Intellectual property data retrieval method and system

Abstract

The invention relates to the technical field of intelligent information retrieval, in particular to an intellectual property data retrieval method and system. According to the method, articles in a database are clustered according to similar characteristics of all text keyword vectors among the articles and in combination with similar characteristics of the number of text keywords; further acquiring a user retrieval tendency parameter of the target text keyword, and acquiring a retrieval matching degree of the target text keyword in combination with similar characteristics of the retrieval keyword and the target text keyword and a reproduction time interval of the retrieval keyword; and finally, based on the retrieval matching degree of the text keyword, utilizing a preset recommendation algorithm to obtain a retrieval result. According to the method, the articles are clustered by analyzing the correlation among the articles, traditional keyword matching is corrected by means of user behavior habits of the same enterprise, the business scope of the enterprise and personal retrieval characteristics of retrieval users, personalized retrieval results are provided, and the retrieval efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent information retrieval, and in particular to an intellectual property data retrieval method and system. Background Art

[0002] Intellectual property data retrieval is based on the deep integration of modern information technology and intellectual property management needs, aiming to provide efficient and accurate intellectual property information retrieval services. With the advancement of global economic integration and the acceleration of scientific and technological innovation, intellectual property plays an increasingly important role in corporate competition and national development, and the requirements for intellectual property retrieval are also getting higher and higher.

[0003] The existing intellectual property data retrieval method is relatively traditional, searching through keyword matching, classification numbers and technical issues. The existing search method is prone to a large number of duplicate files. Users need to query a large number of files to find target matching files. The search matching accuracy is not high. It takes a long time for users to retrieve the intellectual property information files of target matching files. The search efficiency is low and it is difficult to meet the efficient work requirements of modern society. Summary of the invention

[0004] In order to solve the technical problem of low retrieval efficiency of existing intellectual property data retrieval methods, the purpose of the present invention is to provide an intellectual property data retrieval method and system, and the technical solution adopted is as follows:

[0005] A method for retrieving intellectual property data, the method comprising:

[0006] Obtaining the user's identity information and the user's search keywords; the identity information at least includes the business scope of the user's enterprise; obtaining the business keyword vector of the business scope; obtaining the text keywords and text keyword vectors of the articles in the intellectual property information database;

[0007] According to the similarity characteristics of all text keyword vectors between any two articles, combined with the similarity characteristics of the number of text keywords, the correlation parameters between the articles are obtained; and the articles in the database are clustered according to the correlation parameters between the articles;

[0008] Select any of the text keywords as the target text keyword; obtain the user search tendency parameter of the target text keyword based on the similarity characteristics between the historical search word vectors of other users of the same enterprise and the target text keyword vector, combined with the similarity characteristics between all the text keyword vectors and all the business keyword vectors in the cluster to which the target text keyword belongs; obtain the search matching degree of the target text keyword based on the similarity characteristics between the search keyword and the target text keyword, combined with the recurrence time interval of the search keyword and the user search tendency parameter;

[0009] The search results are obtained by using a preset recommendation algorithm based on the search matching degree of the text keywords.

[0010] Furthermore, the method for obtaining the correlation parameters between the articles includes:

[0011] Pair the text keywords between any two different articles to obtain a similarity coefficient between the two paired text keywords; obtain the sum of the similarity coefficients of all paired text keyword vectors between the two articles as the first correlation sub-parameter between the two articles;

[0012] According to the similarity feature of the number of text keywords between the two articles, obtaining the second related sub-parameter between the two articles;

[0013] According to the quantity characteristics of the same characters in the titles of the two articles, the third correlation sub-parameter between the two articles is obtained;

[0014] A correlation parameter between articles is obtained according to the first correlation sub-parameter, the second correlation sub-parameter and the third correlation sub-parameter, wherein the first correlation sub-parameter, the second correlation sub-parameter and the third correlation sub-parameter are all positively correlated with the correlation parameter.

[0015] Furthermore, the method of clustering articles in a database according to the correlation parameters between articles includes:

[0016] The correlation parameter is negatively correlated and mapped to be the cluster distance between the corresponding two articles, and the articles in the database are clustered using a preset clustering algorithm.

