A think tank data recommendation method and system based on LFM model

Through the think tank data recommendation method based on the LFM model, using user historical data and KMP algorithm, the problems of low efficiency and single recommendation mode in the existing technology are solved, and efficient and accurate user interest recognition and recommendation are achieved.

CN118377968BActive Publication Date: 2024-11-08JIAXING HAOHAN SOFTWARE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410597610.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-14
Publication Date
2024-11-08
Estimated Expiration
2044-05-14

AI Technical Summary

Technical Problem

The existing think tank data recommendation method relies on text similarity algorithms, which has high calculation cost and low efficiency, and cannot effectively process user historical data, resulting in a single recommendation mode and cannot adapt to changes in user interest points.

Method used

Using an LFM model-based method, the model is trained through user historical data, the user interest matrix is ​​calculated and matrix decomposed, and the knowledge graph is constructed in combination with the KMP algorithm to achieve efficient think tank data recommendation.

Benefits of technology

It improves the accuracy and efficiency of think tank data recommendation, can better adapt to user interests, and reduces computing costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118377968B_ABST
    Figure CN118377968B_ABST
Patent Text Reader

Abstract

The present invention provides a think tank data recommendation method and system based on the LFM model, the method comprising: pre-constructing a knowledge graph of think tank data, and obtaining triple information of corresponding think tank data nodes according to the knowledge graph, obtaining user input data, and querying the triple information using a KMP algorithm according to the input data to obtain first query data; saving the first query data as historical data of the think tank data, obtaining the user's operation data on the think tank data to construct a user's interest matrix for the think tank data; performing matrix decomposition on the user's interest matrix for the think tank data to obtain a user's latent class interest relationship matrix for the think tank data, a think tank data and a latent class relationship matrix, which are used to construct an LFM model; training the LFM model with latent class data obtained according to the historical data of the think tank data to obtain a trained think tank data recommendation model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of database recommendation technology, and in particular to a think tank data recommendation method and system based on LFM model Background Art

[0002] At present, the existing think tank data usually relies on manual input of relevant search terms to query through search engines and recommend the required think tank data. However, the above method has the following technical problems: 1. Existing think tank data usually uses text similarity algorithms for matching, but think tank data text is generally standardized text, and text similarity calculation requires the help of natural language processing models, which has high calculation costs and low efficiency. 2. Existing think tank data has not yet processed the historical data of query users, so the recommendation method is only based on the text similarity calculation of the query text, resulting in a relatively simple recommendation model that cannot adapt to the think tank information query of user interest points. 3. Users have little interest data actually downloaded from think tanks, and the existing model cannot accurately infer the user's interest data when not downloading. Summary of the invention

[0003] One of the inventive objectives of the present invention is to provide a method and system for recommending think tank data based on the LFM model. The method and system train the LFM model through the historical data of users' operations on think tank data. After the LFM model is initialized, the user interest matrix is ​​calculated using the historical data. Matrix decomposition is performed based on the user interest matrix to obtain the user's latent interest matrix for think tank data and the think tank data and latent interest relationship matrix. The LFM model is trained based on the two decomposed matrices to obtain a user interest prediction model. The present invention can effectively solve the problem of sparse user interest matrix for think tank data based on the latent interest matrix, thereby enhancing the accuracy of think tank data recommendation.

[0004] Another invention object of the present invention is to provide a think tank data recommendation method and system based on the LFM model. The method and system use the KMP algorithm to search the think tank data, pre-construct triple information of knowledge graph specification in the think tank data, query the triple information according to the KMP algorithm, and select the think tank data associated with the entity of the optimal triple as the query output according to the query result. Compared with the traditional similarity algorithm, the KMP algorithm of the present invention has the query advantages of low cost and high efficiency.

[0005] Another object of the present invention is to provide a method and system for recommending think tank data based on the LFM model. The method and system use the KMP algorithm to obtain think tank data with a correlation relationship. The present invention searches for think tank data with a certain correlation based on the KMP algorithm as historical data of the LFM model, and uses information related to the think tank data obtained according to the KMP algorithm as latent data, thereby realizing the rapid construction of historical latent data of think tank data based on efficient query.

[0006] In order to achieve at least one of the above-mentioned invention objects, the present invention further provides a think tank data recommendation method based on the LFM model, the method comprising:

[0007] Pre-constructing a knowledge graph of think tank data, and obtaining triple information of corresponding think tank data nodes according to the knowledge graph, obtaining user input data, and querying the triple information using a KMP algorithm according to the input data to obtain first query data;

[0008] The first query data is saved as historical data of think tank data, and the operation data of the user on the think tank data is obtained to construct the user's interest matrix on the think tank data;

[0009] Performing matrix decomposition on the user's interest matrix for think tank data to obtain a user's latent class interest relationship matrix for think tank data, a think tank data and latent class relationship matrix, which are used to construct an LFM model;

[0010] The LFM model is trained based on latent class data obtained from the historical data of the think tank data to obtain a trained think tank data recommendation model.

