A related document recommendation system and method based on keyword co-occurrence relationship

Abstract

The present invention provides a system and method for recommending related documents based on keyword co-occurrence relationships, wherein the system includes: a data preparation subsystem for cleaning document metadata to obtain target processing data; a parameter calculation subsystem for calculating target parameters based on the target processing data; a relevance calculation subsystem for determining the recommended ranking of related documents based on the target parameters; and a document recommendation subsystem for recommending documents based on the recommended ranking of related documents. The system and method for recommending related documents based on keyword co-occurrence relationships of the present invention cleans document metadata to obtain target processing data, calculates keyword frequency, keyword weight, and keyword co-occurrence network based on the target processing data, and further calculates the recommended ranking of related documents, and recommends documents based on the recommended ranking of related documents, and can quickly recommend extended documents related to the searched document from a large amount of documents based on a document, thereby improving the efficiency of document retrieval.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining, and in particular to a system and method for recommending related documents based on keyword co-occurrence relationships. Background Art

[0002] With the continuous deepening of academic research and the rapid growth of the number of literature, researchers are faced with the challenge of how to quickly and accurately obtain the required information from massive literature. Researchers need to innovate in the process of academic research, and innovations can be mined from interdisciplinary subjects. This requires researchers to not only find literature directly related to the research point, but also find interdisciplinary literature related to the research point, and find innovations by reading and studying interdisciplinary literature.

[0003] Traditional literature retrieval and recommendation methods are mostly based on the references, citing documents and co-cited documents of a document, and are unable to find other documents related to the research point from a large amount of literature.

[0004] In view of this, there is an urgent need for a related document recommendation system and method based on keyword co-occurrence relationships to at least solve the above-mentioned shortcomings. Summary of the invention

[0005] One of the purposes of the present invention is to provide a system and method for recommending related documents based on keyword co-occurrence relationships, which performs data cleaning on the acquired document metadata to obtain target processing data, calculates keyword frequency, keyword weight and keyword co-occurrence network based on the target processing data, and further calculates the recommended ranking of related documents. Document recommendation is performed based on the recommended ranking of related documents, and extended documents related to the search file can be quickly recommended from a large number of documents based on one document, thereby improving the efficiency of document retrieval.

[0006] An embodiment of the present invention provides a related document recommendation system based on keyword co-occurrence relationship, including:

[0007] The data preparation subsystem is used to obtain document metadata and clean the document metadata to obtain target processing data;

[0008] The parameter calculation subsystem is used to process data according to the target and calculate the target parameters, including: keyword frequency, keyword weight and keyword co-occurrence network;

[0009] The relevance calculation subsystem is used to determine the recommended ranking of relevant documents based on the target parameters;

[0010] The document recommendation subsystem is used to recommend documents based on the recommended ranking of related documents.

[0011] Preferably, the data preparation subsystem includes:

[0012] The document metadata preparation module is used to obtain document metadata based on a preset document entry library;

[0013] A target keyword acquisition module is used to acquire target keywords;

[0014] The keyword data cleaning module is used to remove the document metadata without the target keyword to obtain the first processed data, normalize the first processed data based on the preset normalization processing rules to obtain the second processed data, and remove the low-frequency data according to the second processed data to obtain the target processed data.

[0015] Preferably, the parameter calculation subsystem includes:

[0016] The word frequency statistics module is used to process data according to the target, perform keyword frequency statistics, and obtain the keyword frequency matrix and the document keyword relationship matrix;

[0017] The keyword weight calculation module is used to calculate the keyword weight according to the statistical results of keyword frequency statistics;

[0018] The keyword co-occurrence network calculation module is used to analyze the co-occurrence relationship of the target keywords, build a keyword co-occurrence network and calculate the co-occurrence intensity.

[0019] Preferably, the relevance calculation subsystem includes:

[0020] A search keyword acquisition module is used to acquire search keywords;

[0021] The document relevance calculation module is used to calculate the document relevance based on the target parameters and search keywords;

[0022] The relevance ranking module is used to recommend and sort related documents in descending order of their relevance.

[0023] Preferably, the search keyword acquisition module includes:

[0024] The first search keyword acquisition submodule is used to acquire the search keyword input by the target person;

[0025] and / or,

[0026] The second search keyword acquisition submodule is used to determine the search keyword according to the query document input by the target person.

