Personalized recommendation method and system for power industry science and technology frontier and application articles thereof

Abstract

The invention provides a personalized recommendation method and system for power industry science and technology frontier and application articles thereof, a storage medium and electronic equipment, and relates to the field of personalized recommendation. According to the method, related domain coefficients are utilized, sentences, related to the personalized knowledge graph, of an article to be recommended are screened out, and a new text after information filtering is generated; the method comprises the following steps: taking a title of an article to be recommended and a new text as input of a BERT model, respectively obtaining title representation and a plurality of sentence representation, introducing a sentence-level attention mechanism, obtaining an attention weight of the title representation for each sentence representation so as to obtain weighted sentence representation, and finally inputting the weighted sentence representation into a classifier. According to the method, sentence screening, preliminary filtering of irrelevant information, introduction of the sentence level attention model and re-filtering of the preliminarily screened sentences are carried out by utilizing the constructed personalized knowledge graph, so that interference of information irrelevant to themes is greatly reduced, the performance of a classification model is improved, and a personalized recommendation result with high accuracy is quickly generated.

Description

Technical Field

[0001] The present invention relates to the field of personalized recommendation, and in particular to a personalized recommendation method, system, storage medium and electronic device for cutting-edge scientific and technological frontiers and application articles in the electric power industry. Background Art

[0002] In today's era of information explosion, major forums have become an ocean of information, with a large number of articles and complex content. This phenomenon has brought many problems, the most prominent of which is that it is difficult for users to obtain the information they are interested in in a timely manner when faced with massive amounts of information.

[0003] For users such as researchers, forums have certain advantages. Their articles are updated very quickly, and they can keep up to date with the latest research trends in their fields. However, this also brings new challenges. Due to the complexity of forum articles, it is difficult to quickly and comprehensively find professional content related to one's research field. In academic research, time and energy are both valuable resources, and a lot of time is needed to screen and identify this information, which to a certain extent affects their research efficiency. Therefore, it is very valuable to recommend public account articles based on each person's interests and professional fields. This personalized recommendation method can help users filter out a large amount of irrelevant information. For users such as researchers, personalized recommendations enable them to more efficiently acquire professional knowledge related to their research fields, improve the accuracy and effectiveness of knowledge acquisition, and thus better promote the progress of academic research. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the deficiencies in the prior art, the present invention provides a method, system, storage medium and electronic device for personalized recommendation of cutting-edge scientific and technological articles in the electric power industry and their applications, solving the technical problem that it is difficult for users to obtain information of interest in a timely manner.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0008] A personalized recommendation method for power industry science and technology frontiers and their application articles, including:

[0009] Build a knowledge graph for the power industry;

[0010] Based on the power industry knowledge graph, user preference information is integrated to build a personalized knowledge graph;

[0011] Using the related field coefficient, the sentences related to the personalized knowledge graph in the article to be recommended are screened out to generate new text after information filtering;

[0012] The title of the article to be recommended and the new text are used as the input of the BERT model to obtain the title representation and several sentence representations respectively, and introduce the sentence-level attention mechanism to obtain the attention weight of the title representation to each sentence representation to obtain the weighted sentence representation;

[0013] The weighted sentence representation and title representation are concatenated as the input of the classifier to obtain the category probability of whether the user is interested in the recommended article, so as to generate user personalized recommendation results.

[0014] Preferably, the construction of the electric power industry knowledge graph includes:

[0015] Collect and preprocess articles on the cutting-edge technology and its applications in the power industry;

[0016] Based on the preprocessed articles, the graph ontology is constructed, including:

[0017] The word frequency-inverse text frequency algorithm and TextRank algorithm are used to extract the keywords of the article, and the k-Means clustering algorithm is used to cluster the articles. The preliminary domain concepts are obtained by manually observing the keywords and clustering results. The domain concepts are modified and the domain relations and their constraints are defined in combination with the general graph, and several graph ontologies are constructed.

