An information recommendation method and device, electronic equipment, chip and storage medium

Abstract

Embodiments of the present application disclose an information recommendation method, an information recommendation device, an electronic device, a chip and a computer readable storage medium. The method comprises: extracting a professional field term in each piece of information in a plurality of pieces of information; obtaining a recommendation list based on the professional field term, a professional field ontology, user historical browsing information and candidate recommendation information; and the present application recommends the candidate information based on the professional field ontology and the user historical browsing information, and the recommendation result is more accurate.

Description

Technical Field

[0001] This application relates to the field of recommendation algorithm technology, and in particular to an information recommendation method, apparatus, electronic device, chip, and storage medium. Background Technology

[0002] Currently, recommendations for information in a specific professional field rely on the ontology of that field. By acquiring keywords input by the user, processing the keywords, and using them to query related objects to be recommended in the ontology knowledge base, the similarity between the objects to be recommended is calculated based on the ontology knowledge base. This process then corrects the similarity results calculated based on traditional association recommendation algorithms before generating a recommendation list. This recommendation method relies entirely on the ontology and does not take into account other influencing factors, resulting in low accuracy of the recommendation results. In addition, the interpretability of the calculation model is weak, and the recommendation results are difficult to explain convincingly. Summary of the Invention

[0003] This application provides an information recommendation method, apparatus, electronic device, chip, and computer-readable storage medium.

[0004] The technical solution of this application embodiment is implemented as follows:

[0005] In a first aspect, embodiments of this application provide an information recommendation method, including:

[0006] Extract specialized terminology from each of multiple pieces of information;

[0007] Based on the aforementioned professional domain terminology, professional domain ontology, user browsing history, and candidate recommendation information, a recommendation list is obtained.

[0008] Secondly, embodiments of this application provide an information recommendation device, including:

[0009] Extraction Unit: Used to extract specialized terminology from each of multiple pieces of information;

[0010] Generation unit: used to generate a recommendation list based on the professional domain terminology, professional domain ontology, user browsing history information, and candidate recommendation information.

[0011] Thirdly, this application provides an electronic device, including: a processor and a memory, the memory being used to store a computer program, and the processor being used to call and run the computer program stored in the memory to execute any of the information recommendation methods provided in the embodiments of this application.

[0012] Fourthly, this application provides a chip, including: a processor, for calling and running a computer program from a memory, causing a device equipped with the chip to execute any of the information recommendation methods provided in the embodiments of this application.

[0013] Fifthly, this application provides a computer-readable storage medium for storing a computer program that causes a computer to execute any of the information recommendation methods provided in the embodiments of this application.

[0014] The information recommendation method provided in this application extracts professional domain terms from each of multiple pieces of information; based on the professional domain terms, professional domain ontology, user browsing history information, and candidate recommendation information, a recommendation list is obtained; and the candidate information is recommended based on the professional domain ontology and user browsing history information, resulting in more accurate recommendation results. Attached Figure Description

[0015] Figure 1 A schematic diagram illustrating the implementation process of an information recommendation method provided in this application embodiment;

[0016] Figure 2 This is a schematic diagram of a directed acyclic graph provided in an embodiment of this application;

[0017] Figure 3 A schematic diagram illustrating the process of obtaining the seventh representation vector matrix provided in an embodiment of this application;

[0018] Figure 4 This is a schematic diagram of the structure of an information recommendation device provided in an embodiment of this application;

[0019] Figure 5 A schematic structural diagram of an electronic device provided in the embodiments of this application;

[0020] Figure 6 This is a schematic structural diagram of the chip provided in an embodiment of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0022] It should be noted that, in the embodiments of this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, in the embodiments of this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0023] In the description of the embodiments of this application, the term "correspondence" may indicate that there is a direct or indirect correspondence between two things, or that there is an association between two things, or that there is a relationship of instruction and being instructed, configuration and being configured, etc.

[0024] To facilitate understanding of the technical solutions of the embodiments of this application, the relevant technologies of the embodiments of this application are described below. The following relevant technologies are optional solutions and can be combined with the technical solutions of the embodiments of this application in any way, and they all fall within the protection scope of the embodiments of this application.

[0025] Doc2Vec is an open-source tool that can transform high-dimensional sparse vectors into low-dimensional dense vectors, allowing you to flexibly obtain representation vectors for sentences, paragraphs, or documents as needed.

[0026] There are many types of attention mechanisms. If classified according to their scope of attention, they can be divided into global attention, local attention, soft attention, and hard attention.

[0027] GloVe is used to learn the vector space representation of words. This model combines global matrix factorization and a local context window approach. It can effectively utilize statistical information by training only on the non-zero elements of the word-word co-occurrence matrix instead of the entire sparse matrix.

[0028] To gain a more detailed understanding of the features and technical content of this application, the implementation of this application will be described in detail below with reference to the accompanying drawings. The accompanying drawings are for reference and illustration only and are not intended to limit this application.

