CN113158045A - Interpretable recommendation method based on graph neural network reasoning - Google Patents

Abstract

The invention discloses an interpretable recommendation method based on graph neural network reasoning. The method includes: constructing a multi-relational user behavior graph according to the interaction matrix of user behavior; for the user behavior graph, using a user behavior preference understanding model to learn high-level association relationships and disseminate user behavior preferences, and obtain user, item, user-item relationship The associated vector representation; input the user state representation, item state representation, and user-item association representation output by the user behavior preference understanding model into the user behavior preference understanding model to obtain item recommendations for a given user; The fusion of user preference state representation, item state representation, and user-item association representation is used as the input to the explanation generation model, and combined with a set of text reviews, relevant explanations of the items recommended for a given user are obtained. The present invention can generate high-quality, easy-to-understand recommendation explanations while providing high-performance recommendation results.

Description

An Interpretable Recommendation Method Based on Graph Neural Network Reasoning

technical field

The present invention relates to the field of computer technology, and more particularly, to an interpretable recommendation method based on graph neural network reasoning.

Background technique

As an important tool to alleviate the problem of information overload, recommender systems have been widely used in e-commerce, news recommendation, video recommendation and other fields, gradually changing people's lifestyles. For example, an online bookstore recommends books that users like based on their historical behavior information. If users can understand why the book is recommended with an easy-to-understand explanation, it can greatly improve the effectiveness of the recommendation (helping users make decisions quickly) and the persuasiveness of the recommendation (improving the possibility of users buying the book) ), which is the interpretability of recommender systems. The explainable recommendation system presents the reasons and explanations of the recommendation while recommending items to users, which can not only improve the transparency of the recommendation system, improve the user's trust and acceptance of the recommendation system, but also improve the user's satisfaction with the recommended products and services. . Explainable recommender systems have important research significance and application value in fields such as e-commerce, finance, medical care, and law where interpretability and transparency are critical.

In the prior art, explainable recommendation methods mainly include two categories: explainable recommendation methods based on explanation generation and explainable recommendation methods based on knowledge graphs. In recent years, with the development of natural language processing technology, many interpretable recommendation methods based on explanation generation have been proposed, which can generate textual recommendation explanations in the form of real human comments. For example, a research result proposes a convolutional neural network model combined with a dual attention mechanism to learn user preference representation and item feature representation from user/item reviews, and use local and global attention mechanisms to select informative words to assist score prediction , to enhance the interpretability of the model. Another example, a research result proposes a neural scoring regression model to predict user-item ratings, and uses recurrent neural networks such as LongShort-Term Memory (LSTM), based on the encoder-decoder framework. The Gate Recurrent Unit (GRU) generates textual interpretations. Another research result uses matrix factorization for score prediction, and proposes a sequence-to-sequence explanation generative model based on generative adversarial networks.

For interpretable recommendation methods based on knowledge graphs, the introduction of knowledge graphs as auxiliary information into recommender systems has received more and more attention, and knowledge graphs can introduce various types of related information. Researchers use the reasoning path of knowledge graph to enhance the transparency and interpretability of the model, and propose some interpretable recommendation methods based on knowledge graph. In one study, an end-to-end knowledge graph-aware recommendation method, RippleNet, was proposed to automatically discover the propagation path from the associated paths of the knowledge graph, and use the information propagation mechanism to discover the hierarchical potential interests of users. Another research result proposes a recurrent neural network for knowledge-aware paths, which combines the representation of knowledge graph entities and relationships to generate path representations, uses the serialized dependencies of paths to infer user preferences, and applies reasoning paths to provide recommended explanations.

However, although existing interpretable recommendation schemes use text information extraction to generate recommended explanations, or use knowledge graph reasoning paths to enhance model interpretability, and achieve certain results, there are still problems in user behavior analysis modeling and interpretation quality. certain deficiencies. In addition, interpretable recommendation methods based on explanation generation directly use generative models to generate recommended explanations. Although they can generate text explanations that are easy for humans to understand in the form of explanations, due to the problem of data sparsity, the generated explanations are of low quality and poor readability.