[0018] Based on the ontology structure of the graph ontology, entity data is obtained, and the relationship between entities is obtained through manual screening in combination with semi-supervised and remotely supervised methods to construct the power industry knowledge graph.

[0019] Preferably, the method of building a personalized knowledge graph based on the power industry knowledge graph and integrating user preference information includes:

[0020] Classify the user's historical browsing articles according to different topics, and extract key information of each historical browsing article; wherein the key information at least includes key concepts, technologies and trends;

[0021] Traverse all historically browsed articles on each article topic, and search for corresponding nodes or node sets in the power industry knowledge graph. If there is no completely corresponding node, create a new node;

[0022] The key information of each historically browsed article is associated with the corresponding node, and a new relationship connection is established when multiple nodes are involved.

[0023] Preferably, after the personalized knowledge graph is constructed, the personalized knowledge graph is regularly optimized and updated based on new historical browsing articles.

[0024] Preferably, the method of using the related field coefficient to filter out sentences related to the article to be recommended and the personalized knowledge graph to generate a new text after information filtering includes:

[0025] Preprocessing the article to be recommended to obtain a vocabulary list, and counting the frequency of each word in the vocabulary list appearing in the article to be recommended;

[0026] For each word in the vocabulary list, find a corresponding node in the personalized knowledge graph and obtain the betweenness centrality of the node;

[0027] For each sentence of the article to be recommended, add the product of the betweenness centrality and the word frequency of all the words in each sentence to get the relevant field coefficient;

[0028] Sort all sentences according to the domain relevance coefficient, select the top n sentences with the highest score values, and connect them to generate a new text after information filtering.

[0029] Preferably, the sentence-level attention mechanism is introduced to obtain the attention weight of the title representation to each sentence representation to obtain the weighted sentence representation; expressed as:

[0030]

[0031] v seq =∑a i v i ,i∈{1,2,...n}

[0032] Among them, a i is the attention weight, subscript i is the sentence representation index; Q is the vector of the attention query, and the title representation v is taken title ; K i For the vector matrix that needs to be assigned different attention weights, take the sentence representation v in the new text i , the superscript T indicates transposition; d is the scaling factor; v seq is the weighted sentence representation.

[0033] A personalized recommendation system for power industry science and technology frontiers and their application articles, including:

[0034] The first building module is used to build a knowledge graph for the power industry;

[0035] The second construction module is used to build a personalized knowledge graph based on the power industry knowledge graph and integrating user preference information;

[0036] A screening module is used to screen out sentences related to the personalized knowledge graph in the article to be recommended by using the relevant field coefficient, and generate a new text after information filtering;

[0037] The fusion module is used to take the title of the article to be recommended and the new text as the input of the BERT model, obtain the title representation and several sentence representations respectively, and introduce the sentence-level attention mechanism to obtain the attention weight of the title representation to each sentence representation, so as to obtain the weighted sentence representation;

[0038] The classification module is used to concatenate the weighted sentence representation and title representation as the input of the classifier, obtain the category probability of whether the user is interested in the recommended article, and generate user personalized recommendation results.

[0039] A computer-readable storage medium stores a computer program, which can implement the personalized recommendation method for the frontier of science and technology and its application articles in the power industry as described above when a processor executes the computer program.

[0040] An electronic device, comprising:

[0041] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a method for executing the personalized recommendation method of the electric power industry scientific and technological frontiers and their application articles as described above.