[0017] Furthermore, the method for obtaining the user search tendency parameter includes:

[0018] Obtaining a first tendency sub-parameter of the target text keyword based on similar features between the historical search word vectors of other users of the same enterprise and the target text keyword vector;

[0019] Acquire a second tendency sub-parameter of the target text keyword according to similar features of all the text keyword vectors and all the business keyword vectors in the cluster to which the target text keyword belongs;

[0020] According to the first tendency sub-parameter and the second tendency sub-parameter of the target text keyword, a user search tendency parameter of the target text keyword is obtained; the first tendency sub-parameter and the second tendency sub-parameter are both positively correlated with the user search tendency parameter.

[0021] Furthermore, the method for obtaining the first tendency sub-parameter includes:

[0022] The average value of the Euclidean distances between all historical search word vectors of other users of the same enterprise and the target text keyword vector is obtained, and the average value of the Euclidean distances is negatively correlated and mapped as the first tendency sub-parameter of the target text keyword.

[0023] Furthermore, the method for obtaining the retrieval matching degree includes:

[0024] Acquire similar parameters between each of the search keywords and the target text keywords as matching sub-parameters between the target text keywords and each of the search keywords;

[0025] After negative correlation mapping, the recurrence time interval of each of the search keywords is used as the time reference weight of each of the search keywords;

[0026] The retrieval matching degree of the target text keyword is obtained based on the matching sub-parameters of the target text keyword and each of the retrieval keywords, the time reference weight of each of the retrieval keywords and the user retrieval tendency parameters of the target text keyword; the matching sub-parameters, the time reference weight and the user retrieval tendency parameters are all positively correlated with the retrieval matching degree.

[0027] Furthermore, the preset recommendation algorithm is a reinforcement learning recommendation algorithm.

[0028] Furthermore, the method for obtaining text keywords of the article includes:

[0029] At least the title, keywords, abstract and technical issues of the article are decomposed using a preset word segmentation algorithm to obtain the text keywords of the article.

[0030] Furthermore, the preset clustering algorithm is an Isodata clustering algorithm.

[0031] The present invention also proposes an intellectual property data retrieval system, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any one of the steps of the intellectual property data retrieval method when executing the computer program.

[0032] The present invention has the following beneficial effects:

[0033] The present invention first obtains the identity information of the user and the user's search keywords, business keyword vectors, text keywords and text keyword vectors of the article, so as to provide a basis for subsequent analysis; further, according to the similarity characteristics of all text keyword vectors between any two articles, combined with the similarity characteristics of the number of text keywords, the articles in the database are clustered, and the related articles are classified, which is conducive to rapid retrieval and improves retrieval efficiency; further, by analyzing the similarity characteristics of text keywords and business keywords in the cluster clusters, and with the help of the search behavior habits of users of the same enterprise, the user search tendency parameters of the target text keywords are obtained, which increases the multi-dimensional correlation between the search results and the actual needs of the users, reduces the interference of a large amount of useless information, and facilitates the improvement of the search accuracy; further, according to the similarity characteristics of the search keywords and the target text keywords, combined with the recurrence time interval of the search keywords and the user search tendency parameters, the actual needs of the users are associated on the basis of traditional search keyword matching, and the personal habits of the users are integrated to obtain the search matching degree of the target text keywords, so as to provide a basis for finally obtaining the search results; finally, based on the search matching degree of the text keywords, the search results are obtained using a preset recommendation algorithm. The present invention clusters articles by analyzing the correlation between them, and uses the user behavior habits and business scope of the same enterprise, as well as the personal search characteristics of the search user, to modify the traditional keyword matching, provide personalized search results, and improve search efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are 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.

[0035] Figure 1 A flowchart of an intellectual property data retrieval method provided by one embodiment of the present invention;

[0036] Figure 2 A flowchart of a method for obtaining correlation parameters provided by an embodiment of the present invention;

[0037] Figure 3The present invention is a flowchart of a method for obtaining user search tendency parameters provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the intellectual property data retrieval method and system proposed by the present invention, its specific implementation, structure, features and effects, in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0039] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0040] The specific scheme of an intellectual property data retrieval method and system provided by the present invention is described in detail below with reference to the accompanying drawings.

[0041] See also Figure 1 , which shows a flow chart of an intellectual property data retrieval method provided by an embodiment of the present invention, specifically comprising:

[0042] Step S1: Obtain the user's identity information and the user's search keywords; the identity information at least includes the business scope of the user's enterprise; obtain the business keyword vector of the business scope; obtain the text keywords and text keyword vectors of the articles in the intellectual property information database.