In addition, the interpretable recommendation method based on knowledge graph uses the facts and associations introduced from common knowledge bases such as DBPedia and FreeBase to construct knowledge graphs. Although information dissemination and reasoning based on graph models can improve the recommendation performance, it introduces redundant noise. Data, it is difficult to deal with unrelated entities, which affects the recommendation efficiency and results. Moreover, methods based on knowledge graphs need to manually pre-set paths, rules, etc., and lack domain knowledge, resulting in homogenization of recommended interpretation results.

SUMMARY OF THE INVENTION

The purpose of the present invention is to overcome the above-mentioned defects of the prior art and provide an interpretable recommendation method based on graph neural network reasoning.

The technical solution of the present invention is to provide an interpretable recommendation method based on graph neural network reasoning. The method includes the following steps:

According to the interaction matrix of user behavior, construct a multi-relational user behavior graph, in which the node set of the graph is all users and all items, and the edges are user-item interaction relationships;

For the user behavior graph, use the user behavior preference understanding model to learn high-level associations and propagate user behavior preferences, and obtain a vector representation of user, item, and user-item associations;

Inputting the user state representation, the item state representation, and the user-item association representation output by the user behavior preference understanding model into the user behavior preference understanding model, to obtain an item recommendation for a given user;

The fusion of user preference state representation, item state representation, and user-item association representation output by the user behavior preference understanding model is used as the input of the explanation generation model, and combined with the text comment set, the relevant explanation of the item recommended for a given user is obtained. .

Compared with the prior art, the advantage of the present invention is that, for the problem of interpretable recommendation, an interpretable recommendation method based on graph neural network reasoning is proposed, and knowledge reasoning is used to deeply understand user behavior patterns and mine potential intentions of user behaviors, thereby realizing Comprehensive, fine-grained analysis and modeling of user behavior. In addition, the present invention takes into account the accuracy and interpretability of the recommendation, provides personalized recommendations to users while generating high-quality and easy-to-understand recommendation explanations, and generates recommendation explanations (recommendation reasons) while providing recommendation results, thereby improving the recommendation system. The transparency of the recommendation system and users' trust and acceptance of the recommendation system, and improve the user's satisfaction with the recommended products and services.

Other features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the accompanying drawings.

Description of drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention.

1 is a flowchart of an interpretable recommendation method based on graph neural network reasoning according to an embodiment of the present invention;

FIG. 2 is a frame diagram of an interpretable recommendation method based on graph neural network reasoning according to an embodiment of the present invention.

Detailed ways

Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, the numerical expressions and numerical values set forth in these embodiments do not limit the scope of the invention unless specifically stated otherwise.

The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, such techniques, methods, and apparatus should be considered part of the specification.

In all examples shown and discussed herein, any specific values should be construed as illustrative only and not limiting. Accordingly, other instances of the exemplary embodiment may have different values.

It should be noted that like numerals and letters refer to like items in the following figures, so once an item is defined in one figure, it does not require further discussion in subsequent figures.

In the embodiment described below, it is assumed that there are m users in the data set, denoted as u; n items, denoted as v. The interaction behavior between the user and the item is represented as < _ui , _vj , r _ij , d _ij >, where r _ij represents the user _ui 's rating on the item v _j , and d _ij represents the user _ui 's textual comment on the item v _j , The user-item interaction matrix is denoted R, the set of text reviews is denoted D, and the dictionary composed of all text reviews is denoted V _g .

The problem of the present invention can be defined as: given a set of users U, a set of items V, a user-item interaction matrix R and a set of text reviews D, for a given user u, recommend the top-N items (the ones that the user has not interacted with before). item), while generating an explanation Y for the recommended item.

In short, the interpretable recommendation method based on graph neural network reasoning provided by the present invention realizes comprehensive user behavior modeling by constructing a user behavior graph and using graph neural network reasoning to propagate multi-hop information. In general, the method includes four parts, which are user behavior graph construction, user behavior preference understanding based on graph neural network inference, user behavior interaction prediction, and explanation generation model.

Specifically, as shown in FIG. 1 and FIG. 2 , the provided interpretable recommendation method based on graph neural network reasoning includes the following steps.

Step S110, for the interaction matrix of user behavior, construct a user behavior graph.