[0029] Figure 1 A schematic diagram of the implementation process of an information recommendation method provided in this application embodiment. Figure 1 ,like Figure 1 As shown in the figure, this application provides an information recommendation method applied to an information recommendation device. The method includes the following steps:

[0030] Step 101: Extract the professional terminology from each of the multiple pieces of information.

[0031] Here, the professional field can be water conservancy or other professional fields; this application does not limit this. Each piece of information includes a title and a body. In this step, it is necessary to extract the professional field terms from the title and body of each piece of information from multiple pieces of information.

[0032] You can use the jieba word segmentation dictionary to extract specialized terminology. Specifically, you can customize a jieba word segmentation dictionary, add specialized terminology to it, and use the jieba word segmentation function to segment the text, extracting specialized terminology from the title and body of each message, forming a set w1, w2, w3, ..., w... containing all specialized terminology from multiple messages. |w| ∈W, where |w| represents the total number of domain terms in set W.

[0033] Step 102: Based on the professional domain terminology, professional domain ontology, user browsing history information, and candidate recommendation information, a recommendation list is obtained.

[0034] Specifically, the candidate recommendation information is a portion of the multiple pieces of information. In this embodiment, the candidate information is recommended based on the professional domain ontology and the user's browsing history, resulting in more accurate recommendation results.

[0035] Another embodiment of this application provides an information recommendation method, which is similar to... Figure 1 The illustrated embodiment differs in that step 102: based on the professional domain terminology, professional domain ontology, user browsing history information, and candidate recommendation information, a recommendation list is obtained, including the following steps:

[0036] Step 1021: Based on the professional domain terminology, the professional domain ontology, and the user's browsing history information, obtain a first representation vector matrix; wherein, the first representation vector matrix is ​​a matrix formed by the representation vectors of each browsing information in the user's browsing history information.

[0037] Here, the initialization representation vector of each browsing information in the user's browsing history can be obtained by combining professional domain terms and user browsing history information. Schematic, it can be represented by a multi-hot vector. The initialization representation vector contains only 0 or 1. Specifically, for a certain browsing information in the user's browsing history, the value of the corresponding position of its initialization representation vector is 1 if and only if it contains professional domain terms from set W. Schematic, if set W includes three professional domain terms w1, w2, and w3, and a certain browsing history record includes terms w1 and w2 but not w3, then the initialization representation vector of that browsing history record is (0, 0, 1). The fifth representation vector matrix is ​​formed by the initialization representation vectors of each browsing information in the user's browsing history.

[0038] Here, the domain ontology describes domain concepts and the relationships between them, representing the hierarchical structure of domain concepts. Based on relevant knowledge, domain terms can be added to the corresponding concepts in the domain ontology, forming a directed acyclic graph (DAG). In this DAG, leaf nodes represent domain terms, and non-leaf nodes represent the domain ontology. In this embodiment, all nodes in the DAG can be represented as a set Z = W + W', where W' = {w |w|+1 ,w |w|+2 ,...,w |w|+|w'|} is a collection of all professional domain ontology, and all of them are non-leaf nodes in a directed acyclic graph. The parent node in the directed acyclic graph represents a more generalized water conservancy concept compared to its child nodes.

[0039] Furthermore, based on the user's browsing history information and the directed acyclic graph, a sixth representation vector matrix is ​​obtained; wherein, the sixth representation vector matrix is ​​a matrix formed by the representation vectors of each professional domain term in the professional domain terminology.

[0040] Another embodiment of this application provides an information recommendation method, wherein obtaining a sixth representation vector matrix based on the user's historical browsing information and the directed acyclic graph includes:

[0041] Based on the user's historical browsing information and the directed acyclic graph, each browsing information in the user's historical browsing information is enhanced to obtain the enhanced browsing information of each user.

[0042] Based on the professional domain terminology and the enhanced user's browsing history for each time, a co-occurrence matrix composed of all professional domain terminology and professional domain ontology is obtained;

[0043] Based on the co-occurrence matrix, the initialization representation vector of each node in the directed acyclic graph is obtained through the second model;

[0044] Based on the initial representation vector of each node in the directed acyclic graph and the directed acyclic graph, the representation vector of each professional domain term in the professional domain terminology is obtained through the second convolutional neural network, forming the sixth representation vector matrix.

[0045] Specifically, as an example, suppose the list [[a,b,c],[d,e,f],[g,h]] represents a user's complete access record sequence, representing the user's historical browsing information. Here, [a,b,c], [d,e,f], and [g,h] represent the user's single browsing information, and a, b, c, d, e, f, g, and h represent professional domain terms.