[0043] Considering that the main business scope of an enterprise can be roughly determined through the name of the enterprise, such as A Automobile Manufacturing Co., Ltd., combined with the relevant information of the enterprise information websites open on the Internet, such as the industrial and commercial website, the tax website, the enterprise inspection website and other websites, a more detailed business scope of the enterprise can be obtained. For example, according to the A Automobile Manufacturing Co., Ltd. displayed in the user's registration information, it is found that the company mainly produces and sells the main part of the automobile frame. Then the intellectual property information search content display for this user should be biased towards the fields of automobile industry, mechanical processing, etc.

[0044] In an embodiment of the present invention, when a user registers, at least the user's company name, name, ID number, and mobile phone number are collected as the user's identity information, and the company's business scope is obtained through the user's company name. The company's business scope information is incorporated into the user's identity information to facilitate subsequent optimization of search results with the help of the user's company's business scope.

[0045] Considering that the business scope information of the user's enterprise and the articles in the intellectual property information database cannot be directly processed by computers, word segmentation preprocessing is first performed to obtain the business keyword vector of the business scope; and the text keywords and text keyword vectors of the articles in the intellectual property information database are obtained.

[0046] Preferably, in one embodiment of the present invention, the Jieba word segmentation algorithm is used as the preset word segmentation algorithm, and the BERT model (Bidirectional Encoder Representations from Transformers) is used to generate word vectors; the business scope information of the enterprise is segmented, and the business keyword vector of the business scope is obtained;

[0047] Considering that the title, keywords, abstract and technical issues of an article usually contain the core information of the entire article and the amount of text is relatively small, obtaining text keywords through these core information is conducive to improving retrieval efficiency. Therefore, at least the title, keywords, abstract and technical issues of the article are decomposed using the preset word segmentation algorithm to obtain the text keywords of the article and obtain the text keyword vector.

[0048] It should be noted that the Jieba word segmentation algorithm and the BERT model are both existing technologies. In other embodiments of the present invention, the implementer may also use a hidden Markov model for word segmentation or use the Word2Vec method for vectorization, and may also perform word segmentation and vectorization on other article parts such as background technology to obtain the text keywords and text keyword vectors of the article. These are technical means well known to those skilled in the art and will not be elaborated here.

[0049] Step S2: Based on the similarity features of all text keyword vectors between any two articles and the similarity features of the number of text keywords, the correlation parameters between the articles are obtained; and the articles in the database are clustered according to the correlation parameters between the articles.

[0050] Considering that there are a large number of patent documents and intellectual property information in the intellectual property information database, classifying related articles is conducive to fast retrieval and improving retrieval efficiency, so the articles in the database are clustered; considering that the more similar the text keyword vectors between articles and the more similar the number of text keywords are, the more similar the articles are, so based on the similarity features of all text keyword vectors between any two articles and the similarity features of the number of text keywords, the correlation parameters between articles are obtained to characterize the correlation between articles and provide a basis for further clustering.

[0051] Preferably, in one embodiment of the present invention, the method for obtaining the correlation parameter between articles includes: Figure 2, which shows a flow chart of a method for obtaining correlation parameters provided by an embodiment of the present invention, specifically comprising:

[0052] Step S201: Pair the text keywords between any two different articles to obtain the similarity coefficient between the two paired text keywords; obtain the sum of the similarity coefficients of all paired text keyword vectors between the two articles as the first correlation sub-parameter between the corresponding two articles.

[0053] Considering that the larger the similarity coefficient between text keyword vectors, the stronger the similarity between text keyword vectors, the stronger the text consistency, and the stronger the correlation between articles, the text keywords between any two different articles are paired to obtain the similarity coefficient between the two paired text keywords; the sum of the similarity coefficients of all paired text keyword vectors between the two articles is obtained as the first correlation sub-parameter between the corresponding two articles.

[0054] As an example, the cosine similarity is normalized as the similarity coefficient between the text keyword vectors, and linear normalization is specifically used; the calculation formula of the first related sub-parameter includes:

[0055]

[0056] In the calculation formula of the first related sub-parameter, DX1 z,v represents the first correlation sub-parameter corresponding to the z-th article and the v-th article; X represents the number of text keyword vectors contained in the z-th article; x represents the sequence number of the text keyword vector contained in the z-th article; w z,x represents the x text keyword vectors contained in the zth article; Y represents the number of text keyword vectors contained in the vth article; y represents the sequence number of the text keyword vector contained in the vth article; w v,y represents the yth text keyword vector contained in the vth article; cos_sim() represents the cosine similarity function; norn[] represents the linear normalization function.