For the user behavior interaction matrix R, a multi-relational user behavior graph G is constructed, where the node set (node) of the graph G is all users and all items U∪V. Edges are user-item interactions. For a graph G, the graph embedding representation of the initial nodes and edges using random variables. The graph embedding representation of user node is denoted as e _u , the graph embedding representation of item node is denoted as e _v , and the edge representation of user-item interaction relationship is denoted as r _uv . By constructing a user behavior map, it is beneficial to the subsequent understanding of user behavior.

Step S120, using a graph neural network to infer user behavior preference understanding.

In one embodiment, a user behavior preference understanding model based on graph attention network is proposed, which can not only learn high-level association information, but also use a multi-hop propagation mechanism to deeply and comprehensively understand user behavior preferences. , a complete vector representation of users, items, and user-item associations can be obtained. Specifically include the following steps:

Step S121, information dissemination of the user behavior graph attention network.

根据图结构的一阶关联性，从一阶邻居节点

传播到节点e_i的信息定义为：Define the information dissemination mechanism, that is, the information dissemination between a node and its neighbor nodes (directly connected nodes). For node e _i , its first-order neighbor node set is

According to the first-order associativity of the graph structure, from the first-order neighbor nodes

The information propagated to node e _i is defined as:

测量了一对关系节点间的影响程度，δ为非线性的激活函数，可以选用ReLU或者tanh函数等，W_t，W_r，W_e，W_n为学习参数，b是偏置参数。

同时考虑了关联的节点e_i、e_j和它们间关联关系的影响r_ij，γ为正则化后的影响度量值。因此邻居节点向当前节点e_i传播的信息是权重化地邻居节点表示组合

in,

The influence degree between a pair of relational nodes is measured, δ is a nonlinear activation function, ReLU or tanh function can be selected, W _t , W _r , We , W _n are learning parameters, and b _is a bias parameter.

At the same time, the influence r _ij of the associated nodes e _i , e _j and the relationship between them is considered, and γ is the regularized influence metric value. Therefore, the information propagated by the neighbor nodes to the current node e _i is the weighted neighbor node representation combination

Step S122, information fusion of the user behavior graph attention network.

为

其中

是所使用的融合函数，可采用图卷积网络的非线性加权求和等融合方式。After obtaining the information propagation of the (first-order) adjacent nodes, fuse the embedded representation e _i of the node itself and the propagation information from the adjacent nodes

for

in

is the fusion function used, and fusion methods such as nonlinear weighted summation of graph convolutional networks can be used.

Step S123, the user behavior graph attention network is updated.

的表示为：

In one embodiment, the graph neural network employs multi-hop reasoning in a recursive manner to model the higher-order connectivity information of the graph structure. Therefore, the fused representation is subsequently updated to the new representation of the node. For example, at the kth update, the node

is expressed as:

Through the above process, the representations of nodes and edges are finally output, that is, user state representation e _u , item state representation _ev , and user-item association representation r _uv .

In this step S120, aiming at the deficiencies of existing methods in user behavior modeling, it is proposed to construct a user behavior preference understanding model based on graph neural network reasoning, so as to obtain a high-quality user behavior preference representation.

Step S130, using a neural network collaborative filtering framework to predict user behavior interaction.

In one embodiment, based on a neural network collaborative filtering framework, user representation, item representation, and user-item association representation are integrated, and a neural collaboration layer is used to implement recommendation prediction.

For example, a three-layer neural collaborative filtering model (Neural CF layers) is applied to achieve user-item interaction prediction. The network structure of the model follows a tower-like structure, where the lowest layer is the widest, and each subsequent layer has fewer neurons. The output of each layer in the model is used as the input to the next layer. Specifically, the input of the neural network collaboration layer of the first layer is the splicing of user state representation e _u , item state representation _ev , and user-item association representation r _uv , namely [e _u , e _v , r _uv ]. At each neural network synergy layer, a hidden vector is generated, which is expressed as:

为使用的激活函数，可使用修正线性单元ReLU。在最后一层，隐藏向量h_l被映射为预测评分

表示为：where W _l and b _l are the weight matrix and bias matrix of the lth layer,

For the activation function used, a rectified linear unit ReLU can be used. In the last layer, the hidden vector h _l is mapped to the predicted score

Expressed as:

推荐top-N个物品，即前N个推荐物品，N是预先设定的整数。where W is the weight matrix and σ is the sigmoid function. For user u, sort by predicted score

Recommend top-N items, that is, the top N recommended items, where N is a preset integer.

