the one or more programs may be executed by the one or more processors to implement the above-described item recommendation model training method, for example, perform the above-described method steps S110 to S150 in fig. 1, the above-described method steps S210 to S240 in fig. 2, and the above-described method steps S310 to S340 in fig. 3, or implement the above-described item recommendation method, for example, perform the above-described method steps S410 to S420 in fig. 4.
The memory, as a non-transitory computer readable storage medium, may be used to store non-transitory software programs as well as non-transitory computer executable programs. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory optionally includes memory remotely located relative to the processor, the remote memory being connectable to the processor through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
According to the item recommendation model training method, the item recommendation method, the device and the storage medium, a plurality of historical user information, historical item information, historical comment information of the historical item information, a real item label and a real score label are obtained, wherein the historical comment information is matched with the historical user information, the real item label is matched with the historical item information, and the real score label is matched with the historical comment information; inputting the historical user information, any two pieces of historical article information and any two pieces of historical comment information of the historical article information into a preset prediction model to obtain a historical user embedded vector, a historical article embedded vector and a historical comment embedded vector; determining a point-by-point loss function according to the historical user embedded vector, the historical article embedded vector and the real article label; determining predictive score information according to the historical user embedded vector and the historical comment embedded vector, and determining a pair-wise loss function according to the predictive score information and the real score label; training the prediction model based on the point-by-point loss function and the paired loss function to obtain an article recommendation model; based on the method, historical user information, historical article information and historical comment information are used as training data, the training data are input into a prediction model, then a historical user embedded vector, a historical article embedded vector and a historical comment embedded vector are obtained, a point-by-point loss function is determined by combining a real article label, then prediction score information is determined by combining the historical user embedded vector and the historical comment embedded vector, and a pair-by-point loss function is determined by combining the real score label, so that the prediction model is trained by utilizing the point-by-point loss function, more accurate prediction user preference of the article recommendation model can be achieved, the prediction model is trained by utilizing the pair-by-point loss function, the article recommendation model is prevented from being influenced by semantic deviation of comment information, accuracy of the prediction score information is guaranteed, and article recommendation accuracy is improved.
The embodiments described in the embodiments of the present application are for more clearly describing the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application, and as those skilled in the art can know that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the embodiments of the present application are equally applicable to similar technical problems.
It will be appreciated by those skilled in the art that the solutions shown in fig. 1-4 are not limiting to embodiments of the present application, and may include more or fewer steps than illustrated, or may combine certain steps, or different steps.
The above described apparatus embodiments are merely illustrative, wherein the units illustrated as separate components may or may not be physically separate, i.e. may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
Those of ordinary skill in the art will appreciate that all or some of the steps of the methods, systems, functional modules/units in the devices disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof.
The terms "first," "second," "third," "fourth," and the like in the description of the present application and in the above-described figures, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that embodiments of the present application described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
It should be understood that in this application, "at least one" means one or more, and "a plurality" means two or more. "and/or" for describing the association relationship of the association object, the representation may have three relationships, for example, "a and/or B" may represent: only a, only B and both a and B are present, wherein a, B may be singular or plural. The character "/" generally indicates that the context-dependent object is an "or" relationship. "at least one of" or the like means any combination of these items, including any combination of single item(s) or plural items(s). For example, at least one (one) of a, b or c may represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b, c may be single or plural.
In the several embodiments provided in this application, it should be understood that the disclosed apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the above-described division of units is merely a logical function division, and there may be another division manner in actual implementation, for example, a plurality of units or components may be combined or may be 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 above as separate components may or may not be physically separate, and components shown as units may or may not be physical units, may be located in one place, or may be distributed over 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 each embodiment of the present application 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. The integrated units may be implemented in hardware or in software functional units.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including multiple instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of the various embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing a program.
Preferred embodiments of the present application are described above with reference to the accompanying drawings, and thus do not limit the scope of the claims of the embodiments of the present application. Any modifications, equivalent substitutions and improvements made by those skilled in the art without departing from the scope and spirit of the embodiments of the present application shall fall within the scope of the claims of the embodiments of the present application.