[0057] The larger the cosine similarity, the more consistent the vector directions are, the closer the text keywords are in content and semantics, the stronger the correlation is, and the larger the first correlation sub-parameter is.

[0058] It should be noted that cosine similarity is a technical means well known to those skilled in the art. In other embodiments of the present invention, the implementer may also obtain the Jaccard similarity of the text keywords of the two articles as the first related sub-parameter.

[0059] Step S202: According to the similarity feature of the number of text keywords between the two articles, obtain the second related sub-parameter between the two articles.

[0060] Considering that the more obvious the similarity feature of the number of text keywords between articles is, the higher the overlap between the corresponding two articles is, and the stronger the correlation between the articles is, the second correlation sub-parameter between the corresponding two articles is obtained according to the similarity feature of the number of text keywords between the two articles.

[0061] As an example, the absolute value of the difference in the number of text keywords between the two articles is negatively correlated by taking the inverse. Specifically, the inverse of the sum of the absolute value of the difference in the number of text keywords between the two articles and the preset division parameter 1 is taken as the second correlation sub-parameter between the corresponding two articles.

[0062] In other embodiments of the present invention, the implementer may also perform negative correlation mapping through a negative correlation mapping function such as exp(-x) to obtain a second related sub-parameter, such as mapping the absolute value of the difference in the number of text keywords between two articles as an independent variable to the exp(-x) function, and using the mapping value as the second related sub-parameter between the two articles. Where exp() represents an exponential function with the natural constant e as the base; x represents an independent variable.

[0063] Step S203: According to the quantity feature of the same characters in the titles of the two articles, a third correlation sub-parameter between the two articles is obtained.

[0064] Considering that the titles of articles are highly condensed versions of intellectual property articles, the more identical characters there are, the more likely it is that the two articles have the same or related topics and are more likely to be related in content. Therefore, the third correlation sub-parameter between the two articles is obtained based on the number of identical characters in the titles of the two articles.

[0065] As an example, the number of identical characters in the titles of two articles is used to represent the quantity feature, and the number of identical characters in the titles of the two articles is used as the third correlation sub-parameter between the two articles.

[0066] As another example, the quantitative feature is represented by the ratio of the number of identical characters in the titles of two articles. The ratio of the number of identical characters in the titles of the two articles to the number of characters with the largest number of characters in the titles of the two articles is used as the third related sub-parameter between the two articles.

[0067] Step S204: Obtain correlation parameters between articles according to the first correlation sub-parameter, the second correlation sub-parameter and the third correlation sub-parameter.

[0068] The first correlation sub-parameter, the second correlation sub-parameter and the third correlation sub-parameter are all positively correlated with the correlation parameter.

[0069] After obtaining the first relevant sub-parameter, the second relevant sub-parameter and the third relevant sub-parameter, multiple correlation analysis angles can be integrated to obtain the correlation parameters between articles based on the first relevant sub-parameter, the second relevant sub-parameter and the third relevant sub-parameter.

[0070] Considering that the larger the first correlation sub-parameter is, the more similar the text keywords of the two articles are; the larger the second correlation sub-parameter is, the more similar the number of text keywords of the two articles are; the larger the third correlation sub-parameter is, the more similar the titles of the two articles are; therefore, the first correlation sub-parameter, the second correlation sub-parameter and the third correlation sub-parameter are all positively correlated with the correlation parameter.

[0071] As an example, the product of the first correlation sub-parameter, the second correlation sub-parameter and the third correlation sub-parameter is linearly normalized and used as the correlation parameter of the corresponding two articles.

[0072] In other embodiments of the present invention, the implementer may also obtain the Jaccard similarity of the text keywords of the two articles as the fourth relevant sub-parameter, and use the fourth relevant sub-parameter to replace any one of the first relevant sub-parameter, the second relevant sub-parameter and the third relevant sub-parameter to obtain the correlation parameter.

[0073] After analyzing the correlation between the two articles, a correlation parameter is obtained to characterize the correlation between the two articles, that is, the articles in the database are clustered according to the correlation parameter between the articles.

[0074] Preferably, in one embodiment of the present invention, considering that the stronger the correlation between two articles, the larger the correlation parameter, the more they need to be classified into one category, and the closer the corresponding clustering distance needs to be, the correlation parameter is negatively correlated and mapped as the clustering distance between the corresponding two articles, and the articles in the database are clustered using a preset clustering algorithm.

