Keyword matching method and device based on natural language processing

Abstract

The invention discloses a keyword matching method and device based on natural language processing, and relates to the technical field of natural language data processing. The method comprises the following steps: acquiring an input deviation degree parameter of an input text, and analyzing whether text input adjustment is performed or not according to the input deviation degree parameter; after the text input is adjusted to be qualified, keyword extraction parameters of keyword extraction are obtained, and whether keyword extraction is conducted again or not is analyzed according to the keyword extraction parameters; and after the keyword extraction is qualified, obtaining matching coincidence parameters of the matched text, and analyzing whether to perform matching again according to the matching coincidence parameters. Text input adjustment is performed by inputting the deviation degree evaluation coefficient, whether keyword extraction is performed again or not is analyzed according to the extracted keyword coincidence degree coefficient, whether matching is performed again or not is analyzed according to the matching coincidence degree evaluation coefficient, the effect of accurately performing text matching is achieved, and the user experience is improved. The problem of inaccurate keyword matching in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language data processing, and in particular to a keyword matching method and device based on natural language processing. Background Art

[0002] In the field of Natural Language Processing (NLP), keyword matching is a basic and important task. It involves extracting keywords or phrases related to a specific query or topic from a large amount of text data, thereby helping users quickly locate the required information. Keyword matching technology is widely used in many fields such as information retrieval, text classification, entity recognition, and question-answering systems. In question-answering systems, keyword matching can match key information in questions and answers, thereby improving the accuracy of the system's answers.

[0003] Natural language sentence matching technology is an important branch of natural language processing, which aims to evaluate the semantic similarity or equivalence of two or more sentences. With the rapid development of the Internet and social media, massive amounts of text data are constantly being generated and shared, which contains a large amount of repeated or similar information. Sentence matching technology can be applied to question-answering systems to help the system understand whether the questions asked by users are repeated with existing questions, so as to provide more accurate and efficient answers. Therefore, a more accurate keyword matching method is needed, combined with sentence matching technology to more accurately realize the text matching function of the question-answering system.

[0004] Existing keyword matching systems use natural language processing technology to perform semantic analysis on query terms and target documents or data, thereby finding results with similar or related meanings and realizing the keyword matching function.

[0005] For example, the invention patent announcement with announcement number: CN113836942B discloses a text matching method based on latent keywords, including: S1, inputting x and x', using word vectors to convert x and y into a series of vectors ex=E(x) and ey=E(x'); S2, inputting the word vector e into the sentence encoder to obtain a vector h containing context information; S3, calculating the probability that each word belongs to a keyword based on h and the probability formula; S4, calculating the keyword vector of the text; S5, obtaining the final vector of the text, and calculating the matching degree of the two texts by the cosine angle.

[0006] For example, the invention patent with announcement number CN115114932B announces a multi-granularity Chinese short text matching method based on keywords, including: S1: dividing sentences into two granularities, character and word, uniformly filling the sentences of the two granularities to a length of N, training Word2Vec on the corresponding data set, and obtaining the embedding representation of the two granularities of character and word; S2: encoding the sentence vector with two BiGRUs to obtain the context information of the sentence in two directions; S3: using cross attention to obtain the association between the character granularity features and the word granularity features, and then performing reverse average pooling and connecting them to obtain the final representation vector of the sentence; S4: connecting the word granularity embedding vectors of the two sentences, and encoding them with an 11-layer Transformer encoder and a layer of Transformer focusing on keywords; S5: connecting the keyword features and the representation vectors of the two sentences as the final prediction vector.

[0007] However, in the process of implementing the technical solution of the invention in the embodiments of the present application, the present application found that the above technology has at least the following technical problems:

[0008] In the prior art, there is a problem of inaccurate keyword matching due to inaccurate keyword matching of the question-answering system, which in turn affects the text matching function of the question-answering system. Summary of the invention

[0009] The embodiments of the present application solve the problem of inaccurate keyword matching in the prior art by providing a keyword matching method and device based on natural language processing, thereby achieving more accurate keyword matching.

[0010] An embodiment of the present application provides a keyword matching method based on natural language processing, comprising the following steps: obtaining an input deviation degree parameter of an input text, and analyzing whether to adjust the text input according to the input deviation degree parameter; obtaining an extraction keyword parameter for extracting keywords after the text input adjustment is qualified, and analyzing whether to re-extract keywords according to the extraction keyword parameter; obtaining a matching parameter of a matching text after the keyword extraction is qualified, and analyzing whether to re-match according to the matching parameter.

[0011] Furthermore, the input deviation degree parameters include: the number of spelling errors, the number of grammatical errors, the length of the input text, the text input time, the preset number of spelling errors, the preset number of grammatical errors, the preset input text length and the preset text input time; the input deviation degree parameters are used to analyze the input deviation degree evaluation coefficient; the input deviation degree evaluation coefficient is used to reflect the input deviation degree of the input text.