[0011] According to one of the preferred embodiments of the present invention, the search method of the KMP algorithm includes: converting the triple information in the knowledge graph into a text string, and converting the user's input data into a pattern string, performing character matching between the pattern string and the text string until all characters in the current pattern string are matched in the text string, then obtaining the matched triple information, and extracting the entity data in the matched triple information, saving it and using it to construct the historical data of the think tank data.

[0012] According to another preferred embodiment of the present invention, the search method of the KMP algorithm includes: obtaining user input data and converting it into a pattern string, the pattern string includes k characters arranged in sequence, the text string includes m characters, where m>k, starting to match the pattern string of the k characters with the starting position of the text string of the triple information, calculating the characters of the pattern string and whether there are k1 characters and a first matching string with the same order in the text string, where k1<k, then moving the pattern string by x characters to match the text string, so that there are at least the above k1 characters and the first matching string with the same order in the pattern string to match the pattern string, and continuing to move the pattern string until the characters of the k pattern strings are all matched in the text string.

[0013] According to another preferred embodiment of the present invention, when the pattern string of k characters finds n identical pattern strings in the M characters, each triple position where the n pattern strings are found is located, and the number of the pattern strings existing in each triple information is counted, and they are sorted in order from most to least according to the number of pattern strings, and the think tank data corresponding to the triple with the largest number of pattern string matches is output preferentially.

[0014] According to another preferred embodiment of the present invention, when the k pattern strings are not matched in the text string data, at least one character is deleted from the k character pattern strings to obtain k n The k-character pattern string is matched again with the text string, where the k n The method for obtaining a pattern string of k characters includes: extracting keywords from the input sentence using word segmentation and word embedding technology, and converting the extracted at least one keyword into the k n The pattern string of characters is matched with the corresponding text string of M characters to obtain k information of each triple n The number of matches for a pattern string of characters.

[0015] According to another preferred embodiment of the present invention, the method for constructing the LFM model includes: obtaining historical data of the user on the think tank data, wherein the historical data includes download data, click data, browsing data and search data of the user on the think tank data, wherein the user downloads the corresponding type of think tank data to construct the user's interest matrix R ui , and the interest matrix R ui According to the latent type data, matrix decomposition is performed to obtain the latent interest matrix And the latent interest relationship matrix , at this time, the user's interest matrix can be expressed by the following formula: R ui = , where u is the user, T is the transposed matrix identifier, and i is the think tank data type.

[0016] According to another preferred embodiment of the present invention, the method for constructing the LFM model includes: after obtaining the user's click data, browsing data and search data for the corresponding type of think tank, determining the latent data according to the above click data, browsing data and search data, wherein the latent data includes the number of pattern string matches of the above user input data in the corresponding think tank data triple information, the browsing data includes the browsing time of the user in the corresponding triple think tank data, the click data includes the number of clicks on the corresponding think tank data, which is used to determine the latent data, and after determining the latent data, the interest matrix is ​​calculated according to the following formula: preference (u, i) = R u,i = = , where K is the latent class dimension, The u and k in it represent the relationship between user interests and latent data. The k and i in it represent the relationship between latent class data and the corresponding think tank data type.

[0017] According to another preferred embodiment of the present invention, after the user's interest matrix is ​​decomposed to obtain the user's latent class interest relationship matrix for think tank data, think tank data and latent class relationship matrix, the LFM model is further trained according to the acquired historical data, and the user's interest matrix R is calculated. u,i = = The sum of squared errors: E u,i = ,in Represents the true value of the user interest matrix, and minimizes the sum of squared errors minE u,i The LFM model is used as the trained LFM model.

[0018] In order to achieve at least one of the above-mentioned invention purposes, the present invention further provides a think tank data recommendation system based on the LFM model, and the system executes the above-mentioned think tank data recommendation method based on the LFM model.

[0019] The present invention provides a computer-readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the above-mentioned think tank data recommendation method based on the LFM model. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Shown is a flow chart of a think tank data recommendation method based on the LFM model of the present invention.

[0021] Figure 2What is shown is a schematic diagram of KMP algorithm matching in the present invention.