[0046] As an example, for a single browsing record of Vt The browsing information for [a,i,j] is enhanced as follows (see reference). Figure 2 , Figure 2 This is a schematic diagram of a directed acyclic graph provided in an embodiment of this application, using leaf nodes c. a ,c i ,c j To expand V using ancestor nodes t ,like Figure 2 As shown, c a The ancestor node is c d ,c e ,c h ,c k c i The ancestor node is c f ,c b ,c k c j The ancestor node is c g ,c b ,c c ,c k Then you can get the enhanced access record V. t '=[c a ,c d ,c e ,c h ,c k ,c i ,c f ,c b ,c k ,c j ,c g ,c b ,c c ,c k The ] represents the enhanced browsing information. It is understandable that ancestor nodes with broader concepts will appear more frequently in the enhanced browsing information.

[0047] After enhancing all historical browsing information, the total frequency of occurrence of all nodes in the directed acyclic graph across the enhanced historical browsing information is calculated, forming a co-occurrence matrix composed of all domain terms and domain ontology. As a preferred embodiment of this application, further reference is made... Figure 2 In a directed acyclic graph, for the access record V t 'node c i ,c j The co-occurrence frequency is calculated as cooc(c i ,c j V t ')=co(c i V t ')×co(c j Vt '), where co(c i V t ') is c i In V t The number of occurrences of ' in ' can be understood as calculating the number of occurrences of node c for each of the multiple browsing entries. i ,c j The co-occurrence counts of each of the multiple enhanced browsing information are summed to obtain the total co-occurrence count of the two nodes.

[0048] Furthermore, the co-occurrence matrix is ​​input into the second model to obtain the initial representation vector of each node in the directed acyclic graph. In practical applications, the second model can be a GloVe model.

[0049] The leaf nodes in the directed acyclic graph, i.e., the representation vectors of the domain terms, can be calculated using the following formula: Among them, e n This represents the initialization vector of a node, and index(n) represents node c in the directed acyclic graph. n The set of ancestor nodes, α mn Initialize the representation vector e of the nodes trained by the attention mechanism. m With e n The weighted score between them is calculated using the following formula: Where w and b represent the weight matrix and bias vector, respectively, and U is the weight vector. Thus, the representation vector of each leaf node in the directed acyclic graph, i.e., each professional domain term in the aforementioned professional domain terminology, is obtained through the second neural network model, forming the sixth representation vector matrix.

[0050] Furthermore, the first representation vector matrix is ​​obtained based on the fifth representation vector matrix and the sixth representation vector matrix.

[0051] Specifically, the sixth representation vector matrix is ​​multiplied by each vector in the fifth representation vector matrix to obtain the representation vector of each browsing information in the user's browsing history, forming the first representation vector matrix.

[0052] In this embodiment, ontology is used to enhance the recommendation effect. When there are relatively few specialized domain terms in the user's browsing history information, the ontology to which the terms belong can be used to obtain representation vectors of the user's browsing records that incorporate domain knowledge at different levels of importance. This enriches the representation vectors and strengthens the recommendation effect.

[0053] Step 1022: Obtain the seventh representation vector matrix, and based on the seventh representation vector matrix and the user's historical browsing information, obtain the second representation vector matrix; wherein, the seventh representation vector matrix is a matrix formed by the representation vectors of each title text in the multiple pieces of information, and the second representation vector matrix is a matrix formed by the representation vectors of each title text in the user's historical browsing information.

[0054] Here, each title text in the multiple pieces of information is preprocessed, and each preprocessed title text in the multiple pieces of information is used as the input of the first model to train the first model. After the first model converges, the seventh representation vector matrix G = [g1, g2,..., g |G| can be obtained.

[0055] Here, stop words such as "ah" and "de" that often appear in the title text will have a negative impact on the training process. The title text can be preprocessed for word segmentation and stop word removal. Based on this, the preprocessing includes word segmentation and stop word removal. In practical applications, the first model can be Doc2Vec. Since PV-DM can train unlabeled data more efficiently than PV-DBOW, in a preferred embodiment of the present application, the PV-DM model is selected for training.

[0056] Reference Figure 3 , Figure 3 is a schematic flow chart for obtaining the seventh representation vector matrix provided by the embodiment of the present application. In this embodiment, in order to expand the prediction and enhance the reliability of the training result of the first model, the corpus dataset is added to the training set and the first model is trained together with all the title texts in the multiple pieces of information. It can be understood that the corpus dataset also needs to be preprocessed.

[0057] Based on this, in the information recommendation method provided by an embodiment of the present application, the training set includes: the corpus dataset.

[0058] Further, extract all the titles of the user's historical browsing information, and extract all the representation vectors corresponding to all the titles of the user's historical browsing information in the seventh representation vector matrix to obtain the second representation vector matrix [g 1' , g 2' ,..., g T , where T is the total number of the user's historical browsing information.

[0059] Step 1023: Based on the seventh representation vector matrix and the candidate recommendation information, obtain the representation vector of each title text in the candidate recommendation information.

[0060] Specifically, all titles in the candidate recommendation information are extracted, and all representation vectors corresponding to all titles in the user's historical browsing information are extracted in the seventh representation vector matrix to obtain the representation vector of each title text in the candidate recommendation information.