[0027] Preferably, the second acquisition submodule for retrieving keywords includes:

[0028] A historical query record set expansion unit is used to obtain the historical query record set of the query document and expand the historical query record set on the target time axis;

[0029] An average search time acquisition unit, used to acquire an average search time;

[0030] A retrieval time window determination unit, used to determine the retrieval time window according to the average retrieval time;

[0031] A retrieval time window traversal unit, used for moving the retrieval time window on the target time axis;

[0032] A positioning condition satisfaction determination unit, used to determine whether the search time window satisfies the positioning condition;

[0033] A historical query record determination unit, used to determine the historical query record closest to the current time in each positioning time window if the search time window meets the positioning condition;

[0034] A historical query record parsing unit, used to parse the historical query records and determine the target query person and target query keywords;

[0035] A search experience information acquisition unit, used to acquire the search experience information of the target query person;

[0036] The search keyword determination unit is used to determine the target query keyword in the target query record as the search keyword for querying documents according to the search experience information.

[0037] Preferably, the positioning condition satisfaction determination unit includes:

[0038] A comparison record first determination subunit is used to use the historical query records within the retrieval time window as the first comparison record and the historical query records outside the retrieval time window as the second comparison record, and the retrieval time window is divided into left and right sides;

[0039] The comparison record second determination subunit is used to determine the first comparison record closest to the retrieval time window side and use it as the third comparison record;

[0040] The comparison record third determination subunit is used to determine the second comparison record closest to the search time window side and use it as the fourth comparison record;

[0041] A time interval first calculation subunit, used for obtaining a first time interval between a third comparison record and a fourth comparison record closest to the same retrieval time window side;

[0042] The second time interval calculation subunit is used to obtain the second time interval between the third comparison record closest to the same retrieval time window and the corresponding retrieval time window and the third time interval between the fourth comparison record and the corresponding retrieval time window;

[0043] The determination subunit is configured to determine that the retrieval time window satisfies the positioning condition if the first time interval is greater than or equal to a preset time interval threshold and the second time interval is less than a third time interval.

[0044] Preferably, the document recommendation subsystem further includes:

[0045] A historical document browsing information acquisition module is used to obtain the historical document browsing information of the target person before recommending documents according to the recommended order of related documents;

[0046] A browsing effective value determination module, used to determine the browsing effective value of the first historical browsing document according to the historical document browsing information;

[0047] A first target historical browsing document determination module is used to determine the first historical browsing document as the target historical browsing document if the browsing validity value is greater than or equal to a preset browsing validity value threshold;

[0048] A second historical browsing document determination module, for determining the first historical browsing document as the second historical browsing document if the browsing validity value is less than a preset browsing validity value threshold;

[0049] A document feature extraction module, used to extract document features of the second historical browsing document;

[0050] A second target historical browsing document determination module is used to determine the target historical browsing document in the second historical browsing document according to document characteristics and a preset determination document characteristic library;

[0051] The literature recommendation module is used to make literature recommendations after removing the target historically browsed literature from the related literature recommendation sorting;

[0052] Among them, the browsing effective value determination module includes:

[0053] A target paragraph determination submodule, used to determine each target paragraph of the first historical browsing document according to the historical document browsing information;

[0054] The paragraph type acquisition submodule is used to obtain the paragraph type of the target paragraph;

[0055] An ideal viewing time determination submodule is used to determine the ideal viewing time of a target paragraph according to a character count-ideal viewing time comparison library corresponding to the paragraph type;

[0056] The actual browsing time acquisition submodule is used to obtain the actual browsing time of the target paragraph;

[0057] A browsing value determination submodule is used to determine the browsing value based on the actual browsing time and the ideal viewing time;

[0058] The browsing effective value determination submodule is used to determine the browsing effective value of the first historical browsing document according to the browsing value and the preset target weight corresponding to the paragraph type.

[0059] An embodiment of the present invention provides a method for recommending related documents based on keyword co-occurrence relationships, including:

[0060] Step 1: Obtain document metadata and perform data cleaning on the document metadata to obtain target processing data;

[0061] Step 2: Process the data according to the target and calculate the target parameters, which include: keyword frequency, keyword weight and keyword co-occurrence network;

[0062] Step 3: Determine the recommended ranking of relevant literature based on the target parameters;

[0063] Step 4: Recommend literature based on the recommended ranking of relevant literature.