[0042] (III) Beneficial effects

[0043] The present invention provides a personalized recommendation method, system, storage medium and electronic device for the frontier of science and technology and its application articles in the power industry. Compared with the prior art, it has the following beneficial effects:

[0044] In the present invention, the relevant field coefficient is used to screen out sentences related to the article to be recommended and the personalized knowledge graph, and a new text is generated after information filtering; the title of the article to be recommended and the new text are used as the input of the BERT model, and the title representation and several sentence representations are obtained respectively, and the sentence-level attention mechanism is introduced to obtain the attention weight of the title representation to each sentence representation, so as to obtain the weighted sentence representation, and finally input it into the classifier. The present invention uses the constructed personalized knowledge graph to screen sentences, preliminarily filters out irrelevant information, introduces a sentence-level attention model, and filters the sentences after the preliminary screening again, greatly reducing the interference of information irrelevant to the topic, thereby improving the performance of the classification model and quickly generating personalized recommendation results with high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1 A block diagram of a personalized recommendation method for power industry science and technology frontiers and application articles provided by an embodiment of the present invention;

[0047] Figure 2 A flowchart for constructing a knowledge graph for the electric power industry provided by an embodiment of the present invention;

[0048] Figure 3 A schematic diagram of a sentence-level attention mechanism provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The embodiments of the present application solve the technical problem that it is difficult for users to obtain information of interest in a timely manner by providing a personalized recommendation method, system, storage medium and electronic device for the frontier science and technology and application articles of the power industry.

[0051] The technical solution in the embodiment of the present application is to solve the above technical problems, and the overall idea is as follows:

[0052] The purpose of the embodiment of the present invention is to provide users with personalized recommendations on the technological frontiers and their applications in the electric power industry. It aims to provide users with personalized, accurate and novel technological frontier and application-related article recommendation services based on their interests and needs, to help them better understand the technological trends in the electric power industry and improve their professional knowledge level. At the same time, it will also provide companies with valuable market information to help them better seize market opportunities.

[0053] The purpose of the embodiment of the present invention is to build a personalized recommendation system for the frontiers of science and technology and their applications in the power industry. The system aims to provide personalized, accurate and novel recommendation services for articles related to the frontiers of science and technology and their applications according to the interests and needs of users.

[0054] Specifically, the objectives of the embodiments of the present invention include the following aspects:

[0055] 1. Solve the problem of information overload: The amount of information in the frontier of science and technology in the power industry and its application fields is huge and growing, and it is often difficult for users to filter out content that meets their interests and needs from the massive data. Therefore, the embodiment of the present invention aims to help users quickly find high-quality articles related to their interests through a personalized recommendation system, thereby alleviating the trouble caused by information overload.

[0056] 2. Improve user experience: Personalized recommendation systems can tailor recommended content to users based on their preferences and behaviors, providing a more personalized and targeted reading experience. By analyzing users’ reading history, likes, and comments, the system can better understand users’ needs, thereby providing recommendations that are more in line with their interests and improving user satisfaction and loyalty.

[0057] 3. Promote the development of science and technology in the power industry: The development of the power industry's science and technology frontiers and their application fields is crucial to the sustainable development of the energy industry. Through the personalized recommendation system, users can keep abreast of the latest scientific and technological progress, innovative applications and research results, promote technical exchanges and cooperation within the industry, and promote scientific and technological innovation and development in the power industry.

[0058] 4. Support decision-making: Decision-makers in the power industry need to understand the latest technological frontiers and application scenarios when formulating strategies and plans. The personalized recommendation system can provide decision-makers with targeted and practical recommendations for technological frontiers and application-related articles, helping them obtain key information and make wise decisions.

[0059] In summary, the embodiment of the present invention aims to build a personalized recommendation system for the frontiers of science and technology and their applications in the power industry to solve the problem of information overload, improve user experience, promote the development of science and technology in the power industry, and support decision-making. Through accurate, diverse and novel recommendation results, it provides users with valuable content related to the frontiers of science and technology and applications, meets the needs of users, and promotes the development and innovation of the power industry.

