A knowledge expansion based data framing retrieval method and system

Abstract

The application provides a knowledge expansion-based data framing retrieval method and system, which comprises the following steps: retrieving according to selected keywords to obtain a retrieval result; completing the retrieval when the number of cycles reaches a set value, otherwise continuing to execute; when performing for the first time, the selected keywords are user input keywords; mining and analyzing the retrieval result to extract keywords; analyzing the relationship between the extracted keywords and the selected keywords and the relationship between the extracted keywords, and dividing the extracted keywords into multiple groups according to different relationship types; constructing different new query statements according to different groups, verifying the effectiveness of the keyword relationship, and putting the keywords verified as effective into a keyword library; taking the keywords in the keyword library as the selected keywords, increasing the number of cycles by 1, and re-executing the retrieval. The application has the advantages that it can retrieve and return information as comprehensively and accurately as possible, and solves the problem of retrieval concept drift.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of information retrieval and natural language processing, and particularly relates to a data framing retrieval method and system based on knowledge expansion. BACKGROUND

[0002] Information retrieval is an effective means of information acquisition in the context of data explosion. Traditional information retrieval aims to accurately understand user intent and retrieve relevant information. In terms of the breadth of information retrieved, information retrieval focuses more on the in-depth understanding of user intent and the accurate return of information relevant to the user intent due to the limitation of the number of feedback terms that the user can effectively read, and thus focuses on limited retrieval return breadth. Figure One

[0003] However, in actual applications such as data mining, web mining, information acquisition, and data analysis, relevant information is often needed to be returned in the widest possible range to obtain a relevant data set and perform data mining and other work in a specific field.

[0004] Therefore, the existing information retrieval returns too little search result in the fields of data mining, web mining, information acquisition, and data analysis, which cannot meet the retrieval requirements of the relevant fields. SUMMARY

[0005] The present application aims to overcome the defect that the existing retrieval technology returns too little search result, which cannot meet the retrieval requirements of the relevant fields.

[0006] To achieve the above-mentioned purpose, the present application provides a data framing retrieval method based on knowledge expansion, which comprises the following steps:

[0007] Step 1: retrieving according to the selected keywords to obtain the retrieval result; when the number of cycles reaches the set value, completing the retrieval, otherwise continuing to execute step 2; when executing step 1 for the first time, the selected keywords are the user input keywords;

[0008] Step 2: mining and analyzing the retrieval result to extract keywords;

[0009] Step 3: analyzing the relationship between the keywords extracted in step 2 and the keywords selected in step 1 and the relationship between the extracted keywords, and dividing the extracted keywords into multiple groups according to different relationship types;

[0010] Step 4: constructing different new query statements according to different groupings, verifying the effectiveness of the keyword relationship, putting the keywords verified as effective into a keyword library; taking the keywords in the keyword library as the selected keywords, setting the number of cycles to increase by 1, and returning to step 1.

[0011] ​As an improvement of the above method, the step 2 specifically comprises:

[0012] The retrieval results are initially preprocessed, including stop word removal, punctuation normalization and number standardization;

[0013] In the preprocessed retrieval results, first, the key words are obtained by word segmentation, and then the number of occurrences of the key words in the document is counted as the document frequency df1 of the key words;

[0014] Randomly sample n data from the searched data set, and count the document frequency df2 of the key words in the n data;

[0015] The scoring function is set as follows:

[0016]

[0017] Among them, tf_score represents the importance of the key words in the combined set of the retrieval results and the randomly sampled data calculated by the TF-IDF algorithm; tr_score represents the importance of the key words in the key word retrieval results analyzed by the TextRank algorithm; W1 and W2 represent the weights corresponding to the two importance degrees respectively, and W1+W2=1;

[0018] Select the key words with scores higher than the set value calculated by the above scoring function as the extracted key words.

[0019] As an improvement of the above method, the relationship between the extracted key words and the selected key words includes synonym relationship and other relationships;

[0020] The other relationships include: restrictive, mutually exclusive and co-occurrence relationships.