[0064] Preferably, step 1: obtaining document metadata and performing data cleaning on the document metadata to obtain target processing data includes:

[0065] Get target keywords;

[0066] Obtain document metadata based on the preset document entry library;

[0067] Document metadata without target keywords is eliminated to obtain first processed data, the first processed data is normalized based on a preset normalization processing rule to obtain second processed data, and low-frequency data is removed according to the second processed data to obtain target processed data.

[0068] Preferably, step 2: calculating target parameters according to target processing data, the target parameters include: keyword frequency, keyword weight and keyword co-occurrence network, including:

[0069] Process the data according to the target, perform keyword frequency statistics, and obtain the keyword frequency matrix and the literature keyword relationship matrix;

[0070] Calculate keyword weight based on the statistical results of keyword frequency statistics;

[0071] Analyze the co-occurrence relationship of the target keywords, build a keyword co-occurrence network and calculate the co-occurrence intensity.

[0072] The beneficial effects of the present invention are:

[0073] The present invention performs data cleaning on the acquired document metadata to obtain target processing data, calculates keyword frequency, keyword weight and keyword co-occurrence network according to the target processing data, further calculates the recommended ranking of related documents, and recommends documents according to the recommended ranking of related documents. It can quickly recommend extended documents related to the search file from massive documents based on one document, thereby improving the efficiency of document retrieval.

[0074] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in the present application documents.

[0075] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0077] Figure 1 A schematic diagram of a related document recommendation system based on keyword co-occurrence relationship in an embodiment of the present invention;

[0078] Figure 2 Schematic diagram of a method for recommending related documents based on keyword co-occurrence relationships in an embodiment of the present invention. DETAILED DESCRIPTION

[0079] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0080] The embodiment of the present invention provides a related document recommendation system based on keyword co-occurrence relationship, such as Figure 1 As shown, including:

[0081] Data preparation subsystem 1 is used to obtain document metadata and clean the document metadata to obtain target processing data; document metadata refers to the basic information of documents stored in the retrieval database, including author, title, year of publication, journal name, volume number, issue number, and keywords; data cleaning refers to the process of identifying and correcting (or deleting) erroneous data, duplicate data, incomplete data, etc. to improve data quality; target processing data refers to the cleaned document metadata;

[0082] Parameter calculation subsystem 2 is used to process data according to the target and calculate the target parameters. The target parameters include: keyword frequency, keyword weight and keyword co-occurrence network; wherein, keyword frequency is: the frequency of occurrence of keywords in the document; keyword weight represents the importance or relevance of keywords in the document, which can be calculated by different algorithms, such as: TF-IDF (term frequency-inverse document frequency); keyword co-occurrence network is a visual network showing the relationship between keywords in the document, and the nodes in the co-occurrence network represent keywords, and the edges represent the co-occurrence relationship between keywords;

[0083] The relevance calculation subsystem 3 is used to determine the recommended order of related documents according to the target parameters; wherein the recommended order of related documents is: the order from largest to smallest of the document relevance calculated based on the target parameters;

[0084] The document recommendation subsystem 4 is used to recommend documents according to the recommended ranking of related documents. The document recommendation according to the recommended ranking of related documents includes: visually displaying the recommended ranking results of related documents to the search user, and recommending the above-mentioned related documents to the search user.

[0085] The working principle and beneficial effects of the above technical solution are:

[0086] The present invention performs data cleaning on the acquired document metadata to obtain target processing data, calculates keyword frequency, keyword weight and keyword co-occurrence network according to the target processing data, further calculates the recommended ranking of related documents, and recommends documents according to the recommended ranking of related documents. It can quickly recommend extended documents related to the search file from massive documents based on one document, thereby improving the efficiency of document retrieval.

[0087] In one embodiment, the data preparation subsystem includes:

[0088] The document metadata preparation module is used to obtain document metadata based on a preset document entry library; wherein the preset document entry library is: a database for storing and managing documents;

[0089] The target keyword acquisition module is used to acquire the target keyword; the target keyword is: the most representative word selected by researchers when writing a document based on the content of the document, and is acquired from the document metadata;

[0090] The keyword data cleaning module is used to remove the document metadata without the target keyword, obtain the first processed data, perform normalization processing on the first processed data based on the preset normalization processing rule to obtain the second processed data, and remove the low-frequency data according to the second processed data to obtain the target processed data; wherein the preset normalization processing rule is: convert English letters to lowercase; convert English words to singular; only keep one space between words; convert full-width characters to half-width characters; convert traditional Chinese to simplified Chinese; convert Chinese characters and numbers to Arabic numbers; _--&—-,|:,.、。 :&(){}[](){}【】《》| is replaced with a space; ! @$%#^*+="'<>? / ~!@#￥%……*+=;' / “? character is replaced with a blank character; the low-frequency data is: the keyword whose appearance frequency in the document metadata is less than or equal to the preset appearance frequency threshold.