[0012] Furthermore, the specific process of obtaining the input deviation degree assessment coefficient is as follows: performing a ratio operation on the number of spelling errors and a preset number of spelling errors to obtain a spelling error coefficient; performing a ratio operation on the sum of the number of grammatical errors and a preset number of grammatical errors and twice the preset number of grammatical errors to obtain a grammatical error coefficient; performing a processing on the preset input text length and the input text length to obtain an input text length coefficient; performing a processing on the result of a ratio operation on the text input time and the preset text input time to obtain a text input time coefficient; when the input text length coefficient is less than or equal to 1, performing processing based on the spelling error coefficient, the grammatical error coefficient and the input text length coefficient to obtain an input deviation degree assessment coefficient; when the input text length coefficient is greater than 1, performing processing based on the input text length influencing factor, the spelling error coefficient, the grammatical error coefficient and the text input time coefficient to obtain an input deviation degree assessment coefficient.

[0013] Furthermore, the specific process of analyzing whether to perform text input adjustment is: obtaining a preset input deviation degree threshold from a database, and comparing the obtained input deviation degree assessment coefficient with the preset input deviation degree threshold obtained from the database; when the input deviation degree assessment coefficient is less than the preset input deviation degree threshold obtained from the database, no text input adjustment is performed; when the input deviation degree assessment coefficient is not less than the preset input deviation degree threshold obtained from the database, text input adjustment is performed, and the text input adjustment includes input error reminders and re-entry reminders, and when the monitored input deviation degree assessment coefficient is less than the input deviation degree threshold, the text input adjustment is stopped.

[0014] Furthermore, the keyword extraction parameters include: input deviation degree assessment coefficient, number of extracted keywords, number of specific words in the input text, number of synonymous words, extraction response time, standard number of extracted keywords, standard number of specific words in the input text, standard number of synonymous words and standard extraction response time; the keyword extraction parameters are used to analyze and obtain the extraction keyword compliance coefficient; the keyword extraction compliance coefficient is used to reflect the compliance degree of the extracted keywords.

[0015] Furthermore, the specific acquisition process of the coefficient of degree of conformity of extracted keywords is as follows: performing a ratio operation on the number of extracted keywords and the standard number of extracted keywords to obtain an extraction coefficient; performing a ratio operation on twice the standard number of specific words in the input text and the sum of the number of specific words in the input text and the standard number of specific words in the input text to obtain a specific coefficient; processing the result of the ratio operation on the standard number of synonymous words and the number of synonymous words to obtain a synonym coefficient; performing a ratio operation on the standard time length for extraction response and the time length for extraction response to obtain a response coefficient; and processing the input deviation degree assessment coefficient, the extraction coefficient, the specific coefficient, the synonym coefficient and the response coefficient to obtain a coefficient of degree of conformity of extracted keywords.

[0016] Furthermore, the specific analysis process of analyzing whether to re-extract keywords is as follows: obtaining a preset keyword extraction threshold from a database, and comparing the obtained keyword extraction degree coefficient with the preset keyword extraction threshold obtained from the database; when the keyword extraction degree coefficient is greater than the preset keyword extraction threshold obtained from the database, it indicates that the keyword extraction is successful; when the keyword extraction degree coefficient is not greater than the preset keyword extraction threshold obtained from the database, the keywords are re-extracted, and when the monitored keyword extraction degree coefficient is greater than the preset keyword extraction threshold obtained from the database, the re-extraction of keywords is stopped.

[0017] Furthermore, the matching parameters include: input deviation degree evaluation coefficient, keyword extraction degree coefficient, matching text cosine similarity, matching text second similarity, matching time, input deviation degree threshold, keyword extraction threshold, matching text standard cosine similarity, matching text standard second similarity and matching standard time; the matching parameters are used to analyze and obtain the matching degree evaluation coefficient; the matching degree evaluation coefficient is used to reflect the degree of matching of the matching text; the specific process of analyzing whether to re-match is: obtain a preset matching threshold from the database, and compare the obtained matching degree evaluation coefficient with the preset matching threshold obtained from the database; when the matching degree evaluation coefficient is greater than the matching threshold, it indicates that the text match is successful; when the matching degree evaluation coefficient is not greater than the matching threshold, re-matching is performed, and when the monitored matching degree evaluation coefficient is greater than the matching threshold, matching is stopped.

[0018] Furthermore, the limiting expression of the input deviation degree evaluation coefficient is:

[0019]

[0020] Where AT n It is represented as the input deviation evaluation coefficient of the nth text input, n=1,2,...,N, n represents the number of times the user inputs text, and N represents the total number of times the user inputs text. AAP n It is expressed as the spelling error coefficient of the nth text input, AAY n It is expressed as the grammatical error coefficient of the nth text input, AAC n Represents the input text length coefficient for the nth text input, AAD n It is represented as the text input time coefficient at the nth text input, β is represented as the input text length influence factor, and e is represented as a natural constant.