Step S140, using the explanation generation model to obtain an intuitive text explanation for the recommended item.

In one embodiment, an explanation generator based on an encoder-decoder framework is designed, a copy mechanism is introduced to extract relevant information from user/item source text reviews, and a high-quality, easy-to-understand recommendation explanation is generated by combining the generation mode and the copy mode.

Specifically, the recurrent neural network GRU is used as the explanation generator, and the copy mechanism is introduced to extract information from the original comments of users, and the two modes (generating mode and copy mode) are combined to generate an intuitive text explanation Y=[y ₁ , y ₂ , ..y _T ] (word sequence), easy for users to read and understand.

为用户偏好状态表示e_u、物品状态表示e_v、用户-物品关联表示r_uv的融合，用户u_i推荐物品v_j时生成的解释词y_ijt为w_t的概率为P_gen(y_ijt＝w_t)，表示为：For generative mode, GRU is utilized as the decoder, where the initial state of the decoder

is the fusion of user preference state representation e _u , item state representation _ev , and user-item association representation r _uv , the probability that the interpretation word y _ijt generated when user _ui recommends item v _j is w _t is P _gen (y _ijt = w _t ), expressed as:

为在t时刻GRU的隐藏状态，wt是词的向量表示，wt来自整个词典V_g。in,

is the hidden state of the GRU at time t, wt is the vector representation of the word, and wt comes from the entire dictionary V _g .

在拷贝模式下，解释词y_ijt为w_t的概率表示为P_copy(y_ijt＝w_t)：For the copy mode, it is to calculate the probability of the word appearing in the original comment, and when recommending the item v _j to the user _ui , the copy source of the word is the set of words of the user comment document of _ui and the item comment document of v _j .

In copy mode, the probability that the interpreter y _ijt is _wt is expressed as P _copy (y _ijt = _wt ):

Preferably, an explanation generation model is formed by combining the generation mode and the copy mode, and the probability of finally generating the target word is P(y _ijt =wt). Referring to formula (6), a recommendation explanation _{Yi ij} for recommending the item v _j to the user u _i is generated. =[y _{ij1 , y ij2 ,} .. y _ijT ], T is the interpretation length.

P(y _ijt =w _t )=P _gen (y _ijt =w _t )+P _copy (y _ijt =w _t ) (6)

By designing an explanation generation model that combines the generation mode and the copy mechanism, compared with the existing explainable recommendation methods based on explanation generation, high-quality and easy-to-understand recommendation explanations can be obtained, which solves the problems of low explanation quality and poor readability. question.

It should be noted that the method proposed by the present invention is an end-to-end method, so the feature representation learning, recommendation prediction and explanation generation model are trained and optimized by means of joint learning. This end-to-end interpretable recommendation framework reduces the difficulty of training. , speeding up the training process while reducing memory requirements,

In order to verify the effectiveness and advancement of the present invention, a large number of experiments are carried out on two datasets, Amazon-Electronics and Amazon-Music Instrument, using the proposed interpretable method based on graph neural network inference. The experimental results show that the present invention is always better than the current best method for interpretable recommendation, shows great advantages, and has a very broad application prospect.

Compared with the prior art, the present invention has at least the following advantages:

(1) For the interpretable recommendation task, the proposed interpretable recommendation method based on graph neural network reasoning provides high-performance recommendation results while generating high-quality, easy-to-understand recommendation explanations to achieve both recommendation effectiveness and A recommended framework for interpretability.

(2) Based on graph neural network reasoning, a deep-level user behavior understanding model is proposed, which uses the high-order correlation information propagation in user behavior to perform deep-level user behavior reasoning, and realizes comprehensive and deep-level user behavior modeling and understanding. Obtaining high-quality representations in three aspects: user, item, and user-item can effectively improve recommendation performance.