[0022] Figure 3 Shown is a schematic diagram of matrix decomposition in the LFM model of the present invention. DETAILED DESCRIPTION

[0023] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0024] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the element may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0025] Please combine Figure 1-Figure 3, the present invention discloses a method and system for recommending think tank data based on the LFM model, the method mainly includes the following steps: first, it is necessary to construct a knowledge graph for the existing think tank data, wherein the method for constructing the knowledge graph mainly includes: acquiring knowledge from the think tank data, constructing structured data, extracting entity information in the think tank data using methods including but not limited to word embedding, and further constructing the relationship between different entities, thereby obtaining the knowledge graph of the think tank data, the construction method of the knowledge graph is the prior art, the present invention does not improve the knowledge graph construction method itself, so the present invention does not repeat the details of the construction of the knowledge graph. After completing the construction of the knowledge graph of the think tank data, the present invention further obtains the triple information of the knowledge graph, wherein the triple information of the knowledge graph includes [entity, attribute, relationship], the present invention uses the KMP algorithm to convert the user input information into a pattern string, and converts the triple information in the think tank data into a text string, further uses the KMP algorithm to find matching triple information, and saves the search results as the user's historical data for subsequent sample data for the construction of the LFM model of the think tank data. The present invention constructs latent data based on the LFM model, and the latent data can be used to describe the user's interest in different think tank data, thereby improving the user's interest identification and recommendation of think tank data types, and improving the accuracy of interest data recommendation. The latent data in the present invention includes but is not limited to the following operations of users on think tank data: click data, browsing data and search data, wherein the latent data includes the number of pattern string matches of the above-mentioned user input data in the corresponding think tank data triple information, the browsing data includes the user's browsing time in the corresponding triple think tank data, the click data includes the number of clicks on the corresponding think tank data, and the latent data is used for the decomposition of the interest matrix in the LFM model.

[0026] Specifically, the present invention adopts the KMP algorithm to convert the triples [entity, attribute, relationship] in the think tank data into text strings, wherein the strings are converted into corresponding strings in the order of the Chinese expressions of the triples, wherein the conversion method of the strings can be converted according to the encoding method, such as encoding Chinese as "hello", and converting the Chinese "hello" into a hexadecimal encoded text string as "4f60597d". And the user input data is simultaneously converted into a pattern string with the same hexadecimal encoding. It should be noted that in the present invention, the pattern string is used as the search data and the text string of the triple information is matched. When the pattern string matches the text string of the triple information, the triple information can be located, so as to extract the think tank data corresponding to the triple information. It should be noted that the think tank data types described in the present invention include but are not limited to engineering specification think tank data, scientific research paper think tank data, medical specification think tank data, agricultural planting specification think tank data and various standard-related think tank data. The above think tank data is characterized by relatively more standardized text, and the natural language of natural language such as voice data for natural communication has less reliance on text similarity calculation performance. Therefore, the present invention uses the KMP algorithm for relevant matching with scene adaptability, which can effectively improve query efficiency and reduce query calculation costs.

[0027] The implementation method of the KMP algorithm includes: after obtaining the user input data, converting the input data into a pattern string of k characters in length, and converting all triple information [entity, attribute, relationship] in the think tank data into a text string of M characters in length in advance according to the KMP algorithm, where M>k, wherein the pattern string of k characters in length is initialized for matching from the beginning of the text string, assuming that in the initial matching stage there are k1 characters in the sorting and the characters themselves match the pattern string, the k1 strings of the initial matching are defined as the first matching string, where k1<k, at this time, the pattern string is moved x characters to match the text string, so that there is at least the first matching string with the same order of the above k1 characters in the pattern string and the pattern string matches, for example, please refer to Figure 2, if the user inputs a sentence "Airport Road Engineering Specifications", and the Chinese meaning of the text string is "....Urban Construction Engineering Specifications Manual....", the corresponding local text string is "....57ce5e025efa7b515de57a0b89c483038bf4660e4e66....", and the "Airport Road Engineering Specifications" is converted into a hexadecimal pattern string to obtain "673a573a90538def5de57a0b89c48303". At this time, according to the first string matching k1 characters, 5de57a0b89c48303 is obtained. The first string matching k1 characters is the text string of This means "engineering specification", that is, the characters and sequence of the last 16 strings in the pattern string have the corresponding characters and sequence of the first string in the text strings in the M triples. The pattern string is moved so that the next match directly finds the string position of at least the first string 5de57a0b89c48303 in the text string. For example, the second match matches the following second string: 90538def5de57a0b89c48303. At this time, the second string contains all the characters and sequence in the first string. The pattern string is continued to be moved until the characters of the k pattern strings are matched in the text string. That is, the text strings in the M triples are matched. When the pattern string of the k characters finds n identical pattern strings in the M characters, locate each triple position where the n pattern strings are found, count the number of the pattern strings existing in each triple information, and sort them in descending order according to the number of pattern strings, and give priority to outputting the think tank data corresponding to the triples with the largest number of pattern string matches. The number of times the k pattern strings appear in the corresponding triples can be statistically saved as historical data of user searches for subsequent decomposition of the user interest matrix.