[0021] As an improvement of the above method, the determination method of the restrictive, mutually exclusive and co-occurrence relationships is:

[0022] Calculate the conditional probabilities of two key words A and B in the retrieval returned data as P(A|B) and P(B|A) respectively;

[0023] When the following relationship is satisfied, the key words A and B are in a restrictive relationship:

[0024] P(A|B)<p t P(B|A)>p′ t

[0025] Among them, p t and p' t respectively represent the threshold values of the respective conditional probabilities, and the value range is (0, 1);

[0026] When the following formula is satisfied, the key words A and B are in a mutually exclusive relationship:

[0027] P(A|B)<p t P(B|A)<p′ t

[0028] When the following formula is satisfied, the keywords A and B are in co-occurrence relationship:

[0029] P(A|B)>p t P(B|A)>p′ t .

[0030] As an improvement of the above method, p t is 0.3; p' t is 0.7.

[0031] As an improvement of the above method, in the keyword set of the defined relationship, the defined relationship is a directed edge, and the keyword is a node, to form a directed acyclic graph; if there are multiple edges between two nodes, only the longest edge of the defined relationship is retained.

[0032] As an improvement of the above method, the effectiveness of the keyword relationship is verified, and the specific process is:

[0033] For the three types of keywords of defined, mutually exclusive and co-occurrence relationship, the original keyword or synonym is taken as the defined condition, and a new query statement is constructed according to the relationship of the keyword;

[0034] The return data of the newly constructed query is formalized as a probability distribution P(x), that is, the frequency of all words after segmentation is counted as the probability distribution P(x) of the return data;

[0035] The step 1 retrieval result is formalized as a probability distribution Q(x);

[0036] The KL divergence KL(P||Q) of the two probability distributions is calculated:

[0037]

[0038] Wherein, x represents a keyword;

[0039] The relationship with the KL divergence less than a set threshold value is identified as effective.

[0040] The application also provides a data frame defined retrieval system based on knowledge expansion, which is realized based on any of the above methods, and the system comprises:

[0041] A retrieval module for retrieving information according to the selected keywords, obtaining retrieval results; when the number of cycles reaches a set value, the retrieval is completed; when executed for the first time, the selected keywords are user input keywords;

[0042] The keyword extraction module is used for mining analysis on the search result to extract keywords.

[0043] The keyword analysis module is used for analyzing the relationship between the extracted keywords and the selected keywords and the relationship between the extracted keywords, and dividing the extracted keywords into multiple groups according to different relationship types.

[0044] The keyword relationship validity verification module is used for constructing different new query statements according to different groups, verifying the validity of the keyword relationship, and putting the verified keywords into the keyword library.

[0045] Compared with the prior art, the advantages of the present application are that:

[0046] 1. The data framing retrieval method based on knowledge expansion improves the problem of insufficient understanding and description of the user's retrieval intention in the case of less information input (sparse) in the traditional retrieval, and can greatly improve the data breadth returned by the retrieval, better adapt to the data retrieval needs in the case of less information, random input, fuzzy expression and other realistic input possibilities, and meet the diversified needs of information accurate retrieval and intelligence information deep discovery.

[0047] 2. The present application can retrieve related information as comprehensively and accurately as possible, and to some extent, solve the problem of preventing retrieval drift and clear decision boundary, and has high practical value.

[0048] 3. The data framing retrieval method based on knowledge expansion can also be applied to explore the research status or progress of a certain technology, and return data related to the technology from the largest possible range. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 The figure shows a data framing retrieval method based on knowledge expansion. DETAILED DESCRIPTION

[0050] First, the key technical terms involved in the present application are explained:

[0051] Automatic keyword extraction: aims to automatically extract one or more representative words from a given document, which can effectively retrieve the document, and through these keywords, the current document can be distinguished from other documents in terms of meaning. Automatic keyword extraction generally includes three steps: text preprocessing, candidate word selection, and keyword evaluation and determination.