[0075] As an example, negative correlation mapping is performed by taking the inverse, and the inverse of the sum of the correlation parameter and the preset zero division parameter 0.01 is taken as the clustering distance between the corresponding two articles; the preset clustering algorithm is the Isodata clustering algorithm, and the Isodata clustering algorithm is used to cluster the articles in the database.

[0076] It should be noted that the Isodata clustering algorithm is already an existing technology and will not be described in detail.

[0077] Step S3: Select any text keyword as the target text keyword; obtain the user search tendency parameter of the target text keyword based on the similarity features between the historical search word vectors of other users of the same enterprise and the target text keyword vector, combined with the similarity features between all text keyword vectors and all business keyword vectors in the cluster to which the target text keyword belongs; obtain the search matching degree of the target text keyword based on the similarity features between the search keyword and the target text keyword, combined with the recurrence time interval of the search keyword and the user search tendency parameters.

[0078] Select any text keyword as the target text keyword and analyze them one by one; considering that users of the same enterprise have similar search fields or search directions when searching, by analyzing the user behavior of the same enterprise, it is convenient to find similar search habits; at the same time, the more similar the vectors of all text keywords in the cluster to which the target text keyword belongs are to all business keyword vectors, the stronger the correlation between the target text keyword and the business scope of the user enterprise is, so according to the similarity characteristics of the historical search word vectors of other users of the same enterprise and the target text keyword vector, combined with the similarity characteristics of all text keyword vectors in the cluster to which the target text keyword belongs and all business keyword vectors, the user search tendency parameters of the target text keyword are obtained, the potential needs of the current user are inferred, the relevance of the match is increased, the accuracy of intellectual property data retrieval and personalized service capabilities are improved, and the accuracy of the retrieval results is improved.

[0079] Preferably, in one embodiment of the present invention, the method for obtaining the user search tendency parameter includes:

[0080] See also Figure 3 , which shows a flow chart of a method for obtaining a user search tendency parameter provided by an embodiment of the present invention, specifically comprising:

[0081] Step S301: obtaining a first tendency sub-parameter of the target text keyword based on similar features between the historical search word vectors of other users of the same enterprise and the target text keyword vector.

[0082] Considering that the more similar the historical search word vectors of other users of the same enterprise are to the target text keyword vector, the more likely the target text keyword is the current user's search target when the current user searches, and the stronger the search tendency is, so the first tendency sub-parameter of the target text keyword is obtained based on the similarity features between the historical search word vectors of other users of the same enterprise and the target text keyword vector.

[0083] As an example, considering that the smaller the Euclidean distance between text vectors, the closer the two texts are in content and semantics, and the more obvious the similar features are, the average Euclidean distance between all historical search word vectors of other users of the same enterprise and the target text keyword vector is obtained, and the average Euclidean distance is negatively correlated and mapped as the first tendency sub-parameter of the target text keyword.

[0084] It should be noted that the implementer can perform negative correlation mapping by taking the inverse method, and take the inverse of the average Euclidean distance as the first tendency sub-parameter of the target text keyword; among which, the historical search word vectors of other users of the same enterprise cannot be exactly the same as the target text keyword vector, that is, the average Euclidean distance is not zero; the implementer can also perform negative correlation mapping through a negative correlation function such as the exp(-x) function.

[0085] Step S302: Obtain a second tendency sub-parameter of the target text keyword based on similar features between all text keyword vectors and all business keyword vectors in the cluster to which the target text keyword belongs.

[0086] Considering that the more similar all text keyword vectors in the cluster to which the target text keyword belongs are to all business keyword vectors, the more the cluster to which the target text keyword belongs matches the business scope of the user's enterprise, the stronger the possibility of the target text keyword as a search result, and the larger the user's search tendency parameter, the second tendency sub-parameter of the target text keyword is obtained based on the similarity characteristics of all text keyword vectors in the cluster to which the target text keyword belongs and all business keyword vectors.

[0087] As an example, similar to the process of obtaining the first related sub-parameter, a text keyword vector is selected from all text keyword vectors in the cluster to which the target text keyword belongs, and paired with the business keyword vector to obtain the cosine similarity between the paired text keyword vector and the business keyword vector. The cosine similarities of all paired text keyword vectors and the business keyword vector are normalized and summed up, and the sum is used as the second tendency sub-parameter of the target text keyword.