[0091] The working principle and beneficial effects of the above technical solution are:

[0092] The present invention introduces a document entry library to obtain document metadata, and determines target keywords, removes document metadata without target keywords, and introduces preset normalization processing rules to normalize the remaining keyword fields, and then removes low-frequency keywords to obtain cleaned target processed data, thereby improving the rationality of keyword data preparation.

[0093] In one embodiment, the parameter calculation subsystem includes:

[0094] The word frequency statistics module is used to process data according to the target, perform keyword frequency statistics, and obtain a keyword frequency matrix and a document keyword relationship matrix; wherein the keyword frequency matrix is: a description matrix of keywords and corresponding word frequencies; the document keyword relationship matrix is: a description matrix of keywords, document IDs and corresponding word frequencies;

[0095] The keyword weight calculation module is used to calculate the keyword weight according to the statistical results of the keyword frequency statistics; wherein, the calculation formula of the keyword weight is:

[0096] W i =TF i ×IDF i

[0097] Among them, W i represents the weight of the target keyword i, TF i Indicates the frequency of the target keyword appearing in the document, IDF i Represents the inverse document frequency of the target keyword i, IDF i The calculation formula is as follows:

[0098]

[0099] Where N represents the total number of documents, n i represents the number of documents containing the target keyword i, and lg is the logarithm with base 10;

[0100] The keyword co-occurrence network calculation module is used to analyze the co-occurrence relationship of the target keywords, build a keyword co-occurrence network and calculate the co-occurrence intensity. The system analyzes the co-occurrence relationship of the extracted keywords and builds a keyword co-occurrence network. In the co-occurrence network, nodes represent keywords, and the edges between nodes represent the co-occurrence relationship of keywords in the literature. The weight of the edge represents the frequency or intensity of co-occurrence. The calculation formula of co-occurrence frequency is as follows:

[0101] Co ij =∑ d∈D Count(i,j,d)

[0102] Among them, Co ij represents the co-occurrence frequency of target keyword i and target keyword j, D represents the document set, and Count(i,j,d) represents the co-occurrence times of target keyword i and target keyword j in document d;

[0103] Normalize the co-occurrence frequency to get the co-occurrence intensity S ij :

[0104]

[0105] Among them, Co ii represents the frequency of occurrence of target keyword i in the literature where target keyword i is located; Co jj Indicates the frequency of occurrence of target keyword j in the literature where target keyword j is located.

[0106] The working principle and beneficial effects of the above technical solution are:

[0107] The present invention performs keyword frequency statistics on target processing data, calculates keyword weights and constructs a keyword co-occurrence network based on the statistical results, and the process of determining target parameters is more appropriate.

[0108] In one embodiment, the relevance calculation subsystem includes:

[0109] The search keyword acquisition module is used to acquire the search keyword; wherein the search keyword is: the keyword that the searcher needs to search for;

[0110] The document relevance calculation module is used to calculate the document relevance according to the target parameters and the search keywords; the calculation formula of the document relevance is:

[0111] Relevance(A,B)=∑ (ka,kb∈Pairs(KA,KB)) (S ka,kb×W(ka)×W(kb)×MatchFactor(ka,kb,A,B))

[0112] Among them, Relevance(A,B) represents the relevance between document A and document B; KA and KB represent the keyword sets of document A and document B respectively; Pairs(KA,KB) is a set containing all possible keyword pairs (ka,kb) generated after deduplication from the keyword set KA of document A and the keyword set KB of document B, where ka is a keyword after word segmentation of document A and kb represents a keyword after word segmentation of document B; S ka,kb Indicates the co-occurrence strength of keywords ka and kb. If there is no co-occurrence relationship in the keyword network, the value is 0; W(ka) is the weight of keyword ka; W(kb) is the weight of keyword kb; MatchFactor(ka,kb,A,B) is whether the matching positions of keywords ka and kb in document A and document B are consistent. If they are consistent, it is 1, otherwise it is 0.5;

[0113] The relevance ranking module is used to recommend and sort related documents in descending order of their relevance.