[0052] Query automatic expansion: refers to the automatic expansion of query keywords, mainly used to solve the problem that the query keywords input by the user are difficult to meet the needs of high-quality queries in terms of accuracy and effectiveness due to the limitations of query experience and knowledge reserves, and the difference between the query keywords and the system index, so as to more accurately understand the user's intention.

[0053] Text similarity calculation: refers to the calculation of the semantic similarity between texts, also known as text matching, which compares the similarity between two or more texts through a certain method to obtain a quantitative similarity value.

[0054] The technical solutions of the present application will be described in detail below with reference to the accompanying drawings.

[0055] The present application proposes a data frame setting retrieval method and system based on knowledge expansion, which returns information that meets the user's retrieval intention as comprehensively and accurately as possible, and controls noise.

[0056] The method comprises:

[0057] Step 1) According to the given and limited user input keywords (input), retrieve information, and obtain original feedback data;

[0058] Step 2) Mine the feedback data, and use various statistical methods such as TF-IDF, TextRank, etc. Keyword extraction algorithm to automatically extract keywords, so that they can represent the input category as comprehensively as possible and be distinguished from the keywords of other category data;

[0059] According to the initial result text retrieved according to the user given keywords, extract the keywords so that they can fully represent the retrieved category;

[0060] The specific method of extracting keywords is as follows: according to the initial data returned by the initial keyword retrieval, complete the initial preprocessing of the data, including stop word removal, punctuation normalization, and digit standardization operations.

[0061] In the returned retrieval data, the number of times a word appears in a document is counted as the document frequency df1 of the word, which is used to estimate the document distribution of the word in the relevant article;

[0062] A certain amount of data (n, recommended n>20000) is randomly sampled from the searched data set, and the document frequency df2 of the word in the returned data is counted, which is used to estimate the document distribution of the word;

[0063] Combine TF-IDF, TextRank indicators (formally represented as tf_score and tr_score) to establish a keyword evaluation system, and design a comprehensive scoring function as follows:

[0064]

[0065] Wherein, tf_score represents the importance degree of the keyword in the initial keyword search returned document and the randomly sampled data document set calculated using the TF-IDF algorithm, and tr_score represents the importance degree of the keyword in the initial keyword search returned document analyzed by the TextRank algorithm. W1 and W2 are weights for weighted summation of the two indicators, and W1+W2=1.

[0066] In the scoring function, the quotient of df1 and df2 is logarithmized to represent the degree of differentiation of the keyword in the initial keyword search returned document from other categories; the weighted summation of the normalized tf_score and tr_score represents the importance degree of the keyword in the initial keyword search returned document and the randomly sampled data document, and the tr_score represents the importance degree of the keyword in the initial keyword search returned document analyzed by the TextRank algorithm.

[0067] The score calculated by the above scoring function represents the importance evaluation of the keyword in the initial data that is different from the remaining non-returned documents, and the keyword with a higher score is selected as the keyword mining result.

[0068] Step 3) The automatically extracted keywords are combined and subdivided into four relationships of limitation, mutual exclusion, co-occurrence and synonymy, and different query statements are constructed with different relationships, so that the new query statement contains more refined query information, thereby improving the keyword query efficiency.

[0069] The keywords also have certain relationships, which are subdivided into four relationships of limitation, mutual exclusion, co-occurrence and synonymy, and different relationships have different effects in the query.

[0070] Synonyms are an effective bridge for improving the recall rate of the system in the keyword expansion process, greatly weakening the problem of missing effective information in the traditional keyword matching, and breaking through the query barrier between synonyms in the keyword query process. Starting from the relationship with the original keyword, the synonym relationship can be found, and such keywords have similar semantics with the original keyword and can replace the original keyword for retrieval in the query, so as to expand the retrieval range as much as possible under the premise of meeting the user's retrieval demand. The present application uses deep text representation, uses the pre-trained glove model based on wiki encyclopedia to obtain the vector representation of the keyword, calculates the semantic similarity between the original keyword and the mined keyword by using the cosine similarity, and screens the synonyms of the original keyword with a certain threshold. In the next iteration process, the original keyword is replaced by the synonym to construct a new query to expand the query range.

