Knowledge document recommendation method

By building a state transfer matrix and graph network, analyzing user access behaviors and evaluating the correlation between knowledge documents, the problem of difficulty in recommending new users is solved, efficient and personalized knowledge document recommendations are achieved, and user experience is improved.

CN118916479APending Publication Date: 2024-11-08CHINA TELECOM DIGITAL INTELLIGENCE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410992084.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The recommendation algorithm of the traditional knowledge management platform has a "cold start" problem for new users, resulting in poor recommendation results and large errors when judging similarity based on user background information.

Method used

By constructing a state transfer matrix, analyzing user access behavior, evaluating the correlation between knowledge documents, and recommending relevant content based on the converged matrix, overcoming the problem of insufficient history of new users, and at the same time using graph network and matrix iterative algorithms to improve the recommended ability Explanatory.

Benefits of technology

It effectively overcomes the problem of new users' recommendation difficulties, improves the relevance and personalization of recommendations, reduces user search time, and improves user experience and content utilization efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118916479A_ABST
    Figure CN118916479A_ABST
Patent Text Reader

Abstract

The invention relates to a knowledge document recommendation method, which belongs to the technical field of content recommendation, and comprises the following steps: in each statistical period, counting an access sequence of a user to knowledge documents, and constructing a state transition matrix according to access behaviors of the user to the knowledge documents; calculating a converged state transition matrix; according to the element values in the converged state transition matrix, evaluating the correlation among the knowledge document contents; and obtaining the knowledge document currently read by the user, and recommending the knowledge document with high correlation with the currently read knowledge document to the user according to all the obtained document sets with correlation. According to the method disclosed by the invention, the correlation among the contents is mined by utilizing the historical records of content access of the existing users of the platform, so that the cold start problem of a traditional recommendation algorithm for new users is solved, and meanwhile, the method also has relatively high interpretability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of content recommendation, and in particular relates to a knowledge document recommendation method. Background Art

[0002] Knowledge is an important asset of an enterprise, especially for those enterprises that rely on professional knowledge, technology or experience to operate. The knowledge management platform can centrally store, classify and retrieve this knowledge, and provide a centralized, online knowledge base where employees can access and obtain the required knowledge anytime and anywhere, promote knowledge sharing and transfer, and make more effective use of the knowledge resources within the enterprise.

[0003] However, as more and more knowledge is accumulated on the knowledge management platform, the time and energy spent by users to retrieve the knowledge of their interest will gradually increase. Content recommendation algorithms can mine valuable knowledge hidden in a large amount of information based on user behavior, interests, historical data, and the relevance of content, and recommend the most relevant and valuable knowledge content to users. This can greatly improve the user experience for knowledge management platforms and enable users to obtain the required knowledge more efficiently.

[0004] The knowledge management platform can centrally store, classify and retrieve this knowledge, and provide a centralized, online knowledge base where employees can access and obtain the knowledge they need anytime and anywhere, promoting knowledge sharing and transfer, and making more effective use of the knowledge resources within the enterprise. However, as more and more knowledge is accumulated on the knowledge management platform, the time and energy spent by users to retrieve the knowledge of their interest will gradually increase. The content recommendation algorithm mines the valuable knowledge hidden in a large amount of information and recommends the most relevant and valuable knowledge content to users. This can greatly improve the user experience for the knowledge management platform.

[0005] Traditional collaborative filtering algorithms and community discovery all attempt to mine similarities between users to recommend content. However, for new users of knowledge management platforms, due to insufficient historical behavior data, they will inevitably encounter a "cold start" problem, resulting in poor performance of recommendation algorithms.

[0006] The invention patent with application announcement number CN 114969562 A discloses a method for discovering user communities in social networks based on Markov similarity enhancement. The method first extracts the initial matrix and state transition matrix through the topological structure of the network, then obtains the steady-state Markov similarity enhancement matrix through Markov iterative state transition of the initial matrix, obtains the most similar node pair set of the network according to the Markov similarity index, and proposes the connected components therein to complete community discovery. However, this method has the following problems: insufficient historical records for new users lead to the "cold start" problem, resulting in poor recommendation results.

[0007] The invention patent with application announcement number CN 101826114 discloses a content recommendation method based on multiple Markov chains. This method uses the user's click stream information to establish a Markov model, and uses the user's background information to establish a user relationship matrix. Then similar Markov models are merged, and the zero rows of the merged Markov model are filled with sparse items based on the click stream of the similar user set obtained from the user relationship matrix to solve the scarcity of data. However, this method has the following problems: using the user's background information, such as user age, gender, education, work, and region, to judge the similarity between users, resulting in a relatively large recommendation error. Summary of the invention

[0008] In view of the above shortcomings of the prior art, the purpose of the invention is to provide a knowledge document recommendation method, which utilizes the historical records of content access by existing users of the platform to mine the correlation between content, which not only overcomes the "cold start problem" of traditional recommendation algorithms for new users, but also has high interpretability.