Claims (7)
1. A method of training a recommendation model for an item, the method comprising:
acquiring a plurality of historical user information, historical article information, historical comment information of the historical article information, a real article label and a real score label, wherein the historical comment information is matched with the historical user information, the real article label is matched with the historical article information, and the real score label is matched with the historical comment information;
inputting the historical user information, any two pieces of historical article information and any two pieces of historical comment information into a preset prediction model to obtain a historical user embedded vector, a historical article embedded vector and a historical comment embedded vector, wherein the prediction model comprises a user network, an article network, a comment original network and a comment momentum network, the comment momentum network has the same network structure as the comment original network, and the historical comment embedded vector comprises a first historical comment pair embedded vector and a second historical comment pair embedded vector;
determining a point-by-point loss function according to the historical user embedded vector, the historical article embedded vector and the real article tag;
Determining prediction score information according to the historical user embedded vector and the historical comment embedded vector, and determining a pair-wise loss function according to the prediction score information and the real score label;
training the prediction model based on the point-by-point loss function and the paired loss function to obtain an article recommendation model;
the step of inputting the historical user information, any two of the historical item information and any two of the historical comment information into a preset prediction model to obtain a historical user embedded vector, a historical item embedded vector and a historical comment embedded vector comprises the following steps:
inputting the historical user information into the user network to obtain a historical user embedded vector;
inputting any two pieces of historical article information into the article network to obtain two historical article embedded vectors;
inputting the history comment information of any two pieces of history article information into the evaluation original network to obtain the first history comment pair embedded vector;
inputting the history comment information of any two pieces of history article information into the comment quantity network to obtain the second history comment pair embedded vector;
The training of the prediction model based on the point-by-point loss function and the paired loss function to obtain an article recommendation model comprises the following steps:
determining a model total loss function according to the point-by-point loss function and the paired loss functions;
updating model parameters of the user network, the article network and the evaluation original network according to the model total loss function;
determining a momentum update function according to the updated evaluation original network and a preset momentum update coefficient;
and updating model parameters of the assessment momentum network according to the momentum updating function to obtain an article recommendation model.
2. The method of claim 1, wherein the point-wise loss function is formulated as:
wherein L is 1 For the point-wise loss function, y ij For the real article label corresponding to the ith historical user information and the jth historical article information, p ij The predicted article information corresponding to the i-th historical user information and the j-th historical article information is obtained, and n is the number of all the historical article information;
p ij =u i ·v j ,
wherein u is i Embedding the historical user corresponding to the i-th historical user information into the database Quantity, v j Embedding vectors for the history articles corresponding to the j-th history article information;
the formula of the pair-wise loss function is:
wherein L is 2 For the pair-wise loss function, y is For the real scoring tag corresponding to the ith historical user information and the s th historical comment information, y it R for the real scoring label corresponding to the ith historical user information and the tth historical comment information s ' being the second historical comment pair embedded vector corresponding to the s-th historical comment information, r t ' is the second historical comment pair embedded vector corresponding to the t-th historical comment information, m is the number of all the historical user information, and n is the number of all the historical comment information; when y is is <y it ,I(y is <y it ) =1, otherwise I (y is <y it )=0;max(0,u i ·r s ′-u i ·r t ') is used to determine 0 and u i ·r s ′-u i ·r t Maximum in';
the formula of the model total loss function is as follows:
L total =L 1 +λ 1 L 2 +λ 2 L reg ,
wherein L is total Lambda is the total loss function of the model 1 And lambda (lambda) t Is a preset super parameter L reg Is a regularization term;
the formula of the regularization term is:
wherein Θ is k For the kth model parameter in the predictive model, K isThe number of all model parameters in the predictive model.
3. The method of claim 1, wherein the momentum update function is formulated as:
w m =δw m ′+(1-δ)w v ,
Wherein w is m For the model parameters of the theoretical momentum network, delta is the momentum update coefficient, w m ' model parameters for the theoretical momentum network before update, w v And evaluating the model parameters of the original network for updating.