[0071] The words of the remaining three relations, which represent the keywords that cannot be replaced by the original keywords as synonyms, are called candidate words, and the original keywords are used as restrictions to perform a more extensive query. Based on the conditional probability between the candidate words in the initial returned documents, the candidate words are combined into a query statement, which has a crucial influence on the final search results. The candidate words of different relations are divided into three relations, and the specific methods are as follows:

[0072] The conditional probabilities P(A|B) and P(B|A) of two words (or word groups) A and B in the original feedback documents are calculated respectively, and the three types of relations are formally represented as:

[0073] A. restrictive relation:

[0074] P(A|B)<p t P(B|A)>p′ t

[0075] Wherein, p t and p' t represent the threshold value of conditional probability, the value range is (0, 1), and in actual use, p t =0.3 and p' t =0.7, the greater the difference between the two values, the more accurate the relation judgment.

[0076] The above formula shows that A∈B, which belongs to the restrictive relation, indicating that a word is more likely to appear in the text where another word appears, and the latter has a certain degree of restriction on the former.

[0077] In the set of restrictive relations, the restrictive relation is used as a directed edge, and the keyword is used as a node to form a directed acyclic graph, which leads to the occurrence of multiple edges between two nodes, that is, the transmission of the restrictive relation and the multiple restrictive forms of the restrictive relation. For this, the present application selects the longest edge to be retained.

[0078] B. mutual exclusive relation:

[0079] P(A|B)<p t P(B|A)<p' t

[0080] The mutual exclusive relation indicates that two words are almost impossible to appear at the same time, and the two words are probably independent.

[0081] C. co-occurrence relation:

[0082] P(A|B)>p t P(B|A)>p′ t

[0083] The co-occurrence relation indicates that the possibility of the simultaneous appearance of two candidate words is high, and there are two possible situations:

[0084] (1) The combination of A and B is a new keyword group,

[0085] (2) Both of them appear as features of the keyword category.

[0086] Step 4) Effectiveness verification of keyword relationship by combining KL divergence, filtering keyword relationship by certain strategy, and expanding to keyword library in different relationship categories.

[0087] The effectiveness verification of the above three relationships is as follows:

[0088] Using the above three types of relationships, the original keyword or synonym is used as a limiting condition to construct a new query statement, the data corresponding to the constructed query statement is calculated, the probability distribution is calculated, the KL divergence of the probability distribution of the initial data is calculated, and the distance between the two data distributions is calculated to estimate the effectiveness of the expanded keyword. The specific method is as follows:

[0089] The returned data of the newly constructed query is formalized as a probability distribution P(x), that is, the frequency of all words after segmentation is counted as the probability distribution of the returned data;

[0090] The initial data returned by the original keyword search is formalized as a probability distribution Q(x);

[0091] The KL divergence of the two distributions is calculated:

[0092]

[0093] Where x represents the keyword, and in the two distributions P(x) and Q(x), the values of the same keyword in P(x) and Q(x) are calculated when calculating the KL divergence. All keywords are operated, and then the sum is calculated to obtain the value of the KL divergence.

[0094] According to the KL divergence, the distance between the two distributions is calculated, the difference between the corresponding sample distributions of the two groups of data is obtained, and the deviation of the expanded keyword is evaluated. Certain strategy is used to filter keywords, and the present application sets a certain threshold value, considers the relationship with a KL divergence less than the threshold value as effective, and is retained and expanded to the keyword library for the next iteration, so as to obtain a wider range of related search results.