[0060] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0061] Embodiment 1:

[0062] like Figure 1 As shown, the embodiment of the present invention provides a personalized recommendation method for power industry science and technology frontiers and application articles, including:

[0063] S1. Build a knowledge graph for the power industry;

[0064] S2. Based on the power industry knowledge graph, the user's preference information is integrated to build a personalized knowledge graph;

[0065] S3. Using the relevant field coefficient, filter out sentences related to the recommended article and the personalized knowledge graph, and generate new text after information filtering;

[0066] S4. The title of the article to be recommended and the new text are used as the input of the BERT model to obtain the title representation and several sentence representations respectively, and introduce the sentence-level attention mechanism to obtain the attention weight of the title representation to each sentence representation to obtain the weighted sentence representation;

[0067] S5. The weighted sentence representation and title representation are concatenated as the input of the classifier to obtain the category probability of whether the user is interested in the recommended article, so as to generate user personalized recommendation results.

[0068] The embodiment of the present invention utilizes the constructed personalized knowledge graph to screen sentences, preliminarily filters out irrelevant information, introduces a sentence-level attention model, and re-filters the sentences after the preliminary screening, greatly reducing the interference of information irrelevant to the topic, thereby improving the performance of the classification model and quickly generating personalized recommendation results with high accuracy.

[0069] Next, we will introduce the steps of the above scheme in detail:

[0070] In step S1, a knowledge graph of the power industry is constructed.

[0071] It should be noted that building a knowledge base for the power industry is a complex and systematic process that requires in-depth analysis and integration of a wide range of data and information resources from multiple perspectives. First, it is necessary to understand the core components of the power industry knowledge base. It is not only a platform for storing data, but also a framework for structuring and organizing information related to the power industry. The power industry includes multiple aspects such as power generation, transmission, distribution, and new energy technologies, each of which has its own unique data and information needs. Therefore, building a knowledge base that can meet these needs requires multi-level data collection and processing.

[0072] See also Figure 2 , this step specifically includes the following steps:

[0073] S11. Collect and preprocess articles on the cutting-edge technology and its applications in the power industry.

[0074] Data collection is the first step in building a knowledge base for the power industry. Data for the power industry comes from a wide range of sources, including operational data from power companies, statistics from government departments, research papers from academic journals, and industry reports and analyses. These data cover a wide range of content, from the operation of power generation facilities to economic analysis of energy markets. Collecting this data requires not only extensive and systematic searches, but also effective screening and organization to ensure the accuracy and availability of the data.

[0075] Since the collected data is usually divided into structured data sources and unstructured data sources. Structured data sources mainly collect operational data of power companies, government statistics, industry reports, and internal reports of power companies. Unstructured data requires the use of text mining and natural language processing technology to extract information from unstructured data such as news, academic papers, and professional journals. Here, information data from 68 power industry forums such as State Grid News and Electric Times, as well as power industry-related journals, are collected to obtain a large number of articles with professional knowledge of the power industry.

[0076] S12. Construct a graph ontology based on the preprocessed articles.

[0077] It should be noted that ontology is the core model of the knowledge base of the power industry, which defines the concepts of various entities in the power industry (such as power plants, substations, transmission lines, etc.) and the relationships between them. The ontology construction process usually starts with determining the core concepts and entities, and then defining their attributes and relationships.

[0078] For example, a power plant may have attributes such as power generation capacity, power generation technology, and fuel type, and the power plant and the transmission line may be connected through a power supply relationship.

[0079] By establishing an ontology, a clear framework can be provided for the structuring and information retrieval of the knowledge base. The construction of an ontology usually involves the cooperation of domain experts and knowledge engineers to jointly design and optimize the ontology model to ensure its accuracy and practicality. The embodiment of the present invention constructs a graph ontology by using statistical methods combined with manual induction, drawing on high-quality general graphs and expert guidance.

[0080] Specifically, the graph ontology construction process includes:

[0081] 1) The word frequency-inverse text frequency algorithm and TextRank algorithm are used to extract the keywords of the article; 2) The k-Means clustering algorithm is used to cluster the articles, and preliminary domain concepts are obtained by manually observing the keywords and clustering results; 3) The domain concepts are modified in combination with the general graph, and the domain relationships and their constraints are defined, and several graph ontologies are constructed.

[0082] S13. Based on the ontology structure of the graph ontology, entity data is obtained, and the relationship between entities is obtained through manual screening in combination with semi-supervised and remotely supervised methods to construct the power industry knowledge graph.