[0061] Based on this, in one embodiment of this application, obtaining the seventh representation vector matrix and obtaining the second representation vector matrix based on the seventh representation vector matrix and the user's historical browsing information includes:

[0062] Preprocess all title texts of the multiple messages;

[0063] The first model is trained using a training set, and after the first model converges, the seventh representation vector matrix is ​​obtained; wherein, the training set includes: all the title texts of the multiple pieces of information after preprocessing;

[0064] Extract all titles from the user's historical browsing information, and extract all representation vectors corresponding to all titles in the seventh representation vector matrix to obtain the second representation vector matrix;

[0065] The step of obtaining the representation vector for each title text in the candidate recommendation information based on the seventh representation vector matrix and the candidate recommendation information includes:

[0066] Extract all titles from the candidate recommendation information, and extract all representation vectors corresponding to all titles in the seventh representation vector matrix to obtain the representation vector of the title text of each piece of information in the candidate recommendation information.

[0067] Step 1024: Based on the candidate recommendation information and the professional domain terminology, obtain the representation vector of each piece of information in the candidate recommendation information.

[0068] Specifically, the initialization representation vector of each piece of information in the candidate recommendation information is obtained. The method for obtaining the vector is the same as that for obtaining each piece of information in the user's browsing history, and will not be described in detail here.

[0069] The sixth representation vector matrix is ​​multiplied by the initial representation vector of each piece of information in the candidate information to obtain the representation vector of each piece of information in the candidate recommendation information.

[0070] Based on this, in one embodiment of this application, obtaining the first representation vector matrix based on the professional domain terminology, the professional domain ontology, and the user's browsing history information includes:

[0071] Based on the aforementioned professional terminology and the user's historical browsing information, a fifth representation vector matrix is ​​obtained, wherein the fifth representation vector matrix is ​​a matrix formed by the initial representation vector of each browsing information in the user's historical browsing information;

[0072] Based on the terminology and ontology of the aforementioned professional fields, a directed acyclic graph is generated;

[0073] Based on the user's browsing history information and the directed acyclic graph, a sixth representation vector matrix is ​​obtained; wherein, the sixth representation vector matrix is ​​a matrix formed by the representation vectors of each professional domain term in the professional domain terminology;

[0074] Based on the fifth representation vector matrix and the sixth representation vector matrix, the first representation vector matrix is ​​obtained;

[0075] The step of obtaining a representation vector for each piece of information in the candidate recommendation information based on the candidate recommendation information and the professional domain terminology includes:

[0076] Based on the candidate recommendation information and the sixth representation vector matrix, the representation vector of each piece of information in the candidate recommendation information is obtained.

[0077] It should be noted that the dimensions of the representation vectors of each browsing information in the user's browsing history, the representation vectors of each title text in the user's browsing history, the representation vectors of each title text in the candidate recommendation information, and the representation vectors of each piece of information in the candidate recommendation information must be consistent to ensure subsequent processing.

[0078] Step 1025: Based on the first representation vector matrix, the second representation vector matrix, the representation vector of each title text in the candidate recommendation information, and the representation vector of each piece of information in the candidate recommendation information, obtain the click probability of the user facing each piece of information in the candidate recommendation information; based on the click probability of the user facing each piece of information in the candidate recommendation information, obtain the recommendation list.

[0079] Here, the representation vector of the title text and the representation vector of the user's browsing history information are learned by two sub-models respectively. In this embodiment, the representation vector of each browsing information in the user's browsing history information, the representation vector of each title text in the user's browsing history information, and the representation vector of each title text in the candidate recommendation information are fused by a convolutional neural network model to ensure better interconnection between the user's browsing history information and the candidate recommendation information while minimizing information loss.

[0080] Specifically, for a recommendation among the candidate recommendations, its corresponding representation vector g′ and its title text representation vector v′ are copied T times, where T represents the total number of the user's historical browsing information. The activation function is used to transform the representation vectors of the user's historical browsing information, the title text representation vectors of the user's historical browsing information, the T copies of the recommendation information's representation vector, and the T copies of the recommendation information's title text representation vector, ensuring that the four types of representation vectors after transformation maintain the original spatial relationship. The four types of representation vectors after transformation are used as the input of the convolutional neural network model.

[0081] Schematic illustration: The activation function can be the tanh activation function h, and the four representation vectors after activation are as follows:

[0082] h(g 1:T )=[h(g1),h(g2),...,h(g T )]

[0083] h(v 1:T = [h(v1),h(v2),...,h(v)] T )]

[0084] h(g' 1:T = [h(g'),h(g'),...,h(g')]

[0085] (v' 1:T = [h(v'),h(v'),...,h(v')]

[0086] The four transformed vectors are combined into a multi-channel input matrix U∈R. d×T×4 , U=[[h(g1)h(v1)h(g′)h(v′)][h(g2)h(v2)h(g′)h(v′)][h(g T )h(v T )h(g′)h(v′)]], where d represents the dimension of each vector. The dimensions of these four vectors are the same, and 4 represents 4 channels.