[0095] Generally, the above four steps are expanded by three to four iterations under the condition of less (very little) input information, expand the keyword library, and expand the search breadth, which can maximize the return of comprehensive and accurate search information.

[0096] The present application also provides a data frame retrieval system based on knowledge expansion, which is realized based on any of the above methods, and the system comprises:

[0097] The retrieval module is used for retrieving information according to the selected keyword, obtaining a retrieval result, completing the retrieval when the number of cycles reaches a set value, and selecting the keyword as a user input keyword when the retrieval is performed for the first time.

[0098] The keyword extraction module is used for mining and analyzing the retrieval result, and extracting keywords.

[0099] The keyword analysis module is used for analyzing the relationship between the extracted keywords and the selected keyword and the relationship between the extracted keywords, and classifying the extracted keywords into multiple groups according to different relationship types.

[0100] The keyword relationship validity verification module is used for constructing different new query statements according to different groups, verifying the validity of the keyword relationship, and putting the keywords verified as valid into a keyword library.

[0101] The application further provides a computer device, which comprises at least one processor, a memory, at least one network interface and a user interface. The various components in the device are coupled together through a bus system. It can be understood that the bus system is used to realize the connection communication between the components. In addition to the data bus, the bus system further comprises a power supply bus, a control bus and a state signal bus.

[0102] The user interface can comprise a display, a keyboard or a clicking device, for example, a mouse, a trackball, a touchpad or a touch screen.

[0103] It can be appreciated that the memory in the embodiments disclosed in the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (Read-Only Memory, ROM), a programmable read-only memory (Programmable ROM, PROM), an erasable programmable read-only memory (Erasable PROM, EPROM), an electrically erasable programmable read-only memory (Electrically EPROM, EEPROM) or a flash memory. The volatile memory can be a random access memory (Random Access Memory, RAM) used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (Static RAM, SRAM), dynamic random access memory (Dynamic RAM, DRAM), synchronous dynamic random access memory (Synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (Double Data Rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (Enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (Synchlink DRAM, SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DRRAM). The memory described herein is intended to include but not limited to these and any other suitable types of memory.

[0104] In some embodiments, the memory stores elements, executable modules or data structures, or a subset thereof, or an extended set thereof: an operating system and an application program.

[0105] Among them, the operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application program includes various application programs, such as a media player (Media Player), a browser (Browser), etc., for implementing various application services. The program for implementing the method of the embodiments of the present disclosure can be included in the application program.

[0106] In the above-described embodiments, the processor can also be used to execute the steps of the above-described method by invoking the programs or instructions stored in the memory, specifically, the programs or instructions stored in the application program.

[0107] execute the steps of the above-described method.

[0108] The method can be applied to a processor or implemented by the processor. The processor can be an integrated circuit chip having a signal processing capability. In implementation, the steps of the method can be completed by an integrated logic circuit of hardware in the processor or by an instruction in the form of software. The processor can be a general-purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The methods disclosed above can be implemented or executed by the processor. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed above can be directly embodied as a hardware code executed by the processor or a combination of hardware and software modules in the processor. The software module can be located in a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, or other mature storage media in the art. The storage media is located in the storage memory, and the processor reads information in the storage memory and combines the hardware to complete the steps of the method.

[0109] It can be understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or any combination thereof. For a hardware implementation, the processing units can be implemented within one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSP Devices), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, micro-controllers, microprocessors, other electronic units designed to perform the functions described herein, or a combination thereof.

[0110] For a software implementation, the techniques can be implemented with modules (e.g., procedures, functions, and so on) that perform the functions described herein. The software codes can be stored in memory and executed by processors. The memory can be implemented within the processor or external to the processor.