[0021] An embodiment of the present application provides a keyword matching device based on natural language processing, including: a text input adjustment analysis module, a keyword extraction analysis module and a matching analysis module: wherein the text input adjustment analysis module is used to obtain an input deviation degree parameter of an input text, and analyze whether to perform text input adjustment according to the input deviation degree parameter; the keyword extraction analysis module is used to obtain an extraction keyword parameter of an extracted keyword after the text input adjustment is qualified, and analyze whether to re-extract the keyword according to the extraction keyword parameter; the matching analysis module is used to obtain a matching compliance parameter of a matching text after the keyword extraction is qualified, and analyze whether to re-match according to the matching compliance parameter.

[0022] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0023] 1. By obtaining the input deviation degree parameter of the input text, analyzing whether to adjust the text input according to the input deviation degree parameter, obtaining the extracted keyword parameters of the extracted keywords after the text input adjustment is qualified, analyzing whether to re-extract keywords according to the extracted keyword parameters, and obtaining the matching parameters of the matching text after the keyword extraction is qualified, so as to analyze whether to re-match according to the matching parameters, thereby achieving accurate text matching and effectively solving the problem of inaccurate keyword matching in the prior art.

[0024] 2. After the text input is adjusted to be qualified, the keyword extraction parameters of the extracted keywords are obtained, and the keyword extraction conformity coefficient is obtained according to the keyword extraction parameters, so as to analyze whether to re-extract the keywords according to the keyword extraction conformity coefficient, thereby realizing accurate keyword extraction.

[0025] 3. After the keyword extraction is qualified, the matching parameters of the matching text are obtained, and the matching degree evaluation coefficient is obtained according to the matching parameters, so as to analyze whether to re-match according to the matching degree evaluation coefficient, thereby achieving accurate matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flowchart of a keyword matching method based on natural language processing provided in an embodiment of the present application;

[0027] Figure 2 A graph showing the degree of conformity of extracted keywords provided in an embodiment of the present application;

[0028] Figure 3 A structural diagram of a keyword matching device based on natural language processing provided in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The embodiments of the present application solve the problem of inaccurate keyword matching in the prior art by providing a keyword matching method and device based on natural language processing. The method obtains an input deviation degree parameter of the input text, analyzes whether to adjust the text input according to the input deviation degree parameter, obtains extracted keyword parameters for extracting keywords after the text input adjustment is qualified, analyzes whether to re-extract keywords according to the extracted keyword parameters, obtains matching parameters of the matching text after the keyword extraction is qualified, and analyzes whether to re-match according to the matching parameters, thereby achieving accurate text matching.

[0030] The technical solution in the embodiment of the present application is to solve the above-mentioned problem of inaccurate keyword matching, and the overall idea is as follows:

[0031] By obtaining the input deviation degree parameters of the input text, the input deviation degree evaluation coefficient is obtained according to the input deviation degree parameters, and whether to adjust the text input is analyzed according to the input deviation degree evaluation coefficient, the extraction keyword parameters of the extracted keywords are obtained after the text input adjustment is qualified, the extraction keyword compliance coefficient is obtained according to the extraction keyword parameters, and whether to re-extract keywords is analyzed according to the extraction keyword compliance coefficient, and after the keyword extraction is qualified, the matching compliance parameters of the matching text are obtained, so as to obtain the matching compliance evaluation coefficient according to the matching compliance parameters, and whether to re-match is analyzed according to the matching compliance evaluation coefficient, thereby achieving the effect of accurate text matching.

[0032] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0033] like Figure 1 As shown, it is a flow chart of the keyword matching method based on natural language processing provided in an embodiment of the present application. The method is applied to a keyword matching device based on natural language processing, and the method includes the following steps: obtaining an input deviation degree parameter of the input text, and analyzing whether to adjust the text input according to the input deviation degree parameter; obtaining an extraction keyword parameter of the extracted keyword after the text input adjustment is qualified, and analyzing whether to re-extract the keyword according to the extraction keyword parameter; obtaining a matching parameter of the matching text after the keyword extraction is qualified, and analyzing whether to re-match according to the matching parameter.

[0034] It should be added that the input deviation degree parameters include: the number of spelling errors, the number of grammatical errors, the length of the input text, the text input time, the preset number of spelling errors, the preset number of grammatical errors, the preset input text length and the preset text input time; the input deviation degree parameters are used to analyze the input deviation degree evaluation coefficient; the input deviation degree evaluation coefficient is used to reflect the input deviation degree of the input text.

[0035] In this embodiment, after the text is input into the question-and-answer system, the question-and-answer system analyzes whether there is a deviation in the input text. If there is, feedback is given to allow the user to re-input the text. After the text input is qualified, the question-and-answer system analyzes whether the keyword extraction is qualified. If it is not qualified, the keywords are re-extracted until it is qualified, and then the text is matched. After the text is matched, the system analyzes whether the match is qualified. If it is not qualified, the match is re-matched. If it is still unqualified after a certain number of repetitions, manual matching is performed.