[0083] Here, the constructed ontology is then checked and evaluated, and the final popular science ontology is obtained after modification and improvement. According to the defined popular science ontology structure, a large amount of entity data with good structure and high quality is obtained from the knowledge graph publicly available on the Internet, and the relationship between entities is obtained through manual screening by combining semi-supervised and remote supervision methods. Finally, according to the guidance of power industry experts, the constructed index knowledge graph is checked and evaluated.

[0084] In step S2, based on the power industry knowledge graph, the user's preference information is integrated to construct a personalized knowledge graph; including:

[0085] S21. The user's historically browsed articles are classified into different topics, such as new energy power generation technology (including sub-topics such as solar energy, wind energy, and hydropower), smart grid technology, and power market dynamics, etc. The article source and release time can be further recorded. The key information of each historically browsed article can be extracted, and its importance and relevance to the power industry can be further marked.

[0086] S22. Traverse all historically browsed articles on each article topic, and search for corresponding nodes or node sets in the power industry knowledge graph. If there is no completely corresponding node, create a new node.

[0087] S23. Associate the key information of each historically browsed article with the corresponding node, and establish a new relationship connection when multiple nodes are involved.

[0088] In particular, after the personalized knowledge graph is constructed, it may also include:

[0089] S24. Regularly optimize and update the personalized knowledge graph based on new historical browsing articles.

[0090] It should be noted that valuable information filtering is the part of the model in this article that extracts value from text information. It is divided into two steps: First, the sentence correlation coefficient is calculated with the help of the knowledge graph, and the sentences in each text are filtered and selected based on the coefficient. Then, BERT is used as an encoder to extract the features of all sentences in the title and text, and the attention mechanism is used to assign different weights to each sentence to measure their importance to the article.

[0091] For details, please refer to steps S3 to S4:

[0092] In step S3, the relevant field coefficient is used to filter out sentences related to the personalized knowledge graph in the article to be recommended, and generate a new text after information filtering.

[0093] S31. Calculate betweenness centrality:

[0094] The Brandes algorithm is used to calculate the betweenness centrality of each node in the knowledge graph. The basic idea of ​​the Brandes algorithm is to determine the betweenness centrality of a node by calculating the number of times all shortest paths pass through it. The specific calculation formula is as follows:

[0095]

[0096] Among them, C B (v) represents the betweenness centrality of node v, V is the set of all nodes in the knowledge graph, s and t are any two nodes in V except v, σ st represents the total number of shortest paths from node s to node t, σ st (v) represents the number of nodes v passed through in the shortest path from node s to node t.

[0097] S32, field correlation coefficient calculation:

[0098] First, the article to be recommended is preprocessed, including operations such as word segmentation and removal of stop words, to obtain a vocabulary list, and to count the frequency of each word in the vocabulary list appearing in the article to be recommended, that is, the word frequency.

[0099] Secondly, for each word in the vocabulary list, find the corresponding node in the personalized knowledge graph and obtain the betweenness centrality of the node.

[0100] Again, for each sentence of the article to be recommended, add the product of the betweenness centrality and the word frequency of all the words in each sentence to get the relevant field coefficient p. The mathematical expression is:

[0101]

[0102] Among them, tf represents word frequency. Through the above steps, the coefficient of the power industry-related fields based on betweenness centrality and word frequency can be calculated. This coefficient can reflect the degree of relevance of the text to the scientific and technological frontiers of the power industry and its applications.

[0103] Finally, all sentences are sorted according to the domain relevance coefficient, and the top n sentences with the highest score values ​​are selected and connected to generate a new text after information filtering.

[0104] In step S4, the title of the article to be recommended and the new text are used as inputs of the BERT model to obtain title representation and several sentence representations respectively, and a sentence-level attention mechanism is introduced to obtain the attention weight of the title representation to each sentence representation to obtain the weighted sentence representation.