[0111] The application further provides a nonvolatile storage medium for storing the computer program. When the computer program is executed by a processor, each step in the above method embodiment can be implemented.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A data frame retrieval method based on knowledge expansion, the method comprising: Step 1: Search based on the selected keywords and obtain the search results; when the number of iterations reaches the set value, the search is complete; otherwise, continue to Step 2. When performing step 1 for the first time, the selected keyword is the keyword entered by the user; Step 2: Mining and analyzing the search results to extract keywords; Step 3: Analyze the relationship between the keywords extracted in Step 2 and the keywords selected in Step 1, as well as the relationship between the extracted keywords. Divide the extracted keywords into multiple groups according to different relationship types. Step 4: Construct different new query statements based on different groups, verify the validity of keyword relationships, and add the verified valid keywords to the keyword library; Select keywords from the keyword library, increment the loop count by 1, and return to step 1; Step 2 specifically includes: Initial preprocessing of the search results includes stop word removal, punctuation normalization, and number standardization; In the preprocessed search results, the keywords are first obtained by word segmentation, and then the number of times the keywords appear in the document is counted as the document frequency (df1) of the keywords. Randomly sample n data points from the retrieved dataset and count the document frequency (df2) of the keywords in the n data points; The scoring function is defined as follows: ； Where tf_score represents the importance of keywords in the combination of search results and random sampled data calculated using the TF-IDF algorithm; tr_score represents the importance of keywords in the keyword search results derived from data analysis using the TextRank algorithm; and These represent the weights corresponding to the two levels of importance. ; Keywords whose scores calculated by the above scoring function are higher than the set value are selected as the extracted keywords.

2. The data frame retrieval method based on knowledge expansion according to claim 1, characterized in that, The relationship between the extracted keywords and the selected keywords includes synonyms and other relationships; The other relationships include: limiting, mutual exclusion, and co-occurrence relationships.

3. The data frame retrieval method based on knowledge expansion according to claim 2, characterized in that, The methods for determining the limiting, mutually exclusive, and co-occurrence relationships are as follows: Calculate the conditional probabilities of keywords A and B in the retrieved data. ; Keywords A and B are binding relations when the following relationship is satisfied: ； in, and These represent the threshold values ​​for their respective conditional probabilities, with a range of (0,1). Keywords A and B are mutually exclusive when the following condition is met: ； Keywords A and B are co-occurring when the following condition is met: 。 4. The data frame retrieval method based on knowledge expansion according to claim 3, characterized in that: 。 5. The data frame retrieval method based on knowledge expansion according to claim 3, characterized in that, In the set of keywords with limiting relationships, a directed acyclic graph is formed with the limiting relationships as directed edges and the keywords as nodes. If there are multiple edges between two nodes, only the limiting relationship with the longest edge is retained.

6. The data frame retrieval method based on knowledge expansion according to claim 2, characterized in that, The validity of keyword relationships is verified, and the specific process is as follows: For the three types of keywords with limiting, mutually exclusive, and co-occurrence relationships, new query statements are constructed based on the relationship between the keywords, using the original keywords or synonyms as limiting conditions respectively. The newly constructed query return data is formalized into a probability distribution. P(x) That is, after word segmentation, the frequency of all words is counted, and this is used as the probability distribution of the returned data. P(x) ; The search results from step 1 are formalized into a probability distribution. Q(x) ; Calculate the probability distributions of the two probability distributions KL divergence : ； in, x Indicates keywords; Will KL A relationship whose divergence is less than a set threshold is considered valid.

7. A data frame retrieval system based on knowledge expansion, implemented according to the method described in any one of claims 1-6, characterized in that, The system includes: The retrieval module is used to retrieve information based on selected keywords and obtain retrieval results; the retrieval is completed when the number of iterations reaches a set value; the selected keywords are those entered by the user on the first execution. The keyword extraction module is used to mine and analyze search results and extract keywords. The keyword analysis module is used to analyze the relationship between extracted keywords and selected keywords, as well as the relationships between extracted keywords themselves, and to categorize the extracted keywords into multiple groups based on different relationship types; and The keyword relationship validity verification module is used to construct different new query statements based on different groups, verify the validity of keyword relationships, and add the verified valid keywords to the keyword library.

Images