[0009] The present invention proposes a knowledge document recommendation method, comprising:

[0010] S1, in each statistical period, count the user's access sequence to knowledge documents, analyze the user's access behavior to knowledge documents, and construct the state transition matrix P according to the user's access behavior between knowledge documents;

[0011] S2, calculate the state transfer matrix P^' after convergence;

[0012] S3, evaluating the relevance between the contents of the knowledge documents according to the element values ​​in the converged state transition matrix P^', including:

[0013] Set the threshold w, for knowledge document i, traverse all elements in the i-th row of the matrix P^', if there is an element P in the matrix that is greater than the threshold w ij , then it is judged that the contents of knowledge document i and knowledge document j are relevant;

[0014] Since the correlation between documents has a transitive relationship, for other knowledge documents k that are judged to be relevant to knowledge document i, it is considered that this knowledge document k is also relevant to knowledge document j, and so on, traversing all knowledge documents to find all relevant document sets;

[0015] S4, obtaining the knowledge document that the user is currently reading, and recommending to the user knowledge documents that are highly relevant to the knowledge document that the user is currently reading based on the set of all relevant documents obtained in step S3.

[0016] Furthermore, in S1, a state transfer matrix is ​​constructed, including:

[0017] Taking knowledge documents as nodes, if there is a user who visits knowledge document i and then visits knowledge document j, then there is an edge between nodes i and j, and the weight of the edge is the number of users who have such access behavior; if there is no such user, there is no edge between nodes i and j;

[0018] A graph network is constructed based on the sequence of users' access to knowledge documents in the most recent statistical period, and the element values ​​in the state transfer matrix are filled in according to the node connectivity and edge weights of the graph network to complete the construction of the state transfer matrix.

[0019] Further, in S1, the state transfer matrix is:

[0020]

[0021] Among them, P is the state transfer matrix, n is the sum of platform content, P ij It represents the probability that the next document accessed by a user after accessing document i is document j.

[0022] Further,

[0023] Among them, w ij is the weight of the edge between node i and node j, w i is the weight of all edges of node i, w j are the weights of all edges of node j.

[0024] Furthermore, the element values ​​in the matrix are normalized to ensure that the sum of all elements in each row is 1.

[0025]

[0026] Furthermore, the calculation of the converged state transfer matrix P^' includes:

[0027] Multiply the state transfer matrix P by itself to obtain a new state transfer matrix P^'=P*P;

[0028] This process needs to be iterated for multiple rounds. When the state tends to be stably distributed, that is, after multiple state transitions, the change in the value of the element at the same position in the matrix is ​​less than the threshold 1 or the number of state transitions is greater than the threshold 2, it can be stopped. The final result is the converged state transfer matrix P^'.

[0029] Furthermore, in S1, the duration of the statistical cycle is selected according to the actual operation of the knowledge management platform.

[0030] Furthermore, in S1, the statistical period is 1 week, 1 month or half a year.

[0031] Furthermore, in S3, the threshold w is adjustable, wherein when the relevance in the knowledge documents of the knowledge management platform is strong, the value of the threshold w is increased, otherwise the value of w is decreased.

[0032] Furthermore, in S4, other documents that are highly relevant to the knowledge document that the user is currently reading are found as candidate recommended documents; then, based on the user's historical access records, the documents that the user has previously accessed are eliminated from the candidate recommended documents, and the remaining documents are used as recommended items in the final content recommendation list.

[0033] The beneficial effects of the present invention are as follows:

[0034] The method provided by the present invention recommends content that may be of interest to users based on the user's browsing and downloading behavior and the correlation between content. Personalized recommendation is based on the deep mining of various types of data and is widely used in knowledge management, online shopping and other scenarios. It can find content that may be of interest to users from a large amount of content, reduce the time users spend searching for content, and increase the user experience. The present invention uses the historical records of content access by existing users of the platform to mine the correlation between content, which not only overcomes the "cold start problem" of traditional recommendation algorithms for new users, but also has high interpretability. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings are only used to illustrate specific embodiments and are not considered to limit the present invention. In the entire drawings, the same reference symbols represent the same components. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0036] Figure 1 A flowchart of a knowledge document recommendation method according to an embodiment of the present invention;

[0037] Figure 2 Schematic diagram of the transfer relationship of the relevance between documents according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work should fall within the scope of protection of the present invention.

[0039] Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts disclosed in the present invention.