[0036] In this embodiment, the number of spelling errors describes the total number of spelling errors in the input text; the number of grammatical errors describes the total number of grammatical errors in the input text; the input text length describes the total length of the input text; the text input time describes the time from the start of input to the completion of input; the preset number of spelling errors describes the preset maximum total number of spelling errors in the input text; the preset number of grammatical errors describes the preset maximum total number of grammatical errors in the input text; the preset input text length describes the preset minimum total length of the input text; the preset text input time describes the preset maximum time from the start of input to the completion of input.

[0037] For example, the input text is "Please help me find the latest developments in 'artificial intelligence' and its application in 'machine learning'. Also, I want to know about 'deep learning' and the relationship between them." Spelling errors: "Artificial intelligence" should be "artificial intelligence", "knowledge" here is a typo, and the correct one should be "intelligence"; "deep learning" should be "deep learning", "cross" here is a typo, and the correct one should be "degree". Grammatical errors: The sentence structure in the input text is basically correct, there are no obvious grammatical errors, each sentence expresses a complete meaning, and conforms to the grammatical rules of Chinese. Error summary: Spelling errors: 2, "artificial intelligence" and "deep learning"; Grammatical errors: 0, therefore, the number of spelling errors is 2, and the number of grammatical errors is 0.

[0038] By using an existing spelling check library (such as Pospel, Hunspel, etc.) to detect spelling errors in the input text, the number of spelling errors is obtained. The grammatical errors in the input text are checked by the syntax analysis function in the natural language processing library to obtain the number of grammatical errors. The number of characters in the input text is obtained by a text editor or word processing software to obtain the length of the input text. The timer is started when the user starts to input (for example, when the first character is detected to be input) and stopped when the user completes the input (for example, when the input cursor is detected to leave the input box or the user clicks the submit button). By calculating the difference between the end time of the timer and the start time, the duration of the input text is obtained, which is the text input duration.

[0039] The preset number of spelling errors is the maximum number of total spelling errors when the text length of the historical input text in the historical time period is the same as the correct text length of the input text and the input text is qualified. The preset number of grammatical errors is the maximum number of total grammatical errors when the text length of the historical input text in the historical time period is the same as the correct text length of the input text and the input text is qualified. The preset input text length is the minimum value of the text length when the text length of the historical input text in the historical time period is the same as the correct text length of the input text and the input text is qualified. The preset text input time is the maximum value of the input time when the text length of the historical input text in the historical time period is the same as the correct text length of the input text and the input text is qualified.

[0040] By analyzing the number of spelling errors, the number of grammatical errors, the input text length and the text input time during text input during the question and answer process with the preset number of spelling errors, the preset number of grammatical errors, the preset input text length and the preset text input time respectively, the accuracy of the text input is clarified in more detail.

[0041] Furthermore, the specific process of obtaining the input deviation evaluation coefficient is as follows: the number of spelling errors is calculated by ratioing the number of spelling errors to the preset number of spelling errors to obtain the spelling error coefficient (i.e., AAP n ); the sum of the number of grammatical errors and the preset number of grammatical errors is calculated by ratio with twice the preset number of grammatical errors to obtain the grammatical error coefficient (i.e. AAY n ); Process the preset input text length and the input text length to obtain the input text length coefficient (i.e. AAC n ); the result of the ratio operation of the text input time and the preset text input time is processed to obtain the text input time coefficient (i.e. AAD n ); when the input text length coefficient is less than or equal to 1, the input deviation degree evaluation coefficient is obtained according to the spelling error coefficient, grammatical error coefficient and input text length coefficient; when the input text length coefficient is greater than 1, the input deviation degree evaluation coefficient is obtained according to the input text length influencing factor, spelling error coefficient, grammatical error coefficient and text input time coefficient.

[0042] Among them, the number of text inputs is obtained and numbered, n = 1, 2, ..., N, n represents the number of text inputs by the user, N represents the total number of text inputs by the user, and the limiting expression of the input deviation degree evaluation coefficient is:

[0043]

[0044] Where AT n It is expressed as the input deviation evaluation coefficient for the nth text input, SP nRepresents the number of spelling errors in the nth text input, SY n Represents the number of grammatical errors in the nth text input, SC n Indicates the length of the input text at the nth text input, SD n represents the text input time for the nth text input, XP represents the preset number of spelling errors, XY represents the preset number of grammatical errors, XC represents the preset input text length, XD represents the preset text input time, AAP represents the preset number of grammatical errors, n It is expressed as the spelling error coefficient of the nth text input, AAY n It is expressed as the grammatical error coefficient of the nth text input, AAC n Represents the input text length coefficient for the nth text input, AAD n It is represented as the text input time coefficient at the nth text input, β is represented as the input text length influence factor, and e is represented as a natural constant.

[0045] In this embodiment, when the input text length is smaller, the text input time is smaller, and the influence of the text input time on the input deviation degree evaluation coefficient is smaller. Therefore, when the input text length coefficient is less than or equal to 1, the influence of the input text length coefficient on the input deviation degree evaluation coefficient is not considered. When the input text length is larger, the text input time is larger. At this time, the text input time has a greater influence on the input deviation degree evaluation coefficient than the input text length. Therefore, when the input text length coefficient is greater than 1, the input text length influence factor and the text input time are considered at the same time to influence the input deviation degree evaluation coefficient.