(3) In the explanation generation model, a copy mechanism is introduced, and an explanation generation model combining the generation mode and the copy mode is proposed, which is conducive to the generation of high-quality and easy-to-understand recommended explanations.

To sum up, the present invention innovatively proposes a user behavior understanding model based on graph neural network reasoning. , User-item association representation, realizes comprehensive and fine-grained user behavior modeling, and solves the problem of insufficiency of current user behavior analysis and modeling.

It should be understood that, without departing from the spirit and scope of the present invention, those skilled in the art can make appropriate changes or modifications to the above embodiments, for example, using LSTM to replace the GRU network, or using other activation functions for non- Linear processing, etc.

The present invention may be a system, method and/or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the present invention.

A computer-readable storage medium may be a tangible device that can hold and store instructions for use by the instruction execution device. The computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples (non-exhaustive list) of computer readable storage media include: portable computer disks, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM) or flash memory), static random access memory (SRAM), portable compact disk read only memory (CD-ROM), digital versatile disk (DVD), memory sticks, floppy disks, mechanically coded devices, such as printers with instructions stored thereon Hole cards or raised structures in grooves, and any suitable combination of the above. Computer-readable storage media, as used herein, are not to be construed as transient signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (eg, light pulses through fiber optic cables), or through electrical wires transmitted electrical signals.

The computer readable program instructions described herein may be downloaded to various computing/processing devices from a computer readable storage medium, or to an external computer or external storage device over a network such as the Internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and/or edge servers. A network adapter card or network interface in each computing/processing device receives computer-readable program instructions from a network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing/processing device .

The computer program instructions for carrying out the operations of the present invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state setting data, or instructions in one or more programming languages. Source or object code written in any combination, including object-oriented programming languages, such as Smalltalk, C++, Python, etc., and conventional procedural programming languages, such as the "C" language or similar programming languages. The computer readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server implement. In the case of a remote computer, the remote computer may be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (eg, using an Internet service provider through the Internet connect). In some embodiments, custom electronic circuits, such as programmable logic circuits, field programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), can be personalized by utilizing state information of computer readable program instructions. Computer readable program instructions are executed to implement various aspects of the present invention.

Aspects of the present invention are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer readable program instructions.

These computer readable program instructions may be provided to a processor of a general purpose computer, special purpose computer or other programmable data processing apparatus to produce a machine that causes the instructions when executed by the processor of the computer or other programmable data processing apparatus , resulting in means for implementing the functions/acts specified in one or more blocks of the flowchart and/or block diagrams. These computer readable program instructions can also be stored in a computer readable storage medium, these instructions cause a computer, programmable data processing apparatus and/or other equipment to operate in a specific manner, so that the computer readable medium on which the instructions are stored includes An article of manufacture comprising instructions for implementing various aspects of the functions/acts specified in one or more blocks of the flowchart and/or block diagrams.

Computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other equipment to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other equipment to produce a computer-implemented process , thereby causing instructions executing on a computer, other programmable data processing apparatus, or other device to implement the functions/acts specified in one or more blocks of the flowcharts and/or block diagrams.

The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more functions for implementing the specified logical function(s) executable instructions. In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It is also noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented in dedicated hardware-based systems that perform the specified functions or actions , or can be implemented in a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that implementation in hardware, implementation in software, and implementation in a combination of software and hardware are all equivalent.

Various embodiments of the present invention have been described above, and the foregoing descriptions are exemplary, not exhaustive, and not limiting of the disclosed embodiments. Numerous modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein was chosen to best explain the principles of the embodiments, the practical application or technical improvement in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein. The scope of the invention is defined by the appended claims.

Claims (10)

1. An interpretable recommendation method based on graph neural network reasoning comprises the following steps:

step S1: aiming at an interaction matrix of user behaviors, constructing a multi-relationship user behavior graph, wherein a node set of the graph comprises all users and all articles, and the edges are user-article interaction relationships;

step S2: for the user behavior diagram, learning a high-order incidence relation by using a user behavior preference understanding model and transmitting user behavior preference to obtain vector representation of user, article and user-article correlation;

step S3: inputting the user state representation, the article state representation and the user-article association representation output by the user behavior preference understanding model into the user behavior preference understanding model to obtain article recommendation of a given user;

step S4: and taking the fusion of the user preference state representation, the item state representation and the user-item association representation output by the user behavior preference understanding model as the input of an interpretation generation model, and combining a text comment set to obtain the relevant interpretation of the item recommended for the given user.