[0040] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the orientation or position relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. The terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0041] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of methods and systems consistent with some aspects of the present invention as detailed in the appended claims.

[0042] The present invention proposes a knowledge document recommendation method, which constructs a state transition matrix based on the user's access behavior to the content, evaluates the correlation between the contents based on the element values ​​in the converged state transition matrix, and recommends highly correlated contents based on the content that the user is currently reading. The present invention uses the historical records of content access by existing users on the platform to mine the correlation between the contents, which not only overcomes the "cold start problem" of traditional recommendation algorithms for new users, but also has high interpretability.

[0043] like Figure 1 As shown, the knowledge document recommendation method of the embodiment of the present invention includes the following steps:

[0044] S1, in each statistical period, count the user's access sequence to knowledge documents, analyze the user's access behavior to knowledge documents, and construct the state transfer matrix P based on the user's access behavior between knowledge documents.

[0045] In the embodiment of the present invention, the statistical period is a recent period of time, and the length of this statistical period is selected and weighed according to the actual operation of the knowledge management platform. For example, the length of the statistical period can be the last week, the last month, or the last six months. The shorter the time, the easier it is to reflect the changes in short-term interests among users, but the computing overhead will also be greater, and the weighing is made according to the actual operation of the knowledge management platform.

[0046] The specific process of constructing the state transfer matrix is ​​described below:

[0047] Statistics are collected on the sequence of users' access to knowledge documents in the recent period. A graph network is constructed with documents as nodes. If a user visits knowledge document i and then visits document j, there is an edge between nodes i and j, and the weight of the edge is the number of such users. If there is no such user, there is no edge between nodes i and j.

[0048] Let the state transfer matrix be defined as P,

[0049]

[0050] Among them, n is the sum of platform content, P ij It represents the probability that the next document accessed by a user after accessing document i is document j.

[0051]

[0052] Among them, w ij is the weight of the edge between node i and node j, w i is the weight of all edges of node i, w j are the weights of all edges of node j.

[0053] According to the sequence of users' access to knowledge documents in the most recent statistical period, a graph network is constructed, and the element values ​​in the state transfer matrix are filled in according to the node connectivity and edge weights of the graph network to complete the state transfer matrix.

[0054] In order to ensure that the state transfer matrix is ​​convergent, it is necessary to normalize the element values ​​in the matrix, that is, This ensures that the sum of all elements in each row is 1.

[0055] S2, calculate the state transfer matrix P^' after convergence.

[0056] Specifically, the state transfer matrix P is multiplied by itself to obtain a new state transfer matrix P^'=P*P.

[0057] This process needs to be iterated for multiple rounds. When the state tends to be stably distributed, that is, after multiple state transitions, the change amplitude of the value of the element at the same position in the matrix is ​​less than threshold 1 or the number of state transitions is greater than threshold 2, it can be stopped. The final result is the converged state transfer matrix P^'. Among them, the default value of threshold 1 is set to 0.001, and the default value of threshold 2 is set to 20. In actual applications, it can be adjusted according to the actual number of documents on the platform.

[0058] S3, evaluating the relevance between the contents of the knowledge documents according to the element values ​​in the converged state transition matrix P^', including:

[0059] Set the threshold w, for knowledge document i, traverse all elements in the i-th row of the matrix P^', if there is an element P in the matrix that is greater than the threshold w ij , then it is judged that the correlation between knowledge document i and knowledge document j is high.

[0060] In an embodiment of the present invention, the value of the threshold w can be adjusted in consideration of the length of the content recommendation list. Specifically, when the relevance in the documents of the knowledge management platform is generally strong, the value of w can be increased to ensure that only the documents with the highest relevance are used as candidate recommended content; otherwise, the value of w can be lowered to ensure that the content in the content recommendation list is sufficient.

[0061] In addition, the correlation between documents has a transitive relationship. For example, if document i is correlated with document j, and document i is also correlated with document k, then even if P jk The value of is less than w, and it can also be considered that document j and document k are related. Figure 2 shown.

[0062] That is, for other knowledge documents k that are judged to be relevant to knowledge document i, regardless of the P between document i and document k jk If the value of is greater than or less than w, it is directly determined that the knowledge document k is also relevant to the knowledge document j.

[0063] According to the above process, all documents are traversed and all relevant document sets are found.

[0064] S4, obtaining the knowledge document that the user is currently reading, and recommending to the user knowledge documents that are highly relevant to the knowledge document that the user is currently reading based on the set of all relevant documents obtained in step S3.

[0065] Find other documents that are highly relevant to the knowledge document that the user is currently reading as candidate recommended documents; then, based on the user's historical access records, remove the documents that the user has previously visited from the candidate recommended documents, and use the remaining documents as recommended items in the final content recommendation list.