[0088] Step S303: obtaining a user search tendency parameter of the target text keyword according to the first tendency sub-parameter and the second tendency sub-parameter of the target text keyword; the first tendency sub-parameter and the second tendency sub-parameter are both positively correlated with the user search tendency parameter.

[0089] After analyzing the user search tendency characteristics of the target text keywords from different angles, the user search tendency parameters of the target text keywords can be obtained according to the first tendency sub-parameter and the second tendency sub-parameter of the target text keywords, which characterizes the tendency degree of the target text keywords to be searched by users, and provides a basis for obtaining the search matching degree and obtaining accurate search results in the subsequent process. Considering that the larger the first tendency sub-parameter is, the more similar the historical search word vectors of other users of the same enterprise are to the target text keyword vector, the more likely the target text keyword is the search target of the current user when the current user searches, and the stronger the search tendency is; the larger the second tendency sub-parameter is, the more matching the cluster to which the target text keyword belongs is with the business scope of the user's enterprise, the stronger the possibility of the target text keyword as a search result is, and the larger the user search tendency parameter is, so the first tendency sub-parameter and the second tendency sub-parameter are both positively correlated with the user search tendency parameter.

[0090] As an example, the product of the first tendency sub-parameter and the second tendency sub-parameter is linearly normalized and used as the user search tendency parameter of the target text keyword.

[0091] It should be noted that, in other embodiments of the present invention, it is also considered that the more times the target text keyword appears in the business scope of the user enterprise, the more consistent the target text keyword is with the business scope, and the more likely it is to meet the user's search intention. Therefore, the sum of the number of times the target text keyword appears in the business scope of the user enterprise and the constant 1 is used as the third tendency sub-parameter to supplement the first tendency sub-parameter and the second tendency sub-parameter. The product of the first tendency sub-parameter, the second tendency sub-parameter and the third tendency sub-parameter is linearly normalized and used as the user search tendency parameter for the target text keyword.

[0092] Traditional search methods often only consider the matching degree of the keywords themselves. However, the embodiments of the present invention analyze the similar characteristics of text keywords and business operation keywords in the clusters and use the search behavior habits of users of the same enterprise to increase the multi-dimensional correlation between the search results and the actual needs of users, reduce the interference of a large amount of useless information, and enable users to obtain the required intellectual property data faster and more accurately, thereby improving work efficiency.

[0093] Taking into account that the business scope of an enterprise is usually relatively fixed and the business scope remains unchanged over a period of time, and the work tasks that an individual user is responsible for over a period of time will also remain stable, the historical search records of an individual user on the intellectual property information website platform are consistent, which is reflected in the short time interval between the recurrence of the search keywords; at the same time, the similarity between the search keywords and the target text keywords reflects the possibility of the target text keywords as the user's search target; the user's search tendency parameters reflect the possibility of the target text keywords becoming the search results from the perspective of the search habits of other users of the same enterprise and the fit between the cluster where the target text keywords are located and the business scope of the enterprise. Therefore, based on the similarity between the search keywords and the target text keywords, combined with the time interval between the recurrence of the search keywords and the user's search tendency parameters, the search matching degree of the target text keywords is obtained, and the actual needs of the users are associated on the basis of the traditional search keyword matching, and the personal habits of the users are integrated to provide a basis for finally obtaining the search results.

[0094] Preferably, in one embodiment of the present invention, the more similar the target text keyword is to the search keyword, the higher the search matching degree of the target text keyword is; the shorter the recurrence time interval of the search keyword is, that is, the closer the time when the user searches for the same keyword in the current search and in the past, the more the user wants to retrieve the article corresponding to the search keyword, and the greater the importance of the search keyword is; the user search tendency parameter reflects the possibility of the target text keyword becoming the search result, and the larger the user search tendency parameter is, the greater the possibility of the target text keyword becoming the search result is;

[0095] Based on this, similarity parameters between each search keyword and the target text keyword are obtained as matching sub-parameters between the target text keyword and each search keyword;

[0096] After negative correlation mapping of the recurrence time interval of each search keyword, it is used as the time reference weight of each search keyword;

[0097] The retrieval matching degree of the target text keyword is obtained according to the matching sub-parameters between the target text keyword and each search keyword, the time reference weight of each search keyword and the user search tendency parameter of the target text keyword; the matching sub-parameters, time reference weight and user search tendency parameters are all positively correlated with the retrieval matching degree.