[0114] The working principle and beneficial effects of the above technical solution are:

[0115] The present invention calculates the document relevance according to the acquired search keywords and target parameters, and sorts the document relevance in descending order for subsequent recommendation, thereby achieving higher recommendation efficiency.

[0116] In one embodiment, the search keyword acquisition module includes:

[0117] The first search keyword acquisition submodule is used to acquire the search keywords input by the target person; wherein the target person is: a document search person;

[0118] and / or,

[0119] The second search keyword acquisition submodule is used to determine the search keyword according to the query document input by the target person.

[0120] The working principle and beneficial effects of the above technical solution are:

[0121] There are two situations when obtaining search keywords. The first situation is that the document search personnel directly input the search keywords for document search. The second situation is that the document search personnel input the search query documents, and the search keywords of related documents can be extracted based on the query documents. Introducing two methods to obtain search keywords is more comprehensive.

[0122] In one embodiment, the second keyword acquisition submodule includes:

[0123] The historical query record set expansion unit is used to obtain the historical query record set of the query document and expand the historical query record set on the target time axis; wherein the historical query record set includes: a collection of records containing historical query documents; when the historical query record set is expanded on the target time axis, the query time corresponding to the historical query record and the point on the target time axis correspond one to one;

[0124] The average search time acquisition unit is used to acquire the average search time; wherein the average search time is: the average time required for a user who searches for documents to complete a document search. A complete document search includes: inputting keywords, browsing results and selecting related documents;

[0125] A retrieval time window determination unit is used to determine a retrieval time window according to an average retrieval time; wherein the retrieval time window is: a time period having a time interval equal to the average retrieval time corresponding to a framed area on the time axis;

[0126] A retrieval time window traversal unit, used for moving the retrieval time window on the target time axis;

[0127] A positioning condition satisfaction determination unit is used to determine whether the search time window satisfies the positioning condition; wherein the positioning condition is: a search time window movement stop condition;

[0128] A historical query record determination unit, used to determine the historical query record closest to the current time in each positioning time window if the search time window meets the positioning condition;

[0129] A historical query record parsing unit, used to parse the historical query records and determine the target query person and target query keywords;

[0130] A search experience information acquisition unit is used to acquire the search experience information of the target query personnel; wherein the search experience information includes: search training experience, search practice years and search success rate, etc.;

[0131] The search keyword determination unit is used to determine the target query keyword in the target query record as the search keyword for querying documents based on the search experience information. When determining the target query keyword in the target query record as the search keyword for querying documents, the experience degree represented by the search experience information is quantified as an experience value, and the target query keyword in the target query record corresponding to the largest experience value is used as the search keyword for querying documents.

[0132] The working principle and beneficial effects of the above technical solution are:

[0133] The present invention obtains a historical query record set of historical queries for searching documents, and expands it on a target time axis based on the search time corresponding to each historical query record in the historical query record set; since the search is generally not successful at one time, and often requires multiple and repeated searches within a period of time, the average search time of the search process to complete a search task is obtained, and a time period with a time interval of the average search time is determined to correspond to a search time window on the time axis; the search time window is moved on the time axis to determine whether the search time window meets a stop condition, and when the stop condition is met, the historical query record closest to the current time in the positioning time window is obtained, that is, the query record containing relevant information of the final search formula, and the target query person and the target query keyword of the corresponding historical query record are obtained; in addition, considering that each target query person has different search experience, the search experience information is introduced to quantify the search experience of the target query person, and the target query keyword in the corresponding target query record with the largest experience value is screened as the search keyword for querying documents, thereby improving the acquisition accuracy of the search keyword.

[0134] In one embodiment, the positioning condition satisfaction determination unit includes:

[0135] A comparison record first determination subunit is used to use the historical query records within the retrieval time window as the first comparison record and the historical query records outside the retrieval time window as the second comparison record, and the retrieval time window is divided into left and right sides;

[0136] The comparison record second determination subunit is used to determine the first comparison record closest to the retrieval time window side and use it as the third comparison record;

[0137] The comparison record third determination subunit is used to determine the second comparison record closest to the search time window side and use it as the fourth comparison record;

[0138] The first time interval calculation subunit is used to obtain a first time interval between the third comparison record and the fourth comparison record closest to the same search time window side; wherein the first time interval is: the time difference between the recording time of the third comparison record and the fourth comparison record closest to the same search time window side, for example: 2 seconds;