[0087] Furthermore, matrix U is processed using the following function: Where r∈R d×l×4 , represents the convolution kernel. Applying multiple convolution kernels with different window sizes l can yield multiple feature sets. Max pooling can be used on multiple feature sets to obtain the final feature set q. r' The pooling layer has a stride of l and a window size of 1×(T-l+1). The pooled features are concatenated to obtain the user's representation vector k(t)∈R when faced with the candidate recommendation information. 1×卷积核总个数The user's representation vector when faced with the candidate recommendation is then used as input to a multilayer perceptron. An activation function is applied to activate the perceptron, yielding the user's click probability p∈[0,1]. Following this method, the click probability of each candidate recommendation is obtained. These probabilities are then sorted from highest to lowest to obtain the final recommendation list.

[0088] It is understandable that T can be the total number of historical browsing information of the user, or it can be taken sequentially from [1, max] to train the convolutional neural network model, where max is the total number of historical browsing information of the user. When the training result is optimal, the T value under the optimal condition can be used to calculate the user's click probability.

[0089] Based on this, another embodiment of the present application provides an information recommendation method, wherein the step of obtaining the click probability of a user facing each piece of information in the candidate recommendation information based on the first representation vector matrix, the second representation vector matrix, the representation vector of each title text in the candidate recommendation information, and the representation vector of each piece of information in the candidate recommendation information includes:

[0090] Based on the first representation vector matrix, the second representation vector matrix, the representation vector of each title text in the candidate recommendation information, and the representation vector of each piece of information in the candidate recommendation information, the representation vector when the user faces each piece of information in the candidate recommendation information is obtained through the first convolutional neural network.

[0091] Based on the user's representation vector when facing each piece of information in the candidate recommendation information, the click probability of the user when facing each piece of information in the candidate recommendation information is obtained through a multilayer perceptron.

[0092] Another embodiment of this application provides an information recommendation method in the field of water conservancy, which includes the following steps:

[0093] Step S1: Extract all water-related terms from the titles and texts of multiple water conservancy information entries.

[0094] Step S2: Associate all water conservancy terms with the water conservancy entity to form a directed acyclic graph.

[0095] Step S3: Process the user's historical click sequence and use the nodes of the directed acyclic graph to form a two-level list representing the user's historical click sequence.

[0096] Here, the user's historical click sequence is a sequence of access records representing the user's historical browsing information, and the two-level list is the enhanced access record sequence.

[0097] Step S4: Initialize the representation vector of each node in the directed acyclic graph using the GloVe algorithm.

[0098] Step S5: Use the attention mechanism to calculate the final representation vector d of the leaf nodes in the directed acyclic graph, so that the representation vectors of the leaf nodes form a matrix D;

[0099] Step S6: Perform a dot product between matrix D and the user access records represented by multi-hot vectors to obtain the representation vector of each user access record; perform a dot product between matrix D and the candidate recommendation information represented by multi-hot vectors to obtain the representation vector of each candidate recommendation information.

[0100] Step S7: Use Doc2Vec to train and obtain the representation vector of the title of each water conservancy information in the user's browsing history; obtain the representation vector of the title of each water conservancy information in the candidate recommendation information.

[0101] Step S8: Based on the text representation vector of the title of each piece of water conservancy information in the user's historical browsing record obtained in Step S7, the representation vector of each user access record obtained in Step S6, and the two vectors corresponding to the candidate recommended information, the click probability of the user when facing the candidate water conservancy information is obtained; the recommended list of candidate recommended information is obtained in descending order of click probability.

[0102] The information recommendation method provided in this application enhances the interpretability of the model. Deep learning can be viewed as a black box; although it has shown excellent performance on many tasks, its persuasiveness hinges on whether the public can understand why the model makes these decisions. This application uses an attention mechanism, which can flexibly integrate domain knowledge according to different levels of importance. By extracting attention weights after model training is complete, the recommendation effect of the model can be explained to a certain extent. Specifically, when the model encounters rarely appearing water conservancy terms in the information, how does it utilize the concept of the ancestor node to which the term node belongs to enhance the user's browsing history representation vector?

[0103] Furthermore, this application embodiment uses a CNN to fuse representation vectors trained by different models. The text representation vector and the user browsing history representation vector are learned by two separate sub-models; simply concatenating these two vectors is inappropriate. This paper employs a convolutional neural network to fuse them, enabling better correlation between user browsing history and candidate recommended content with minimal information loss.

[0104] This application embodiment also provides an information recommendation device 400; Figure 4 A schematic diagram of the structure of the information recommendation device 400 provided in the embodiments of this application is shown below. Figure 4 As shown, the information recommendation device 400 provided in this application embodiment includes:

[0105] Extraction unit 401: Used to extract specialized terminology from each of multiple pieces of information;

[0106] Generation unit 402: is used to generate a recommendation list based on the professional domain terminology, professional domain ontology, user browsing history information, and candidate recommendation information.