[0046] The preset input text length influence factor is obtained from the database according to the input text length, and the input text length influence factor indicates the influence of the input text length on the input deviation degree evaluation coefficient. For example, a mapping set is formed by forming the input text length and the input text length influence factor corresponding to the preset input text length in the database, wherein the mapping relationship can be one-to-one or many-to-one, and the current input text length is input into the corresponding mapping set to obtain the input text length influence factor corresponding to the input text length, and the value range of the input text length influence factor is (0,1).

[0047] The algorithm of this embodiment combines the input deviation degree parameter for comprehensive analysis to obtain the input deviation degree evaluation coefficient. The input deviation degree parameter in the algorithm of this embodiment does not exist independently, but is interrelated. Generally, when the length of the input text is smaller, the fewer spelling errors and grammatical errors may occur, and the time required for input is shorter, resulting in a smaller number of spelling errors, a smaller number of grammatical errors, and a smaller text input time, and vice versa. However, when the number of spelling errors is larger, the number of grammatical errors is larger, the length of the input text is smaller, and the text input time is longer, the text input effect is worse. Therefore, when the number of spelling errors is larger, the number of grammatical errors is larger, the length of the input text is smaller, and the text input time is longer, the spelling error coefficient, the grammatical error coefficient, the input text length coefficient, and the text input time coefficient are all larger, which in turn leads to a larger input deviation degree evaluation coefficient. When the input deviation degree evaluation coefficient is larger, the input text error degree is larger, and the text input adjustment is more needed, and vice versa, the text input adjustment is less needed.

[0048] By comprehensively analyzing the number of spelling errors, the number of grammatical errors, the length of the input text and the length of time it takes to input text, an input deviation degree assessment coefficient is obtained, which makes the input deviation degree assessment coefficient more accurate and helps to more accurately adjust text input based on the input deviation degree assessment coefficient.

[0049] Furthermore, the specific process of analyzing whether to make text input adjustments is as follows: obtaining a preset input deviation degree threshold from the database, and comparing the obtained input deviation degree assessment coefficient with the preset input deviation degree threshold obtained from the database; when the input deviation degree assessment coefficient is less than the preset input deviation degree threshold obtained from the database, no text input adjustment is made; when the input deviation degree assessment coefficient is not less than the preset input deviation degree threshold obtained from the database, text input adjustment is made, and the text input adjustment includes input error reminders and re-entry reminders. When the monitored input deviation degree assessment coefficient is less than the input deviation degree threshold, text input adjustment is stopped. If the monitored input deviation degree assessment coefficient is not less than the input deviation degree threshold after a certain number of repetitions (set by the preset personnel, generally 3 times), text input is stopped, the question and answer system is restarted and check whether the system has been updated. If so, try to update the system.

[0050] In this embodiment, the larger the input deviation degree evaluation coefficient is, the greater the error degree of the input text is, and the more text input adjustment is needed; the smaller the input deviation degree evaluation coefficient is, the smaller the error degree of the input text is, and the less text input adjustment is needed.

[0051] The input error reminder specifically means that when the input deviation degree evaluation coefficient is not less than the preset input deviation degree threshold obtained from the database, the user is prompted that the text input is wrong. The re-entry reminder means that the user is reminded to re-enter the text after being prompted that the text input is wrong.

[0052] Furthermore, the keyword extraction parameters include: input deviation degree assessment coefficient, number of extracted keywords, number of specific words in the input text, number of synonymous words, extraction response time, standard number of extracted keywords, standard number of specific words in the input text, standard number of synonymous words and standard extraction response time; the keyword extraction parameters are used to analyze and obtain the extracted keyword compliance coefficient; the keyword extraction compliance coefficient is used to reflect the compliance degree of the extracted keywords.

[0053] In this embodiment, the number of extracted keywords indicates the total number of keywords extracted after the text input is adjusted to be qualified; the number of specific words in the input text indicates the total number of keywords with similar meanings among the keywords extracted after the text input is adjusted to be qualified; the number of synonymous words indicates the total number of words with similar meanings contained in the keywords extracted after the text input is adjusted to be qualified; the extraction response duration indicates the duration of extracting keywords after the text input is adjusted to be qualified; the standard number of extracted keywords indicates the minimum total number of keywords extracted after the preset text input is adjusted to be qualified; the standard number of specific words in the input text indicates the maximum total number of keywords with similar meanings contained in the keywords extracted after the preset text input is adjusted to be qualified; the standard number of synonymous words indicates the maximum total number of words with similar meanings contained in the keywords extracted after the preset text input is adjusted to be qualified; the standard extraction response duration indicates the maximum duration of extracting keywords after the preset text input is adjusted to be qualified.