[0139] The second time interval calculation subunit is used to obtain the second time interval between the third comparison record closest to the same retrieval time window side and the corresponding retrieval time window side and the third time interval between the fourth comparison record and the corresponding retrieval time window side; wherein the second time interval is: the time difference between the recording time of the third comparison record closest to the same retrieval time window side and the cutoff time of the corresponding closest retrieval time window side and the time axis, such as: 10 seconds; the third time interval is: the time difference between the recording time of the fourth comparison record closest to the same retrieval time window side and the cutoff time of the corresponding closest retrieval time window side and the time axis, such as: 2 hours;

[0140] The determination subunit is used to determine that if the first time interval is greater than or equal to a preset time interval threshold, and the second time interval is less than a third time interval, the search time window meets the positioning condition. The preset time interval threshold is manually preset, such as the average search time.

[0141] The working principle and beneficial effects of the above technical solution are:

[0142] The present invention determines the third comparison record and the fourth comparison record closest to the same retrieval time window side, calculates the first time interval, and when the first time interval is large, it means that the retrieval time interval between the third comparison record and the fourth comparison record is long, and the possibility of belonging to the same retrieval task is low, then the first comparison record contained in the retrieval time window has a greater probability of belonging to the same retrieval task; obtains the second time interval of the third comparison record closest to the same retrieval time window side from the corresponding retrieval time window side and the third time interval of the fourth comparison record from the corresponding retrieval time window side, and when the second time interval is less than the third time interval, it means that the closer the third comparison record is to the retrieval time window and within the retrieval time window, the more likely it is that it belongs to the same retrieval task with other first comparison records within the retrieval time window, and utilizes the above two conditions to constrain the positioning of the retrieval time window, thereby improving the record extraction accuracy of multiple retrieval records of the same retrieval task within a period of time, and improving the accuracy of subsequent determination of effective retrieval records within the period of time.

[0143] In one embodiment, the document recommendation subsystem further includes:

[0144] The historical document browsing information acquisition module is used to obtain the historical document browsing information of the target person before recommending documents according to the relevant document recommendation ranking; the historical document browsing information is: the document browsing records from the time when the target person performs this search task to the current time;

[0145] The browsing effective value determination module is used to determine the browsing effective value of the first historical browsing document according to the historical document browsing information; wherein the first historical browsing document is: the browsing document obtained by parsing the historical document browsing information; the browsing effective value represents the viewing effectiveness of the first historical browsing document;

[0146] A first target historical browsing document determination module is used to determine the first historical browsing document as the target historical browsing document if the browsing validity value is greater than or equal to a preset browsing validity value threshold; wherein the preset browsing validity value threshold is manually preset;

[0147] A second historical browsing document determination module, for determining the first historical browsing document as the second historical browsing document if the browsing validity value is less than a preset browsing validity value threshold;

[0148] The document feature extraction module is used to extract the document features of the second historical browsing document; wherein the document feature is a characteristic representation of the document information, such as: the author is someone;

[0149] The second target historical browsing document determination module is used to determine the target historical browsing document in the second historical browsing document according to the document characteristics and the preset determination document characteristic library; wherein the preset determination document characteristic library stores multiple determination document characteristics, such as: the operation is the target personnel;

[0150] The literature recommendation module is used to make literature recommendations after removing the target historically browsed literature from the related literature recommendation sorting;

[0151] Among them, the browsing effective value determination module includes:

[0152] The target paragraph determination submodule is used to determine each target paragraph of the first historical browsing document according to the historical document browsing information; wherein the target paragraph is divided based on the structure of the first historical browsing document, such as: abstract paragraph, body paragraph, etc.;

[0153] The paragraph type acquisition submodule is used to acquire the paragraph type of the target paragraph; wherein the paragraph type is a paragraph type divided based on the structure of the first historical browsed document, such as abstract, body, etc.;

[0154] The ideal viewing time determination submodule is used to determine the ideal viewing time of the target paragraph according to the character number-ideal viewing time comparison library corresponding to the paragraph type; wherein the character number-ideal viewing time comparison library stores the corresponding relationship between the character number and the ideal viewing time, and the corresponding relationship is in a proportional relationship;

[0155] The actual browsing time acquisition submodule is used to acquire the actual browsing time of the target paragraph; wherein the actual browsing time is acquired based on the sight line information of the target paragraph;