[0028] Furthermore, since the user input data may be long or contain irrelevant text, the phenomenon that the pattern string is not matched in the text string of the M triple information may occur during the actual matching process. In one of the preferred embodiments of the present invention, the present invention further performs word segmentation on the pattern string to obtain keywords of the user input data, and further uses the KMP algorithm to perform relevant data processing on the keywords after word segmentation. In another optional preferred embodiment of the present invention, the keywords obtained after word segmentation can be embedded to obtain similar keywords, and the similar keywords can be used to expand the matching range of the input sentence. Specifically: when the k pattern strings are not matched in the text string data, at least one character is deleted from the k character pattern strings to obtain k nThe k-character pattern string is matched again with the text string, where the k n The method for obtaining a pattern string of k characters includes: extracting keywords from the input sentence using a word embedding technology, and converting the extracted at least one keyword into the k n The pattern string of characters is matched with the corresponding text string of M characters to obtain k information of each triple n The number of matches of the pattern string of characters is the word embedding in the present invention. The above-mentioned word segmentation processing of the user input data can effectively avoid the interference of matching irrelevant characters, and the word segmentation text is only used for the user input data, which will not significantly increase the calculation cost of the search.

[0029] Furthermore, the specific method for constructing the LFM model of the present invention includes: obtaining the historical data of the user on the think tank data, wherein the historical data includes the download data, click data, browsing data and search data of the user on the think tank data, wherein the user downloads the corresponding type of think tank data to construct the user's interest matrix R u,i , and the interest matrix R u,i According to the latent type data, matrix decomposition is performed to obtain the latent interest matrix And the latent interest relationship matrix , at this time, the user's interest matrix can be expressed by the following formula: R u,i = , where u is the user, T is the transposed matrix identifier, and i is the think tank data type.

[0030] It is worth mentioning that the system automatically obtains click data, browsing data and search data of the corresponding think tank data in the historical data. The above data is based on the search and matching mode of the KMP algorithm. Therefore, the historical data in the present invention is also generated based on the matching mode of the KMP algorithm. According to the user's click data, browsing data and search data of the corresponding type of think tank data, the above click data, browsing data and search data are used to determine the hidden data, wherein the hidden data includes the number of pattern string matches of the above user input data in the corresponding think tank data triple information, the browsing data includes the user's browsing time in the corresponding triple think tank data, and the click data includes the number of clicks on the corresponding think tank data, which is used to determine the hidden data. After the hidden data is determined, the interest matrix is ​​calculated according to the following formula: preference (u, i) = R u,i = = , where K is the latent class dimension, The u and k in it represent the relationship between user interests and latent data. The k and i in the table represent the relationship between latent data and corresponding think tank data types. Please refer to Figure 3 , Figure 3 Columns A, B, C, and D represent user interests, while V, W, X, and Y represent the corresponding think tank data types, and k1 and k2 represent latent class data. The above matrix sizes and dimensions are only examples, and the present invention is not limited thereto.

[0031] When the system decomposes the user's interest matrix to obtain the user's latent interest relationship matrix for think tank data, think tank data and latent relationship matrix, the LFM model is further trained according to the acquired historical data, and the user's interest matrix R is calculated. u,i = = The sum of squared errors: E u,i = ,in Represents the true value of the user interest matrix, and minimizes the sum of squared errors minE u,i The LFM model is used as the trained LFM model.

[0032] After completing the LFM model, the LFM model is used to identify the user's identity information after the user logs into the system, and the trained LFM model is used to recommend think tank data types to the user, thereby realizing the mining and recommendation of think tank data on user hidden interest data and improving the recommendation experience of think tank data.

[0033] The embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. The embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium. When the computer program is executed by the central processing unit (CPU), the above-mentioned functions defined in the method of the present application are executed. It should be noted that the above-mentioned computer-readable medium of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of an electrical, magnetic, optical, electromagnetic, infrared segment, or semiconductor, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection with one or more wire segments, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, electrical wire, optical cable, RF, etc., or any suitable combination of the foregoing.

[0034] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0035] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments. Without departing from the principles, the implementation methods of the present invention may be deformed or modified in any way.