[0107] In other embodiments of this application, the generation unit 402 is specifically used to preprocess all the title text of the multiple pieces of information;

[0108] The first model is trained using a training set, and after the first model converges, the seventh representation vector matrix is ​​obtained; wherein, the training set includes: all the title texts of the multiple pieces of information after preprocessing;

[0109] Extract all titles from the user's historical browsing information, and extract all representation vectors corresponding to all titles in the seventh representation vector matrix to obtain the second representation vector matrix;

[0110] The step of obtaining the representation vector for each title text in the candidate recommendation information based on the seventh representation vector matrix and the candidate recommendation information includes:

[0111] Extract all titles from the candidate recommendation information, and extract all representation vectors corresponding to all titles in the seventh representation vector matrix to obtain the representation vector of the title text of each piece of information in the candidate recommendation information.

[0112] Obtain a seventh representation vector matrix; based on the seventh representation vector matrix and the user's browsing history information, obtain a second representation vector matrix; wherein the seventh representation vector matrix is ​​a matrix formed by the representation vectors of each title text in the multiple pieces of information, and the second representation vector matrix is ​​a matrix formed by the representation vectors of each title text in the user's browsing history information; based on the seventh representation vector matrix and candidate recommendation information, obtain the representation vector of each title text in the candidate recommendation information; based on the candidate recommendation information and the professional domain terminology, obtain the representation vector of each piece of information in the candidate recommendation information; based on the first representation vector matrix, the second representation vector matrix, the representation vectors of each title text in the candidate recommendation information, and the representation vectors of each piece of information in the candidate recommendation information, obtain the user's click probability for each piece of information in the candidate recommendation information; based on the user's click probability for each piece of information in the candidate recommendation information, obtain a recommendation list.

[0113] In other embodiments of this application, the generation unit 402 is further specifically configured to: obtain a fifth representation vector matrix based on the professional domain terminology and the user's historical browsing information, wherein the fifth representation vector matrix is ​​a matrix formed by the initialization representation vector of each browsing information in the user's historical browsing information; generate a directed acyclic graph based on the professional domain terminology and the professional domain ontology; obtain a sixth representation vector matrix based on the user's historical browsing information and the directed acyclic graph; wherein the sixth representation vector matrix is ​​a matrix formed by the representation vector of each professional domain term in the professional domain terminology; obtain a first representation vector matrix based on the fifth representation vector matrix and the sixth representation vector matrix; the step of obtaining the representation vector of each piece of information in the candidate recommendation information based on the candidate recommendation information and the professional domain terminology includes: obtaining the representation vector of each piece of information in the candidate recommendation information based on the candidate recommendation information and the sixth representation vector matrix.

[0114] In other embodiments of this application, the generation unit 402 is further specifically used to preprocess all title texts of the multiple pieces of information; train a first model using a training set, and obtain the seventh representation vector matrix after the first model converges; wherein, the training set includes: all title texts of the preprocessed multiple pieces of information; extract all titles of the user's historical browsing information, and extract all representation vectors corresponding to all titles of the user's historical browsing information from the seventh representation vector matrix to obtain the second representation vector matrix; the step of obtaining the representation vector of each title text in the candidate recommendation information based on the seventh representation vector matrix and the candidate recommendation information includes: extracting all titles in the candidate recommendation information, and extracting all representation vectors corresponding to all titles of the candidate recommendation information from the seventh representation vector matrix to obtain the representation vector of the title text of each piece of information in the candidate recommendation information.

[0115] In other embodiments of this application, the training set further includes: a corpus dataset.

[0116] In other embodiments of this application, the generation unit 402 is further specifically used to obtain the representation vector of each piece of information in the candidate recommendation information when the user faces it through a first convolutional neural network based on the first representation vector matrix, the second representation vector matrix, the representation vector of each title text in the candidate recommendation information, and the representation vector of each piece of information in the candidate recommendation information; and to obtain the click probability of each piece of information in the candidate recommendation information when the user faces it through a multilayer perceptron based on the representation vector of each piece of information in the candidate recommendation information.

[0117] In other embodiments of this application, the generation unit 402 is further specifically configured to: enhance each browsing information in the user's historical browsing information based on the user's historical browsing information and the directed acyclic graph, to obtain enhanced browsing information for each user; obtain a co-occurrence matrix composed of all professional domain terms and professional domain ontology based on the professional domain terms and the enhanced browsing records for each user; obtain the initial representation vector of each node in the directed acyclic graph through a second model based on the co-occurrence matrix; and obtain the representation vector of each professional domain term in the professional domain terms through a second convolutional neural network based on the initial representation vector of each node in the directed acyclic graph and the directed acyclic graph, forming the sixth representation vector matrix.

[0118] Those skilled in the art should understand that Figure 4 The functions of each unit in the information recommendation device shown can be understood by referring to the relevant description of the aforementioned method. Figure 4 The functions of each unit in the information recommendation device shown can be implemented by a program running on a processor or by specific logic circuits.