[0054] The keywords in the input text are extracted by NLTK (Natural Language Toolkit) in natural language processing to obtain the number of extracted keywords. The semantic similarity between the keywords and other words is calculated by a semantic similarity calculation tool (such as Word2Vec, GloVe, GloVe, etc.) or an algorithm (such as BERT (Bidirectional Encoder Representations from Transformers, an algorithm based on a pre-trained language model)), and the keywords with similar meanings are determined and the number of keywords is counted to obtain the number of specific words in the input text. The semantic similarity between the keywords and other words is calculated by using a semantic similarity calculation tool or an algorithm, and the words with similar meanings to the keywords are determined and the number of words is counted to obtain the number of synonymous words. For example, the extracted keyword is "happy", and the words with similar meanings to "happy" are only "pleasant" and "happy", then the number of specific words in the input text is 1, and the number of synonymous words is 2. By calculating the timer, a start time point is recorded when keyword extraction begins, and an end time point is recorded after keyword extraction is completed, and the time difference between the two time points is calculated, that is, the extraction response duration.

[0055] The standard number of extracted keywords is the minimum value of the total number of extracted keywords when the above-mentioned historical input text successfully extracts keywords within the historical time period. The standard number of input text specific vocabulary is the maximum value of the total number of keywords with similar meanings when the above-mentioned historical input text successfully extracts keywords within the historical time period. The standard number of synonymous vocabulary is the maximum value of the total number of vocabulary with similar meanings to the keywords when the above-mentioned historical input text successfully extracts keywords within the historical time period. The standard extraction response time is the maximum value of the keyword extraction response time when the above-mentioned historical input text successfully extracts keywords within the historical time period.

[0056] Furthermore, the specific process of obtaining the extraction keyword compliance coefficient is as follows: the extraction coefficient (ie, TTG) is obtained by performing a ratio operation on the number of extracted keywords and the number of extracted keyword standards. b ); Calculate the ratio of twice the standard number of input text specific words to the sum of the number of input text specific words and the standard number of input text specific words to obtain the specific coefficient (i.e., TTH b ) ; The result of the ratio operation between the standard number of synonymous words and the number of synonymous words is processed to obtain the synonym coefficient (i.e. TTJ b ); the response coefficient (i.e., TTS) is obtained by calculating the ratio of the extraction response standard time to the extraction response time. b); according to the input deviation degree evaluation coefficient, extraction coefficient, specific coefficient, synonym coefficient and response coefficient, the extraction keyword matching degree coefficient is obtained.

[0057] Specifically, the number of keyword extraction times after the text input adjustment is qualified is obtained, and the keyword extraction times are numbered, b=1, 2, ..., B, b represents the number of keyword extraction times, and B represents the total number of keyword extraction times; the specific restriction expression of the keyword extraction compliance coefficient is:

[0058]

[0059]

[0060] Where, TN b It is expressed as the extraction keyword matching degree coefficient of the b-th extraction keyword, AT N It is expressed as the input deviation evaluation coefficient for the Nth text input, QG b It is expressed as the number of extracted keywords for the bth extraction of keywords, QH b It is represented as the number of specific words in the input text for the bth extraction of keywords, QJ b It is expressed as the number of synonyms of the keyword extracted for the bth time, QS b represents the extraction response time of the bth extraction of keywords, GG represents the standard number of extracted keywords, GH represents the standard number of specific words in the input text, GJ represents the standard number of synonymous words, GS represents the standard extraction response time, TTG b It is expressed as the extraction coefficient of the b-th keyword extraction, TTH b It is expressed as the specific coefficient of the b-th extracted keyword, TTJ b It is expressed as the synonym coefficient of the b-th extracted keyword, TTS b It is expressed as the response coefficient of the b-th extracted keyword.

[0061] The algorithm of this embodiment combines the extraction keyword parameters for comprehensive analysis to obtain the extraction keyword compliance coefficient. The extraction keyword parameters in the algorithm of this embodiment do not exist independently, but are interrelated. When the number of extracted keywords is larger, in general, the number of specific words in the input text corresponding to the extracted keywords, the number of synonymous words, and the extraction response time are larger. When the number of extracted keywords is larger and the number of specific words in the input text, the number of synonymous words, and the extraction response time are smaller, the effect of subsequent matching based on keywords is better. When the input deviation degree evaluation coefficient is smaller, that is, the more accurate the text input is, the better the effect of keyword extraction based on the input text. Therefore, when the input deviation degree evaluation coefficient is smaller, the number of extracted keywords is larger, and the number of specific words in the input text, the number of synonymous words, and the extraction response time are smaller, the extraction coefficient, specific coefficient, synonym coefficient, and response coefficient are larger, resulting in a larger extraction keyword compliance coefficient, and vice versa.

[0062] The extracted keyword conformity coefficient is obtained by comprehensively considering the input deviation degree evaluation coefficient, the number of extracted keywords, the number of specific words in the input text, the number of synonymous words and the extraction response time, so that the extracted keyword conformity coefficient is more accurate, which is conducive to more accurate re-extraction of keywords based on the extracted keyword conformity coefficient.