[0098] As an example, the calculation formula for the search matching degree includes:

[0099]

[0100] Among them, P arepresents the retrieval matching degree of the ath target text keyword in the current retrieval process; norm[] represents the linear normalization function; E a represents the user's search tendency parameter for the ath target text keyword in the current search process; l represents the sequence number of the user's current search keyword; L represents the number of the user's current search keywords; exp() represents an exponential function with the natural constant e as the base; g a,l represents the Hamming distance between the user's current lth search keyword and the ath target text keyword, exp(―g a,l ) represents the matching sub-parameter corresponding to the user's current lth search keyword and the ath target text keyword; T l Indicates the recurrence time interval of the user's current lth search keyword.

[0101] In the calculation formula of the search matching degree, the Hamming distance between the search keyword and the target text keyword is obtained, and the Hamming distance is negatively correlated through the negative correlation function exp(-x). The mapping result is used as the similarity parameter between each search keyword and the target text keyword, and also as the matching subparameter between the target text keyword and each search keyword. The larger the matching subparameter, the more the target text keyword matches the user's search, and the greater the search matching degree; the time interval between the user's current search keyword and the last appearance of the same search keyword in the user's historical search record is used as the recurrence time interval of the search keyword; since the time interval is not 0, a negative correlation mapping is performed by taking the reciprocal, and the reciprocal of the recurrence time interval of each search keyword is taken as the time reference weight of each search keyword; the larger the time reference weight, the more important the similarity feature between the corresponding search keyword and the target text keyword is, and the more attention the user pays to the corresponding search keyword; E a The larger it is, the greater the possibility that the target text keyword will become the search result, and the greater the search match degree.

[0102] It should be noted that when the user searches for the first time, the time reference weight is not calculated, and the formula is transformed into: Obtain the search matching degree; when the search keyword appears for the first time, the time interval between the first search record and the current search is used as the recurrence time interval of the search keyword.

[0103] Step S4: Obtain search results using a preset recommendation algorithm based on the search matching degree of the text keywords.

[0104] After evaluating the search matching degree of the text keywords, the search results can be obtained using a preset recommendation algorithm based on the search matching degree of the text keywords.

[0105] Preferably, in one embodiment of the present invention, the preset recommendation algorithm is a reinforcement learning recommendation algorithm (Reinforcement Learning-Based Recommendations), and the average value of the search matching degrees of all text keywords in the same article is used as the weight of the content display of the search results. The larger the average value of the search matching degrees, the greater the weight of the content display, indicating that the higher the content relevance of the intellectual property information data to which the text keywords belong, the higher the display result. At the same time, on the intellectual property information website, for highly relevant intellectual property information data, color gradient can be used to represent the data content, and the more top-ranked the information data, the darker the color, so as to better assist users in viewing the retrieved content.

[0106] It should be noted that the reinforcement learning recommendation algorithm is already an existing technology and will not be described in detail here.

[0107] An embodiment of the present invention further provides an intellectual property data retrieval system, which includes a memory, a processor and a computer program, wherein the memory is used to store the corresponding computer program, and the processor is used to run the corresponding computer program. When the computer program runs in the processor, it can implement an intellectual property data retrieval method described in steps S1-S4.

[0108] In summary, in order to solve the technical problem of low retrieval efficiency of existing intellectual property data retrieval methods, the present invention proposes an intellectual property data retrieval method and system. The present invention first obtains the user's identity information and the user's search keywords, business keyword vectors, text keywords and text keyword vectors; further clusters the articles in the database based on the similarity features of all text keyword vectors between articles and the similarity features of the number of text keywords; further obtains the user's search tendency parameters for the target text keywords, combines the similarity features of the search keywords and the target text keywords, and the recurrence time interval of the search keywords, to obtain the search matching degree of the target text keywords; finally, based on the search matching degree of the text keywords, a preset recommendation algorithm is used to obtain the search results. The present invention clusters the articles by analyzing the correlation between the articles, and uses the user behavior habits and business scope of the same enterprise, as well as the personal search characteristics of the search user, to modify the traditional keyword matching, provide personalized search results, and improve the search efficiency.