[0119] Figure 5 This is a schematic structural diagram of an electronic device 500 provided in an embodiment of this application. Figure 5 The illustrated electronic device 500 includes a processor 510, which can call and run computer programs from memory to implement the methods in the embodiments of this application.

[0120] Optionally, such as Figure 5 As shown, the electronic device 500 may further include a memory 520. The processor 510 can retrieve and run computer programs from the memory 520 to implement the methods described in the embodiments of this application.

[0121] The memory 520 can be a separate device independent of the processor 510, or it can be integrated into the processor 510.

[0122] Optionally, such as Figure 5 As shown, the electronic device 500 may also include a transceiver 530, which the processor 510 can control to communicate with other devices. Specifically, it can send information or data to other devices or receive information or data sent by other devices.

[0123] The transceiver 530 may include a transmitter and a receiver. The transceiver 530 may further include antennas, and the number of antennas may be one or more.

[0124] The electronic device 500 may specifically be an information recommendation device in the embodiments of this application, and the electronic device 500 may implement the corresponding processes implemented by the information recommendation device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0125] Figure 6 This is a schematic structural diagram of the chip according to an embodiment of this application. Figure 6 The chip 600 shown includes a processor 610, which can call and run computer programs from memory to implement the methods in the embodiments of this application.

[0126] Optionally, such as Figure 6 As shown, chip 600 may further include memory 620. Processor 610 can retrieve and run computer programs from memory 620 to implement the methods described in this embodiment.

[0127] The memory 620 can be a separate device independent of the processor 610, or it can be integrated into the processor 610.

[0128] Optionally, the chip 600 may also include an input interface 630. The processor 610 can control the input interface 630 to communicate with other devices or chips; specifically, it can acquire information or data sent by other devices or chips.

[0129] Optionally, the chip 600 may also include an output interface 640. The processor 610 can control the output interface 640 to communicate with other devices or chips, specifically, to output information or data to other devices or chips.

[0130] This chip can be applied to the information recommendation device in the embodiments of this application, and the chip can implement the corresponding processes implemented by the information recommendation device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0131] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0132] It should be understood that the processor in the embodiments of this application may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method embodiments can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor described above can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0133] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0134] It should be understood that the above-described memory is exemplary and not a limiting description. For example, the memory in the embodiments of this application may also be static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM), etc. That is to say, the memory in the embodiments of this application is intended to include, but is not limited to, these and any other suitable types of memory.

[0135] This application also provides a computer-readable storage medium for storing a computer program. This computer-readable storage medium can be applied to the information recommendation apparatus in the embodiments of this application, and the computer program causes a computer to execute the corresponding processes implemented by the information recommendation apparatus in the various methods of the embodiments of this application; for brevity, these will not be elaborated further here.

[0136] This application also provides a computer program product, including computer program instructions. This computer program product can be applied to the information recommendation device in this application, and the computer program instructions cause a computer to execute the corresponding processes implemented by the information recommendation device in the various methods of this application. For simplicity, further details are omitted here.

[0137] This application also provides a computer program. This computer program can be applied to the information recommendation device in this application. When the computer program runs on a computer, it causes the computer to execute the corresponding processes implemented by the information recommendation device in the various methods of this application. For simplicity, these will not be elaborated further here.

[0138] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0139] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0140] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0141] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0142] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0143] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0144] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An information recommendation method, characterized in that, include: Extract specialized terminology from each of multiple pieces of information; Based on the aforementioned professional domain terminology, professional domain ontology, and user browsing history information, a first representation vector matrix is ​​obtained; wherein, the first representation vector matrix is ​​a matrix formed by the representation vectors of each browsing information in the user's browsing history information. Obtain a seventh representation vector matrix, and based on the seventh representation vector matrix and the user's browsing history information, obtain a second representation vector matrix; wherein, the seventh representation vector matrix is ​​a matrix formed by the representation vectors of each title text in the multiple pieces of information, and the second representation vector matrix is ​​a matrix formed by the representation vectors of each title text in the user's browsing history information; Based on the seventh representation vector matrix and the candidate recommendation information, the representation vector of each title text in the candidate recommendation information is obtained; Based on the candidate recommendation information and the professional domain terminology, a representation vector is obtained for each piece of information in the candidate recommendation information; Based on the first representation vector matrix, the second representation vector matrix, the representation vector of each title text in the candidate recommendation information, and the representation vector of each piece of information in the candidate recommendation information, the click probability of the user facing each piece of information in the candidate recommendation information is obtained; A recommendation list is obtained based on the click probability of each piece of information in the candidate recommendation information.