4. A method of recommending items, the method comprising:
acquiring target user information, a plurality of target item information and target comment information of the target item information, and inputting the target user information, the target item information and the target comment information into an item recommendation model to obtain prediction score information of each target item information, wherein the item recommendation model is trained by the item recommendation model training method according to any one of claims 1 to 3;
item recommendation information is determined among the plurality of target item information based on the predictive scoring information.
5. An article recommendation model training device, the device comprising:
the information processing device comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring a plurality of pieces of historical user information, historical article information, historical comment information of the historical article information, real article labels and real grading labels, wherein the historical comment information is matched with the historical user information, the real article labels are matched with the historical article information, and the real grading labels are matched with the historical comment information;
The input unit is used for inputting the historical user information, any two pieces of historical article information and any two pieces of historical comment information into a preset prediction model to obtain a historical user embedded vector, a historical article embedded vector and a historical comment embedded vector, wherein the prediction model comprises a user network, an article network, a comment original network and a comment momentum network, the comment momentum network has the same network structure as the comment original network, and the historical comment embedded vector comprises a first historical comment pair embedded vector and a second historical comment pair embedded vector;
a first determining unit configured to determine a point-by-point loss function according to the historical user embedded vector, the historical article embedded vector, and the real article tag;
the second determining unit is used for determining prediction score information according to the historical user embedded vector and the historical comment embedded vector and determining a pair-wise loss function according to the prediction score information and the real score label;
the training unit is used for training the prediction model based on the point-by-point loss function and the paired loss function to obtain an article recommendation model;
The step of inputting the historical user information, any two of the historical item information and any two of the historical comment information into a preset prediction model to obtain a historical user embedded vector, a historical item embedded vector and a historical comment embedded vector comprises the following steps:
inputting the historical user information into the user network to obtain a historical user embedded vector;
inputting any two pieces of historical article information into the article network to obtain two historical article embedded vectors;
inputting the history comment information of any two pieces of history article information into the evaluation original network to obtain the first history comment pair embedded vector;
inputting the history comment information of any two pieces of history article information into the comment quantity network to obtain the second history comment pair embedded vector;
the training of the prediction model based on the point-by-point loss function and the paired loss function to obtain an article recommendation model comprises the following steps:
determining a model total loss function according to the point-by-point loss function and the paired loss functions;
updating model parameters of the user network, the article network and the evaluation original network according to the model total loss function;
Determining a momentum update function according to the updated evaluation original network and a preset momentum update coefficient;
and updating model parameters of the assessment momentum network according to the momentum updating function to obtain an article recommendation model.
6. An electronic device comprising a memory, a processor, a program stored on the memory and executable on the processor, and a data bus for enabling a connection communication between the processor and the memory, the program when executed by the processor implementing the article recommendation model training method according to any one of claims 1 to 3.
7. A storage medium, which is a computer-readable storage medium, for computer-readable storage, characterized in that the storage medium stores one or more programs executable by one or more processors to implement the item recommendation model training method of any one of claims 1 to 3, or the item recommendation method of claim 4.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210906295.3A CN115222486B (en) | 2022-07-29 | 2022-07-29 | Article recommendation model training method, article recommendation method, device and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210906295.3A CN115222486B (en) | 2022-07-29 | 2022-07-29 | Article recommendation model training method, article recommendation method, device and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115222486A CN115222486A (en) | 2022-10-21 |