[0063] like Figure 2 As shown, it is a curve diagram of the extraction keyword matching degree coefficient provided by the embodiment of the present application, with the number of extracted keywords as the independent variable and the extraction keyword matching degree coefficient as the dependent variable. It is assumed that the input deviation degree evaluation coefficient AT of the Nth text input N is e-1, with the number of extracted keywords as the independent variable and the coefficient of the degree of conformity of the extracted keywords as the dependent variable. Through the increase and decrease analysis of curves 1, 2 and 3, it can be seen that the standard number of extracted keywords is set to 10, the standard number of specific words in the input text is 5, the standard number of synonymous words is 20, and the standard time for extraction response is 2 seconds. When the number of extracted keywords is larger, the coefficient of the degree of conformity of the extracted keywords is larger; when the extraction coefficient is smaller, the coefficient of the degree of conformity of the extracted keywords is smaller. The statistical table of the relevant data of the coefficient of conformity of the extracted keywords is shown in Table 1:

[0064] Table 1 Statistics of relevant data of extracted keyword matching coefficient

[0065]

[0066] When the number of extracted keywords of Curve 1, Curve 2 and Curve 3 is 10 and the extraction response time is 2 seconds, and the number of specific words and synonyms in the input text are as shown in the above table, the extracted keyword conformity coefficients of Curve 1, Curve 2 and Curve 3 are 2.51, 2.66 and 2.90 respectively. It can be seen that when the number of extracted keywords remains unchanged, the smaller the number of specific words and synonyms in the input text, the greater the extracted keyword conformity coefficient, and the less need there is to re-extract keywords. When it is analyzed based on the extracted keyword conformity coefficient that keyword extraction needs to be re-performed, keyword extraction is performed again in time to avoid using unqualified keywords for subsequent matching.

[0067] Furthermore, the specific analysis process for analyzing whether to re-extract keywords is as follows: obtain the preset keyword extraction threshold from the database, and compare the obtained keyword extraction degree coefficient with the preset keyword extraction threshold obtained from the database; when the keyword extraction degree coefficient is greater than the preset keyword extraction threshold obtained from the database, it indicates that the keyword extraction is successful; when the keyword extraction degree coefficient is not greater than the preset keyword extraction threshold obtained from the database, the keywords are re-extracted, and when the monitored keyword extraction degree coefficient is greater than the preset keyword extraction threshold obtained from the database, the re-extraction of keywords is stopped. When the monitored keyword extraction degree coefficient is not greater than the preset keyword extraction threshold obtained from the database after a certain number of repetitions (set by the preset personnel, generally 3 times), the question-answering system is checked for errors or loopholes and repaired in a timely manner.

[0068] In this embodiment, when the extracted keyword matching degree coefficient is larger, the keyword extraction effect is better, and the keyword extraction does not need to be repeated. When the extracted keyword matching degree coefficient is smaller, the keyword extraction effect is worse, and the keyword extraction needs to be repeated.

[0069] Further, the matching parameters include: input deviation degree evaluation coefficient, keyword extraction degree coefficient, matching text cosine similarity, matching text second shape similarity, matching time, input deviation degree threshold, keyword extraction threshold, matching text standard cosine similarity, matching text standard second shape similarity and matching standard time; matching parameters are used to analyze and obtain matching degree evaluation coefficient; matching degree evaluation coefficient is used to reflect the matching degree of matching text; the specific process of analyzing whether to re-match is: obtain a preset matching threshold from the database, compare the obtained matching degree evaluation coefficient with the preset matching threshold obtained from the database; when the matching degree evaluation coefficient is greater than the matching threshold, it means that the text is successfully matched; when the matching degree evaluation coefficient is not greater than the matching threshold, re-matching is performed, and when the monitored matching degree evaluation coefficient is greater than the matching threshold, matching is stopped. When the monitored matching degree evaluation coefficient is not greater than the matching threshold after a certain number of repetitions (set by the preset personnel, generally 5 times), the manual matching mechanism is introduced for matching.

[0070] In this embodiment, the matching text cosine similarity represents the cosine similarity between the input text and the matching text; the matching text second similarity represents the second similarity between the input text and the matching text; the matching duration represents the duration of the matching text; the matching text standard cosine similarity represents the preset minimum cosine similarity between the input text and the matching text; the matching text standard second similarity represents the preset minimum second similarity between the input text and the matching text; the matching standard duration represents the maximum duration of the matching text.

[0071] The cosine similarity between the input text and the matching text is obtained by the cosine similarity algorithm, which is the cosine similarity of the matching text. The Jaccard similarity between the input text and the matching text is obtained by the Jaccard similarity algorithm, which is the second similarity of the matching text. A computer timer is used to record a start time point when the matching starts, and an end time point is recorded after the matching is completed. The time difference between the two time points is calculated, which is the matching duration.