2. The information recommendation method according to claim 1, characterized in that, The first representation vector matrix obtained based on the professional domain terminology, the professional domain ontology, and the user's browsing history information includes: Based on the aforementioned professional terminology and the user's historical browsing information, a fifth representation vector matrix is ​​obtained, wherein the fifth representation vector matrix is ​​a matrix formed by the initial representation vector of each browsing information in the user's historical browsing information; Based on the terminology and ontology of the aforementioned professional fields, a directed acyclic graph is generated; Based on the user's browsing history information and the directed acyclic graph, a sixth representation vector matrix is ​​obtained; wherein, the sixth representation vector matrix is ​​a matrix formed by the representation vectors of each professional domain term in the professional domain terminology; Based on the fifth representation vector matrix and the sixth representation vector matrix, the first representation vector matrix is ​​obtained; The step of obtaining a representation vector for each piece of information in the candidate recommendation information based on the candidate recommendation information and the professional domain terminology includes: Based on the candidate recommendation information and the sixth representation vector matrix, the representation vector of each piece of information in the candidate recommendation information is obtained.

3. The information recommendation method according to claim 2, characterized in that, The step of obtaining the seventh representation vector matrix, and obtaining the second representation vector matrix based on the seventh representation vector matrix and the user's historical browsing information, includes: Preprocess all title texts of the multiple messages; The first model is trained using a training set, and after the first model converges, the seventh representation vector matrix is ​​obtained; wherein, the training set includes: all the title texts of the multiple pieces of information after preprocessing; Extract all titles from the user's historical browsing information, and extract all representation vectors corresponding to all titles in the seventh representation vector matrix to obtain the second representation vector matrix; The step of obtaining the representation vector for each title text in the candidate recommendation information based on the seventh representation vector matrix and the candidate recommendation information includes: Extract all titles from the candidate recommendation information, and extract all representation vectors corresponding to all titles in the seventh representation vector matrix to obtain the representation vector of the title text of each piece of information in the candidate recommendation information.

4. The information recommendation method according to claim 3, characterized in that, The training set also includes: a corpus dataset.

5. The information recommendation method according to any one of claims 2-4, characterized in that, The step of obtaining the click probability of a user facing each piece of information in the candidate recommendation information based on the first representation vector matrix, the second representation vector matrix, the representation vector of each title text in the candidate recommendation information, and the representation vector of each piece of information in the candidate recommendation information includes: Based on the first representation vector matrix, the second representation vector matrix, the representation vector of each title text in the candidate recommendation information, and the representation vector of each piece of information in the candidate recommendation information, the representation vector when the user faces each piece of information in the candidate recommendation information is obtained through the first convolutional neural network. Based on the user's representation vector when facing each piece of information in the candidate recommendation information, the click probability of the user when facing each piece of information in the candidate recommendation information is obtained through a multilayer perceptron.

6. The information recommendation method according to any one of claims 2-4, characterized in that, The sixth representation vector matrix obtained based on the user's historical browsing information and the directed acyclic graph includes: Based on the user's historical browsing information and the directed acyclic graph, each browsing information in the user's historical browsing information is enhanced to obtain the enhanced browsing information of each user. Based on the professional domain terminology and the enhanced user's browsing history for each time, a co-occurrence matrix composed of all professional domain terminology and professional domain ontology is obtained; Based on the co-occurrence matrix, the initialization representation vector of each node in the directed acyclic graph is obtained through the second model; Based on the initial representation vector of each node in the directed acyclic graph and the directed acyclic graph, the representation vector of each professional domain term in the professional domain terminology is obtained through the second convolutional neural network, forming the sixth representation vector matrix.

7. An information recommendation device, characterized in that, include: Extraction Unit: Used to extract specialized terminology from each of multiple pieces of information; Generation unit: used to obtain a first representation vector matrix based on the professional domain terminology, professional domain ontology and user browsing history information; wherein, the first representation vector matrix is ​​a matrix formed by the representation vectors of each browsing information in the user browsing history information; Obtain a seventh representation vector matrix, and based on the seventh representation vector matrix and the user's browsing history information, obtain a second representation vector matrix; wherein, the seventh representation vector matrix is ​​a matrix formed by the representation vectors of each title text in the multiple pieces of information, and the second representation vector matrix is ​​a matrix formed by the representation vectors of each title text in the user's browsing history information; Based on the seventh representation vector matrix and the candidate recommendation information, the representation vector of each title text in the candidate recommendation information is obtained; Based on the candidate recommendation information and the professional domain terminology, a representation vector is obtained for each piece of information in the candidate recommendation information; Based on the first representation vector matrix, the second representation vector matrix, the representation vector of each title text in the candidate recommendation information, and the representation vector of each piece of information in the candidate recommendation information, the click probability of the user facing each piece of information in the candidate recommendation information is obtained; A recommendation list is obtained based on the click probability of each piece of information in the candidate recommendation information.

8. An electronic device, characterized in that, include: A processor and a memory, the memory for storing a computer program, the processor for calling and running the computer program stored in the memory to perform the information recommendation method as described in any one of claims 1-6.

9. A chip, characterized in that, include: A processor for retrieving and running a computer program from memory, causing a device with the chip mounted to perform the information recommendation method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, Used to store a computer program that causes a computer to perform the information recommendation method as described in any one of claims 1-6.

Images