CN115222486B true CN115222486B (en) | 2024-02-02 |
Family
ID=83613249
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210906295.3A Active CN115222486B (en) | 2022-07-29 | 2022-07-29 | Article recommendation model training method, article recommendation method, device and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115222486B (en) |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2014074961A (en) * | 2012-10-02 | 2014-04-24 | Nippon Telegr & Teleph Corp <Ntt> | Commercial product recommendation device, method and program |
CN111275521A (en) * | 2020-01-16 | 2020-06-12 | 华南理工大学 | Commodity recommendation method based on user comment and satisfaction level embedding |
CN113254785A (en) * | 2021-06-21 | 2021-08-13 | 腾讯科技(深圳)有限公司 | Recommendation model training method, recommendation method and related equipment |
CN113256367A (en) * | 2021-04-25 | 2021-08-13 | 西安交通大学 | Commodity recommendation method, system, equipment and medium based on user behavior historical data |
CN113850654A (en) * | 2021-10-26 | 2021-12-28 | 北京沃东天骏信息技术有限公司 | Training method of item recommendation model, item screening method, device and equipment |
WO2022016522A1 (en) * | 2020-07-24 | 2022-01-27 | 华为技术有限公司 | Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium |
CN114764471A (en) * | 2021-01-12 | 2022-07-19 | 腾讯科技(深圳)有限公司 | Recommendation method, recommendation device and storage medium |
-
2022
- 2022-07-29 CN CN202210906295.3A patent/CN115222486B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2014074961A (en) * | 2012-10-02 | 2014-04-24 | Nippon Telegr & Teleph Corp <Ntt> | Commercial product recommendation device, method and program |
CN111275521A (en) * | 2020-01-16 | 2020-06-12 | 华南理工大学 | Commodity recommendation method based on user comment and satisfaction level embedding |
WO2022016522A1 (en) * | 2020-07-24 | 2022-01-27 | 华为技术有限公司 | Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium |
CN114764471A (en) * | 2021-01-12 | 2022-07-19 | 腾讯科技(深圳)有限公司 | Recommendation method, recommendation device and storage medium |
CN113256367A (en) * | 2021-04-25 | 2021-08-13 | 西安交通大学 | Commodity recommendation method, system, equipment and medium based on user behavior historical data |
CN113254785A (en) * | 2021-06-21 | 2021-08-13 | 腾讯科技(深圳)有限公司 | Recommendation model training method, recommendation method and related equipment |
CN113850654A (en) * | 2021-10-26 | 2021-12-28 | 北京沃东天骏信息技术有限公司 | Training method of item recommendation model, item screening method, device and equipment |
Non-Patent Citations (1)
Title |
---|
基于用户评论的深度情感分析和多视图协同融合的混合推荐方法;张宜浩;朱小飞;徐传运;董世都;;计算机学报(第06期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN115222486A (en) | 2022-10-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112164391A (en) | Statement processing method and device, electronic equipment and storage medium | |
CN115917535A (en) | Recommendation model training method, recommendation device and computer readable medium | |
CN112380449B (en) | Information recommendation method, model training method and related device | |
US11874798B2 (en) | Smart dataset collection system | |
CN112288554B (en) | Commodity recommendation method and device, storage medium and electronic device | |
CN115659008B (en) | Information pushing system, method, electronic equipment and medium for big data information feedback | |
CN114201516B (en) | User portrait construction method, information recommendation method and related devices | |
CN116739695A (en) | Electronic commerce management system and method based on big data | |
US11269896B2 (en) | System and method for automatic difficulty level estimation | |
Khan et al. | Comparative analysis on Facebook post interaction using DNN, ELM and LSTM | |
CN117217284A (en) | Data processing method and device | |
CN118043802A (en) | Recommendation model training method and device | |
CN116821516B (en) | Resource recommendation method, device, equipment and storage medium | |
CN113836390B (en) | Resource recommendation method, device, computer equipment and storage medium | |
CN113592593A (en) | Training and application method, device, equipment and storage medium of sequence recommendation model | |
CN115222433A (en) | Information recommendation method and device and storage medium | |
CN115222486B (en) | Article recommendation model training method, article recommendation method, device and storage medium | |
CN113159877B (en) | Data processing method, device, system and computer readable storage medium | |
CN111915339A (en) | Data processing method, device and equipment | |
KR102653187B1 (en) | web crawling-based learning data preprocessing electronic device and method thereof | |
CN113221534B (en) | Text emotion analysis method and device, electronic equipment and storage medium | |
CN117132810B (en) | Target detection method, model training method, device, equipment and storage medium | |
CN116188053A (en) | Data processing method, device, computer equipment and storage medium | |
CN117436980A (en) | Insurance product recommendation method and device, equipment and storage medium | |
CN116957641A (en) | Click behavior prediction method and device, electronic 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 |