[0105] It should be noted that BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained language model based on the Transformer architecture. It learns rich language knowledge and semantic representations by performing unsupervised learning on large-scale text. In text classification, the main role of BERT is to extract features and learn representations of text. It can convert the input text into low-dimensional dense vector representations, which contain the semantic information of the text. The BERT model consists of a multi-layer bidirectional Transformer decoder, and the core of the Transformer model is the attention mechanism. Transformer captures long-distance dependencies in text through a multi-head attention mechanism (Multi-HeadAttention), which can better obtain the semantic information of the target word and its context in a variety of semantic scenarios. The calculation formula of H is:

[0106] H = Concat(h 1, h2,…,h i )*W 0

[0107] Among them, W 0 represents the weight matrix, which redistributes the weights for the output vector; h i represents the output vector of the i-th head, h i The calculation formula is:

[0108] h i =Attention(Q i ,K i ,V i )

[0109]

[0110] Among them, Q i ,K i ,V i are query, key, and value matrices, d k is the dimension of the key. Through the multi-head attention mechanism, we can simultaneously focus on different positions and semantic information in the text.

[0111] In the BERT model, [CLS] is a special character that does not have any semantics. It can be used to represent the comprehensive semantic information of the input text. The embodiment of the present invention uses the embedded representation of this character as the feature vector v of the input text. The new text and title obtained in the previous step are respectively input into the BERT model, and the vector where the [CLS] character is located is taken as the representation of each sentence and the title representation, which are v and v respectively. doc =[v1,v2,...,v n ]、v title .

[0112] It should be noted that not every short sentence in a long text contains semantic information that is equally relevant to the text topic. In order to evaluate the importance of semantic information of different short sentences in a long text for the classification task, the attention mechanism is used in this model to obtain the attention weight of the title for each short sentence. The sentence vector of each short sentence is weighted and summed according to the attention weight of each short sentence to obtain a short sentence feature vector with more significant classification features. In a long natural language text, the title at the beginning of the text often contains the topic sentence of the article or a general sentence. Its semantic content helps to extract and analyze the content features of the text. Therefore, the model concatenates the short sentence feature vector and the title feature vector as the text classification feature vector.

[0113] Therefore, when obtaining the representation v of each sentence doc =[v1,v2,...,v n ] and title representation title Afterwards, Figure 3 As shown, this step introduces a sentence-level attention mechanism to obtain the attention weight of the title representation to each sentence representation to obtain the weighted sentence representation; it is expressed as:

[0114]

[0115] v seq =∑a i v i ,i∈{1,2,...n}

[0116] Among them, a i is the attention weight, subscript i is the sentence representation index; Q is the vector of the attention query, and the title representation v is taken title ; K i For the vector matrix that needs to be assigned different attention weights, take the sentence representation v in the new text i , the superscript T indicates transposition; d is the scaling factor; v seq is the weighted sentence representation.

[0117] In step S5, the weighted sentence representation and title representation are concatenated as the input of the classifier to obtain the category probability of whether the user is interested in the recommended article, so as to generate a user personalized recommendation result.

[0118] like Figure 3 As shown in Figure 2, through the attention mechanism, the attention of the sentence vector corresponding to the classification is obtained. Then the sentence vector is aggregated through attention to obtain v output , which contains the text information after being double-filtered by the filter and the attention mechanism.

[0119] In this step, v output Input into the classifier consisting of a fully connected layer and a Softmax function to obtain the category probability:

[0120] p=Softmax(Wv output +b)

[0121] Finally, based on the category probability of whether the user is interested in the recommended article, a user-personalized recommendation result is generated.

[0122] For example: the closer the category probability is to 1, the more interested the user is, and the article to be recommended can be recommended to the user; conversely, the closer the category probability is to 1, the less interested the user is, and the article to be recommended may not be recommended to the user.

[0123] So far, the embodiment of the present invention has completed the entire process of the personalized recommendation method for the frontier science and technology and its application articles in the power industry.