[0156] The browsing value determination submodule is used to determine the browsing value according to the actual browsing time and the ideal viewing time; wherein the browsing value is: the actual browsing time divided by the ideal viewing time;

[0157] The browsing effective value determination submodule is used to determine the browsing effective value of the first historical browsing document according to the browsing value and the preset target weight corresponding to the paragraph type. The preset target weight corresponding to the paragraph type is manually preset, for example: the abstract is 0.4, the text is 0.3, and the acknowledgment is 0.05. The sum of the target weights of all paragraph types is 1. The browsing effective value is the sum of the corresponding products of the browsing value corresponding to each target paragraph of the first historical browsing document and the target weight of the paragraph type of the target paragraph.

[0158] The working principle and beneficial effects of the above technical solution are:

[0159] When the system recommends relevant documents, there may be a situation where the target person has already consulted the recommended documents directly generated based on the input keywords, so there is no need to recommend them again. Therefore, the historical document browsing information of the target person is introduced to determine multiple target paragraphs in the first historical browsing document; the paragraph type of the target paragraph is obtained, the ideal viewing time reference library corresponding to the paragraph type is determined, and the ideal viewing time of the target paragraph is determined based on the number of characters of the target paragraph, and the actual browsing time is determined based on the viewing line of sight of the target paragraph, and the browsing value is obtained by dividing the actual browsing time by the ideal viewing time; since the content of each paragraph type is different, its value weight (i.e., target weight) is also different, for example: the abstract is a brief summary of the full text, which can help readers understand the content of the article in the shortest time, so it is given a higher weight, and the acknowledgments are mostly the author's insights into the writing and learning process, so they are given a lower weight. In general, the sum of the target weights of all paragraph types is 1, and the sum of the products of the browsing value corresponding to each target paragraph in the first historical browsing document and the target weight of the paragraph type is taken as the browsing effective value, and the browsing effective value is more accurate.

[0160] The embodiment of the present invention provides a method for recommending related documents based on keyword co-occurrence relationship. Figure 2 As shown, including:

[0161] Step 1: Obtain document metadata and perform data cleaning on the document metadata to obtain target processing data;

[0162] Step 2: Process the data according to the target and calculate the target parameters, which include: keyword frequency, keyword weight and keyword co-occurrence network;

[0163] Step 3: Determine the recommended ranking of relevant literature based on the target parameters;

[0164] Step 4: Recommend literature based on the recommended ranking of relevant literature.

[0165] In one embodiment, step 1: obtaining document metadata and performing data cleaning on the document metadata to obtain target processed data includes:

[0166] Get target keywords;

[0167] Obtain document metadata based on the preset document entry library;

[0168] Document metadata without target keywords is eliminated to obtain first processed data, the first processed data is normalized based on a preset normalization processing rule to obtain second processed data, and low-frequency data is removed according to the second processed data to obtain target processed data.

[0169] In one embodiment, step 2: Calculate target parameters according to target processing data, the target parameters include: keyword frequency, keyword weight and keyword co-occurrence network, including:

[0170] Process the data according to the target, perform keyword frequency statistics, and obtain the keyword frequency matrix and the literature keyword relationship matrix;

[0171] Calculate keyword weight based on the statistical results of keyword frequency statistics;

[0172] Analyze the co-occurrence relationship of the target keywords, build a keyword co-occurrence network and calculate the co-occurrence intensity.

[0173] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A related document recommendation system based on keyword co-occurrence relationship, characterized in that: include: The data preparation subsystem is used to obtain document metadata and clean the document metadata to obtain target processing data; The parameter calculation subsystem is used to process data according to the target and calculate the target parameters, including: keyword frequency, keyword weight and keyword co-occurrence network; A relevance calculation subsystem is used to obtain a historical query record set of the query document and expand the historical query record set on a target time axis; Get the average search time; Determine the retrieval time window based on the average retrieval time; Move the retrieval time window on the target time axis; The historical query records within the retrieval time window are used as the first comparison records, and the historical query records outside the retrieval time window are used as the second comparison records, and the retrieval time window is divided into left and right sides; Determine the first comparison record closest to the retrieval time window side and use it as the third comparison record; Determine a second comparison record closest to the retrieval time window side and use it as a fourth comparison record; Obtaining a first time interval between a third comparison record and a fourth comparison record that are closest to the same retrieval time window side; Obtain a second time interval between a third comparison record closest to the same retrieval time window and the corresponding retrieval time window, and a third time interval between a fourth comparison record and the corresponding retrieval time window; If the first time interval is greater than or equal to the preset time interval threshold, and the second time interval is less than the third time interval, then the search time window meets the positioning condition; If the search time window meets the positioning condition, determine the historical query record closest to the current time in each positioning time window; Analyze historical query records to determine target query personnel and target query keywords; Obtain the search experience information of the target query personnel; According to the search experience information, the target query keyword in the target query record is determined as the search keyword for searching the document; Determine the recommended ranking of relevant literature based on target parameters and search keywords; The document recommendation subsystem is used to recommend documents based on the recommended ranking of related documents.