[0066] The knowledge document recommendation method provided by the present invention can be applied to scenarios such as knowledge management platforms and e-commerce platforms, can promote knowledge sharing within the group, accelerate the efficiency and accuracy of content search, and provide better related document recommendation functions. In terms of market benefits, on the one hand, it can accelerate the growth of grassroots employees, especially grassroots account managers, and reach more business opportunities. On the other hand, it can support commercial delivery to customers with similar knowledge management platform needs.

[0067] The knowledge document recommendation method provided by the present invention recommends content that may be of interest to users based on the user's browsing and downloading behavior and the correlation between the content. Personalized recommendation is based on the deep mining of various types of data and is widely used in scenarios such as knowledge management and online shopping. It can find content that may be of interest to users from a large amount of content, reduce the time users spend searching for content, and increase the user experience. The present invention uses the historical records of content access by existing users of the platform to mine the correlation between content, which not only overcomes the "cold start problem" of traditional recommendation algorithms for new users, but also has high interpretability.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed in the present invention should be covered within the protection scope of the present invention.

Claims

1. A knowledge document recommendation method, It is characterized in that include: S1, in each statistical period, count the user's access sequence to knowledge documents, analyze the user's access behavior to knowledge documents, and construct the state transition matrix P according to the user's access behavior between knowledge documents; S2, calculate the state transfer matrix P^' after convergence; S3, evaluating the relevance between the contents of the knowledge documents according to the element values ​​in the converged state transition matrix P^', including: Set the threshold w, for knowledge document i, traverse all elements in the i-th row of the matrix P^', if there is an element P in the matrix that is greater than the threshold w ij , then it is judged that the contents of knowledge document i and knowledge document j are relevant; Since the correlation between documents has a transitive relationship, for other knowledge documents k that are judged to be relevant to knowledge document i, it is considered that this knowledge document k is also relevant to knowledge document j, and so on, traversing all knowledge documents to find all relevant document sets; S4, obtaining the knowledge document that the user is currently reading, and recommending to the user knowledge documents that are highly relevant to the knowledge document that the user is currently reading based on the set of all relevant documents obtained in step S3.

2. A knowledge document recommendation method according to claim 1, It is characterized in that In S1, construct the state transfer matrix, including: Taking knowledge documents as nodes, if there is a user who visits knowledge document i and then visits knowledge document j, then there is an edge between nodes i and j, and the weight of the edge is the number of users who have such access behavior; if there is no such user, there is no edge between nodes i and j; A graph network is constructed based on the sequence of users' access to knowledge documents in the most recent statistical period, and the element values ​​in the state transfer matrix are filled in according to the node connectivity and edge weights of the graph network to complete the construction of the state transfer matrix.

3. A knowledge document recommendation method according to claim 2, It is characterized in that In S1, the state transfer matrix is: Among them, P is the state transfer matrix, n is the sum of platform content, P ij It represents the probability that the next document accessed by a user after accessing document i is document j.

4. A knowledge document recommendation method according to claim 3, It is characterized in that Among them, w ij is the weight of the edge between node i and node j, w i is the weight of all edges of node i, w j are the weights of all edges of node j.

5. A knowledge document recommendation method according to claim 4, It is characterized in that Normalize the element values ​​in the matrix to ensure that the sum of all elements in each row is 1.

6. A knowledge document recommendation method according to claim 1, It is characterized in that The state transfer matrix P^' after the calculation convergence includes: Multiply the state transfer matrix P by itself to obtain a new state transfer matrix P^'=P*P; This process needs to be iterated for multiple rounds. When the state tends to be stably distributed, that is, after multiple state transitions, the change in the value of the element at the same position in the matrix is ​​less than the threshold 1 or the number of state transitions is greater than the threshold 2, it can be stopped. The final result is the converged state transfer matrix P^'.

7. A knowledge document recommendation method according to claim 1, It is characterized in that In S1, the duration of the statistical cycle is selected according to the actual operation of the knowledge management platform.

8. A knowledge document recommendation method according to claim 7, It is characterized in that In S1, the statistical period is 1 week, 1 month or half a year.

9. A knowledge document recommendation method according to claim 1, It is characterized in that In S3, the threshold w is adjustable, wherein when the relevance in the knowledge documents of the knowledge management platform is strong, the value of the threshold w is increased, otherwise the value of w is decreased.

10. A knowledge document recommendation method according to claim 1, It is characterized in that In S4, other documents that are highly relevant to the knowledge document that the user is currently reading are found as candidate recommended documents; Then, according to the user's historical access records, the documents that the user has previously accessed are eliminated from the candidate recommended documents, and the remaining documents are used as recommended items in the final content recommendation list.