[0124] Embodiment 2:

[0125] The embodiment of the present invention provides a personalized recommendation system for power industry science and technology frontiers and their application articles, including:

[0126] The first building module is used to build a knowledge graph for the power industry;

[0127] The second construction module is used to build a personalized knowledge graph based on the power industry knowledge graph and integrating user preference information;

[0128] A screening module is used to screen out sentences related to the personalized knowledge graph in the article to be recommended by using the relevant field coefficient, and generate a new text after information filtering;

[0129] The fusion module is used to take the title of the article to be recommended and the new text as the input of the BERT model, obtain the title representation and several sentence representations respectively, and introduce the sentence-level attention mechanism to obtain the attention weight of the title representation to each sentence representation, so as to obtain the weighted sentence representation;

[0130] The classification module is used to concatenate the weighted sentence representation and title representation as the input of the classifier, obtain the category probability of whether the user is interested in the recommended article, and generate user personalized recommendation results.

[0131] Embodiment 3:

[0132] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When a processor executes the computer program, the personalized recommendation method for the frontier science and technology of the electric power industry and its application articles as described in Example 1 can be implemented.

[0133] Embodiment 4:

[0134] An embodiment of the present invention provides an electronic device, including:

[0135] One or more processors; a memory; and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a method for executing the personalized recommendation method of the electric power industry scientific and technological frontiers and their application articles as described in Example 1.

[0136] It is understandable that the personalized recommendation system for the electric power industry scientific and technological frontiers and their application articles, computer-readable storage medium and electronic device provided in the embodiments of the present invention correspond to the provided personalized recommendation method for the electric power industry scientific and technological frontiers and their application articles, and the explanations, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the personalized recommendation method for the electric power industry scientific and technological frontiers and their application articles, which will not be repeated here.

[0137] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0138] 1. The embodiment of the present invention uses the constructed personalized knowledge graph to screen sentences, preliminarily filters out irrelevant information, introduces a sentence-level attention model, and re-filters the sentences after the preliminary screening, greatly reducing the interference of information irrelevant to the topic, thereby improving the performance of the classification model and quickly generating personalized recommendation results with high accuracy.

[0139] 2. The embodiment of the present invention combines knowledge graphs for sentence screening: In view of the problem that the length of popular science articles makes it difficult for the model to focus on key information and the poor classification performance of traditional models, it is proposed to combine knowledge graphs for sentence screening to identify the value of professional generated content in the power field. Filter out irrelevant information, reduce the interference of irrelevant information to the subject, and improve the classification performance of the model.

[0140] 3. The embodiment of the present invention introduces a sentence-level attention model: In order to more effectively reduce the interference of information irrelevant to the topic, thereby improving the performance of the classification model, this article introduces a sentence-level attention model. The function of this model is to re-filter the sentence information after preliminary screening. Specifically, different sentences in an article contribute differently to the understanding of the article content. The sentence-level attention model can calculate the contribution of sentence vectors to distinguishing the categories of articles. In this way, more attention can be paid to sentence information that is more important to the classification task, and less relevant information can be further removed, thereby improving the accuracy and performance of the model.

[0141] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0142] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A personalized recommendation method for power industry science and technology frontiers and their application articles, characterized by: include: Construct a knowledge graph for the power industry; Based on the power industry knowledge graph, user preference information is integrated to build a personalized knowledge graph; Using the relevant field coefficient, the sentences related to the personalized knowledge graph in the article to be recommended are screened out to generate new text after information filtering; The title of the article to be recommended and the new text are used as the input of the BERT model to obtain the title representation and several sentence representations respectively, and introduce the sentence-level attention mechanism to obtain the attention weight of the title representation to each sentence representation to obtain the weighted sentence representation; The weighted sentence representation and title representation are concatenated as the input of the classifier to obtain the category probability of whether the user is interested in the recommended article, so as to generate user personalized recommendation results.