2. The method according to claim 1, wherein step S2 includes:

for the user behavior graph G, according to the first-order relevance of the graph structure, the first-order neighbor nodes

Propagate to node e_iThe information of (2) is defined as:

wherein,

is a pair of relational nodes e_jAnd e_iDegree of influence between, delta being a non-linear activation function, W_t，W_r，W_e，W_nIs the correlation term weight, b is the offset, r_ijExpressing the influence of the incidence relation between the nodes, wherein gamma is the influence metric value after regularization, and the neighbor node points to the current node e_iThe propagated information is a weighted combination of the representations of the neighboring nodes

Node e_iIs a first-order neighbor node set

After information propagation of the first-order adjacent nodes is obtained, the embedded expression e of the fusion node is self_iAnd propagation information from neighboring nodes

Is shown as

Wherein

As a fusion function used;

modeling the high-order connection information of the graph structure by using multi-hop inference in a recursive mode, and finally outputting the representation of nodes and edges, wherein the representation is marked as a user state e_uArticle state e_vUser-item association representation r_uv。

3. The method according to claim 1, wherein step S3 includes:

the user behavior preference understanding model adopts a neural collaborative filtering model with a tower-shaped structure, the input of a first layer of neural network collaborative layer is the splicing of user state representation, article state representation and user-article association representation, and for each neural network collaborative layer, the generation of a hidden vector is shown as follows:

wherein, W_lAnd b_lThe weight matrix and the bias matrix for the l-th layer,

is an activation function, h_l-1Is the vector of the l-1 layer output;

at the last layer of the neural collaborative filtering model, the vector h is hidden_lPredicted scores mapped to recommended items

Scoring based on prediction for a given user

The items are sorted and recommended.

4. The method of claim 1, wherein in step S4, the given user u is generated by extracting information from the user' S original comments in conjunction with the copy mode_iRecommending items v_jRecommended explanation of (Y)_ij＝[y_ij1，y_ij2，..y_ijT]Expressed as:

P(y_ijt＝w_t)＝P_gen(y_ijt＝w_t)+P_copy(y_ijt＝w_t)

wherein T is the interpretation length, P_gen(y_ijt＝w_t) Is a user u obtained by using a generation pattern_iRecommending items v_jInterpreter y generated at the time_ijtIs w_tProbability of (P)_copy(y_ijt＝w_t) Is a user u obtained by a copy mechanism_iRecommendation materialArticle v_jInterpreter y generated at the time_ijtIs w_tThe probability of (c).

5. Method according to claim 4, characterized in that for the generation mode a recurrent neural network GRU is used as decoder, the initial state of which is the initial state

Representing e for user preference status_uArticle status representation e_vUser-item association representation r_uvFusion of (1), user u_iRecommending items v_jInterpreter y generated at the time_ijtIs w_tThe probability of (d) is expressed as:

wherein,

is a hidden state of GRU at time t, w_tIs a vector representation of a word, w_tFrom a dictionary V_g。

6. The method of claim 5, wherein in copy mode, user u_iRecommending items v_jInterpreter y generated at the time_ijtIs w_tThe probability of (d) is expressed as:

wherein, V_cIs u_iV and_jof item review documents。

7. The method of claim 3, wherein the hidden vector h is hidden at the last layer of the neural collaborative filtering model_lPredicted scores mapped to recommended items

Expressed as:

wherein W is a weight matrix, sigma is a sigmoid function, and h_lIs the concealment vector and l is the layer index.

8. The method of claim 2, wherein the fusion function

A non-linear weighted sum.

9. A computer-readable storage medium, on which a computer program is stored, which program, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 8.

10. A computer device comprising a memory and a processor, on which memory a computer program is stored which is executable on the processor, characterized in that the steps of the method of any of claims 1 to 8 are implemented when the processor executes the program.

Images