2. A related document recommendation system based on keyword co-occurrence relationship as claimed in claim 1, characterized in that: Data preparation subsystem, including: The document metadata preparation module is used to obtain document metadata based on a preset document entry library; A target keyword acquisition module is used to acquire target keywords; The keyword data cleaning module is used to remove the document metadata without the target keyword to obtain the first processed data, normalize the first processed data based on the preset normalization processing rules to obtain the second processed data, and remove the low-frequency data according to the second processed data to obtain the target processed data.

3. A related document recommendation system based on keyword co-occurrence relationship as claimed in claim 1, characterized in that: Parameter calculation subsystem, including: The word frequency statistics module is used to process data according to the target, perform keyword frequency statistics, and obtain the keyword frequency matrix and the document keyword relationship matrix; The keyword weight calculation module is used to calculate the keyword weight according to the statistical results of keyword frequency statistics; The keyword co-occurrence network calculation module is used to analyze the co-occurrence relationship of the target keywords, build a keyword co-occurrence network and calculate the co-occurrence intensity.

4. A method for recommending related documents based on keyword co-occurrence relationships, characterized in that: include: Step 1: Obtain document metadata and perform data cleaning on the document metadata to obtain target processing data; Step 2: Process the data according to the target and calculate the target parameters, which include: keyword frequency, keyword weight and keyword co-occurrence network; Step 3: Obtain the historical query record set of the query document, and expand the historical query record set on the target time axis; Get the average search time; Determine the retrieval time window based on the average retrieval time; Move the retrieval time window on the target time axis; The historical query records within the retrieval time window are used as the first comparison records, and the historical query records outside the retrieval time window are used as the second comparison records, and the retrieval time window is divided into left and right sides; Determine the first comparison record closest to the retrieval time window side and use it as the third comparison record; Determine a second comparison record closest to the retrieval time window side and use it as a fourth comparison record; Obtaining a first time interval between a third comparison record and a fourth comparison record that are closest to the same retrieval time window side; Obtain a second time interval between a third comparison record closest to the same retrieval time window and the corresponding retrieval time window, and a third time interval between a fourth comparison record and the corresponding retrieval time window; If the first time interval is greater than or equal to the preset time interval threshold, and the second time interval is less than the third time interval, then the search time window meets the positioning condition; If the search time window meets the positioning condition, determine the historical query record closest to the current time in each positioning time window; Analyze historical query records to determine target query personnel and target query keywords; Obtain the search experience information of the target query personnel; According to the search experience information, the target query keyword in the target query record is determined as the search keyword for searching the document; Determine the recommended ranking of relevant literature based on target parameters and search keywords; Step 4: Recommend literature based on the recommended ranking of relevant literature.

5. A method for recommending related documents based on keyword co-occurrence relationship as claimed in claim 4, characterized in that: Step 1: Obtain document metadata and clean the document metadata to obtain target processing data, including: Get target keywords; Obtain document metadata based on the preset document entry library; Document metadata without target keywords is eliminated to obtain first processed data, the first processed data is normalized based on a preset normalization processing rule to obtain second processed data, and low-frequency data is removed according to the second processed data to obtain target processed data.

6. A method for recommending related documents based on keyword co-occurrence relationship as claimed in claim 4, characterized in that: Step 2: Process the data according to the target and calculate the target parameters, which include: keyword frequency, keyword weight and keyword co-occurrence network, including: Process the data according to the target, perform keyword frequency statistics, and obtain the keyword frequency matrix and the literature keyword relationship matrix; Calculate keyword weight based on the statistical results of keyword frequency statistics; Analyze the co-occurrence relationship of the target keywords, build a keyword co-occurrence network and calculate the co-occurrence intensity.

Images