2. The personalized recommendation method according to claim 1, characterized in that: The construction of the power industry knowledge graph includes: Collect and preprocess articles on the cutting-edge technology and its applications in the power industry; Based on the preprocessed articles, the graph ontology is constructed, including: The word frequency-inverse text frequency algorithm and TextRank algorithm are used to extract the keywords of the article, and the k-Means clustering algorithm is used to cluster the articles. The preliminary domain concepts are obtained by manually observing the keywords and clustering results. The domain concepts are modified and the domain relations and their constraints are defined in combination with the general graph, and several graph ontologies are constructed. Based on the ontology structure of the graph ontology, entity data is obtained, and the relationship between entities is obtained through manual screening in combination with semi-supervised and remotely supervised methods to construct the power industry knowledge graph.

3. The personalized recommendation method according to claim 1, characterized in that: The method of constructing a personalized knowledge graph based on the power industry knowledge graph and integrating user preference information includes: Classify the user's historical browsing articles according to different topics, and extract key information of each historical browsing article; wherein the key information at least includes key concepts, technologies and trends; Traverse all historically browsed articles on each article topic, and search for corresponding nodes or node sets in the power industry knowledge graph. If there is no completely corresponding node, create a new node; The key information of each historically browsed article is associated with the corresponding node, and a new relationship connection is established when multiple nodes are involved.

4. The personalized recommendation method according to claim 3, characterized in that: After the personalized knowledge graph is constructed, the personalized knowledge graph is regularly optimized and updated based on new historical browsing articles.

5. The personalized recommendation method according to claim 1, characterized in that: The method uses the related field coefficient to filter out sentences related to the article to be recommended and the personalized knowledge graph, and generates a new text after information filtering; including: Preprocess the article to be recommended to obtain a vocabulary list, and count the frequency of each word in the vocabulary list appearing in the article to be recommended; For each word in the vocabulary list, find a corresponding node in the personalized knowledge graph and obtain the betweenness centrality of the node; For each sentence of the article to be recommended, add the product of the betweenness centrality and the word frequency of all the words in each sentence to get the relevant field coefficient; Sort all sentences according to the domain relevance coefficient, select the top n sentences with the highest score values, and connect them to generate a new text after information filtering.

6. The personalized recommendation method according to claim 5, characterized in that: The sentence-level attention mechanism is introduced to obtain the attention weight of the title representation to each sentence representation to obtain the weighted sentence representation; it is expressed as: v seq =∑a i v i ,i∈{1,2,...n} Among them, a i is the attention weight, subscript i is the sentence representation index; Q is the vector of the attention query, and the title representation v is taken title ; K i For the vector matrix that needs to be assigned different attention weights, take the sentence representation v in the new text i , the superscript T indicates transposition; d is the scaling factor; v seq is the weighted sentence representation.

7. A personalized recommendation system for power industry science and technology frontiers and their application articles, characterized by: include: The first building module is used to build a knowledge graph for the power industry; The second construction module is used to build a personalized knowledge graph based on the power industry knowledge graph and integrating user preference information; A screening module is used to screen out sentences related to the personalized knowledge graph in the article to be recommended by using the relevant field coefficient, and generate a new text after information filtering; The fusion module is used to take the title of the article to be recommended and the new text as the input of the BERT model, obtain the title representation and several sentence representations respectively, and introduce the sentence-level attention mechanism to obtain the attention weight of the title representation to each sentence representation, so as to obtain the weighted sentence representation; The classification module is used to concatenate the weighted sentence representation and title representation as the input of the classifier, obtain the category probability of whether the user is interested in the recommended article, and generate user personalized recommendation results.

8. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when a processor executes the computer program, a personalized recommendation method for the frontier science and technology of the electric power industry and its application articles as described in any one of claims 1 to 6 can be implemented.

9. An electronic device, characterized in that: include: one or more processors; Memory; And one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the programs include a method for executing the personalized recommendation method of the electric power industry scientific and technological frontier and its application articles as described in any one of claims 1 to 6.

Images