Claims

1. A think tank data recommendation method based on the LFM model, characterized in that: The method comprises: Pre-constructing a knowledge graph of think tank data, and obtaining triple information of corresponding think tank data nodes according to the knowledge graph, obtaining user input data, and querying the triple information using a KMP algorithm according to the input data to obtain first query data; The first query data is saved as historical data of think tank data, and the operation data of the user on the think tank data is obtained to construct the user's interest matrix on the think tank data; Performing matrix decomposition on the user's interest matrix for think tank data to obtain a user's latent class interest relationship matrix for think tank data, a think tank data and latent class relationship matrix, which are used to construct an LFM model; The LFM model is trained according to the latent data obtained from the historical data of the think tank data to obtain a trained think tank data recommendation model; The method for constructing the LFM model includes: obtaining historical data of users on the think tank data, wherein the historical data includes download data, click data, browsing data and search data of the think tank data by the users, and constructing the user's interest matrix R according to the think tank data of the corresponding type downloaded by the users. u,i , and the interest matrix R u,i According to the latent type data, matrix decomposition is performed to obtain the latent interest matrix And the latent interest relationship matrix , at this time, the user's interest matrix can be expressed by the following formula: R u,i = , where u is the user, T is the transposed matrix identifier, and i is the think tank data type; The method for constructing the LFM model includes: after obtaining the click data, browsing data and search data of the user for the corresponding type of think tank, determining the latent data according to the click data, browsing data and search data, wherein the latent data includes the number of pattern string matches of the user input data in the corresponding think tank data triple information, the browsing data includes the browsing time of the user in the corresponding triple think tank data, and the click data includes the number of clicks on the corresponding think tank data, which is used to determine the latent data. After determining the latent data, the interest matrix is ​​calculated according to the following formula: preference (u, i) = R u,i = = , where K is the latent class dimension, The u and k in it represent the relationship between user interests and latent data. The k and i in it represent the relationship between latent class data and the corresponding think tank data type.

2. According to the method for recommending think tank data based on the LFM model in claim 1, it is characterized in that: The search method of the KMP algorithm includes: converting the triple information in the knowledge graph into a text string, and converting the user's input data into a pattern string, performing character matching between the pattern string and the text string until all characters in the current pattern string are matched in the text string, then obtaining the matched triple information, and extracting the entity data in the matched triple information, saving it and using it to construct the historical data of the think tank data.

3. According to claim 2, a method for recommending think tank data based on the LFM model is characterized in that: The search method of the KMP algorithm includes: obtaining user input data and converting it into a pattern string, the pattern string includes k characters arranged in sequence, the text string includes m characters, wherein m>k, starting to match the pattern string of k characters with the starting position of the text string of the triple information, calculating that there are k1 characters of the pattern string and a first matching string with the same sequence in the text string, wherein k1<k, then moving x characters of the pattern string to match the text string, so that there are at least the above k1 characters and the first matching string with the same sequence in the pattern string to match the pattern string, and continuing to move the pattern string until the characters of the k pattern strings are all matched in the text string.

4. According to claim 3, a method for recommending think tank data based on the LFM model is characterized in that: When the k pattern strings find n identical pattern strings in the M characters, locate each triple position where the n pattern strings are found, count the number of the pattern strings existing in each triple information, and sort them in descending order according to the number of pattern strings, and give priority to outputting the think tank data corresponding to the triple with the largest number of pattern string matches.

5. According to claim 3, a method for recommending think tank data based on the LFM model is characterized in that: When the k-character pattern string is not matched in the text string data, at least one character is deleted from the k-character pattern string to obtain k n The k-character pattern string is matched again with the text string, where the k n The method for obtaining a pattern string of k characters includes: extracting keywords from an input sentence using word segmentation and word embedding technology, and converting the extracted at least one keyword into the k n The pattern string of characters is matched with the corresponding text string of M characters to obtain k information of each triple n The number of matches for a pattern string of characters.

6. The method for recommending think tank data based on the LFM model according to claim 1, characterized in that: After the user's interest matrix is ​​decomposed to obtain the user's latent interest relationship matrix for think tank data, think tank data and latent relationship matrix, the LFM model is further trained according to the acquired historical data, and the user's interest matrix R is calculated. u,i = = The sum of squared errors: E u,i = ,in represents the true value of the user interest matrix, and the minE with the smallest sum of squared errors u,i The LFM model is used as the trained LFM model.

7. A think tank data recommendation system based on the LFM model, characterized in that: The system executes a think tank data recommendation method based on the LFM model as described in any one of claims 1 to 6 above.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement a think tank data recommendation method based on the LFM model as described in any one of claims 1 to 6.