[0109] It should be noted that the sequence of the above embodiments of the present invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0110] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for retrieving intellectual property data, characterized in that: The method comprises: Obtaining the user's identity information and the user's search keywords; the identity information at least includes the business scope of the user's enterprise; obtaining the business keyword vector of the business scope; obtaining the text keywords and text keyword vectors of the articles in the intellectual property information database; According to the similarity characteristics of all text keyword vectors between any two articles, combined with the similarity characteristics of the number of text keywords, the correlation parameters between the articles are obtained; and the articles in the database are clustered according to the correlation parameters between the articles; Select any of the text keywords as the target text keyword; obtain the user search tendency parameter of the target text keyword based on the similarity characteristics between the historical search word vectors of other users of the same enterprise and the target text keyword vector, combined with the similarity characteristics between all the text keyword vectors and all the business keyword vectors in the cluster to which the target text keyword belongs; obtain the search matching degree of the target text keyword based on the similarity characteristics between the search keyword and the target text keyword, combined with the recurrence time interval of the search keyword and the user search tendency parameter; The search results are obtained by using a preset recommendation algorithm based on the search matching degree of the text keywords.

2. The method for retrieving intellectual property data according to claim 1, characterized in that: The method for obtaining the correlation parameters between the articles includes: Pair the text keywords between any two different articles to obtain a similarity coefficient between the two paired text keywords; obtain the sum of the similarity coefficients of all paired text keyword vectors between the two articles as the first correlation sub-parameter between the two articles; According to the similarity feature of the number of text keywords between the two articles, obtaining the second related sub-parameter between the two articles; According to the quantity characteristics of the same characters in the titles of the two articles, the third correlation sub-parameter between the two articles is obtained; A correlation parameter between articles is obtained according to the first correlation sub-parameter, the second correlation sub-parameter and the third correlation sub-parameter, wherein the first correlation sub-parameter, the second correlation sub-parameter and the third correlation sub-parameter are all positively correlated with the correlation parameter.

3. The method for retrieving intellectual property data according to claim 2, characterized in that: The method for clustering articles in a database according to the correlation parameters between the articles comprises: The correlation parameter is negatively mapped and used as the clustering distance between the corresponding two articles, and the articles in the database are clustered using a preset clustering algorithm.

4. The method for retrieving intellectual property data according to claim 1, characterized in that: The method for obtaining the user search tendency parameter includes: Obtaining a first tendency sub-parameter of the target text keyword based on similar features between the historical search word vectors of other users of the same enterprise and the target text keyword vector; Acquire a second tendency sub-parameter of the target text keyword according to similar features of all the text keyword vectors and all the business keyword vectors in the cluster to which the target text keyword belongs; According to the first tendency sub-parameter and the second tendency sub-parameter of the target text keyword, a user search tendency parameter of the target text keyword is obtained; the first tendency sub-parameter and the second tendency sub-parameter are both positively correlated with the user search tendency parameter.

5. The method for retrieving intellectual property data according to claim 4, characterized in that: The method for obtaining the first tendency sub-parameter includes: The average value of the Euclidean distances between all historical search word vectors of other users of the same enterprise and the target text keyword vector is obtained, and the average value of the Euclidean distances is negatively correlated and mapped as the first tendency sub-parameter of the target text keyword.

6. The method for retrieving intellectual property data according to claim 1, characterized in that: The method for obtaining the retrieval matching degree includes: Acquire similar parameters between each of the search keywords and the target text keywords as matching sub-parameters between the target text keywords and each of the search keywords; After negative correlation mapping, the recurrence time interval of each of the search keywords is used as the time reference weight of each of the search keywords; The retrieval matching degree of the target text keyword is obtained based on the matching sub-parameters of the target text keyword and each of the retrieval keywords, the time reference weight of each of the retrieval keywords and the user retrieval tendency parameters of the target text keyword; the matching sub-parameters, the time reference weight and the user retrieval tendency parameters are all positively correlated with the retrieval matching degree.

7. The method for retrieving intellectual property data according to claim 1, characterized in that: The preset recommendation algorithm is a reinforcement learning recommendation algorithm.

8. The method for retrieving intellectual property data according to claim 1, characterized in that: The method for obtaining text keywords of the article includes: At least the title, keywords, abstract and technical issues of the article are decomposed using a preset word segmentation algorithm to obtain the text keywords of the article.

9. The method for retrieving intellectual property data according to claim 3, characterized in that: The preset clustering algorithm is the Isodata clustering algorithm.

10. An intellectual property data retrieval system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the intellectual property data retrieval method as claimed in any one of claims 1 to 9 are implemented.

Images