[0072] The minimum cosine similarity, the minimum second similarity and the maximum duration when the text matches are qualified are obtained from the database, namely, the matching text standard cosine similarity, the matching text standard second similarity and the matching standard duration.

[0073] Furthermore, the specific process of obtaining the matching degree evaluation coefficient is as follows: the cosine similarity coefficient (ie, YXX h); the sum of the second similarity of the matching text and the second similarity of the matching text standard is processed with the result of the ratio operation of 2 times the second similarity of the matching text standard to obtain the second similarity coefficient (ie, YKK h ); the result of the ratio operation between the matching standard duration and the matching duration is processed to obtain the matching duration coefficient (i.e. YZZ h );According to the input deviation degree evaluation coefficient, the extracted keyword matching degree coefficient, the cosine similarity coefficient, the second shape similarity coefficient and the matching time coefficient, the matching matching degree evaluation coefficient is obtained.

[0074] Specifically, the number of text matches after keyword extraction is qualified is obtained, and the text matches are numbered, h=1, 2, ..., H, where h represents the number of text matches and H represents the total number of text matches; the specific restriction expression of the matching compliance evaluation coefficient is:

[0075]

[0076] In the formula, YU h It is expressed as the matching degree evaluation coefficient of the hth text matching, AT N It is expressed as the input deviation evaluation coefficient of the Nth text input, TN B It is expressed as the coefficient of the degree of matching of the extracted keywords after the successful extraction of the keywords for the Bth time, RX h Represented as the cosine similarity of the matching texts for the hth text matching, RK h Represented as the second similarity of the matching text in the hth text match, RZ h represents the matching time of the hth text matching, ATT represents the input deviation threshold, TNN represents the keyword extraction threshold, BX represents the standard cosine similarity of the matching text, BK represents the standard second similarity of the matching text, BZ represents the standard matching time, and YXX represents the standard matching time. h Expressed as the cosine similarity coefficient of the hth text match, YKK h Expressed as the second similarity coefficient of the hth text match, YZZ h Expressed as the matching duration coefficient of the hth text match.

[0077] The algorithm of this embodiment combines the matching parameters for comprehensive analysis to obtain the matching conformity evaluation coefficient. The matching conformity parameters in the algorithm of this embodiment do not exist independently, but are interrelated. When the input deviation degree evaluation coefficient is smaller and the keyword extraction conformity coefficient is larger, the input text and the extracted keywords are more accurate, which is conducive to making the accuracy of text matching higher, and vice versa. The larger the cosine similarity of the matching text and the second shape similarity of the matching text, both reflect the similarity between the matching text and the input text. The larger the cosine similarity of the matching text, the larger the second shape similarity of the matching text. Conversely, when the cosine similarity of the matching text is larger and the second shape similarity of the matching text is larger, the similarity between the input text and the matching text is higher, and the matching is more accurate. When the matching time is shorter, it means that the matching efficiency is higher. Therefore, when the input deviation degree evaluation coefficient is smaller, the keyword extraction conformity coefficient is larger, the cosine similarity of the matching text is larger, the second shape similarity of the matching text is larger, and the matching time is smaller, the cosine similarity coefficient, the second shape similarity coefficient and the matching time coefficient are larger, which leads to a larger matching conformity evaluation coefficient, and vice versa.

[0078] The matching degree evaluation coefficient is obtained by comprehensively analyzing the input deviation degree evaluation coefficient, the extracted keyword matching degree coefficient, the matching text cosine similarity, the matching text second similarity and the matching time. The higher the accuracy of the matching degree evaluation coefficient, the more conducive it is to accurately analyze whether re-matching is needed.

[0079] When the matching degree evaluation coefficient is larger, the text matching effect is better, and the need for re-matching is less; when the matching degree evaluation coefficient is smaller, the text matching effect is worse, and the need for re-matching is more.

[0080] like Figure 3 As shown, it is a structural diagram of a keyword matching device based on natural language processing provided in an embodiment of the present application. The keyword matching device based on natural language processing provided in an embodiment of the present application includes: a text input adjustment analysis module, a keyword extraction analysis module and a matching analysis module: wherein, the text input adjustment analysis module is used to obtain an input deviation degree parameter of the input text, and analyze whether to adjust the text input according to the input deviation degree parameter; the keyword extraction analysis module is used to obtain an extraction keyword parameter of the extracted keyword after the text input adjustment is qualified, and analyze whether to re-extract the keyword according to the extraction keyword parameter; the matching analysis module is used to obtain a matching parameter of the matching text after the keyword extraction is qualified, and analyze whether to re-match according to the matching parameter.

[0081] In this embodiment, the algorithm obtains an input deviation degree evaluation coefficient through an input deviation degree parameter. The data in the input deviation degree parameter are not independent of each other, but influence each other. These data jointly influence the input deviation degree evaluation coefficient. The extracted keyword compliance coefficient is obtained by extracting the keyword parameters. The data in the extracted keyword parameters influence each other and jointly influence each other to obtain the extracted keyword compliance coefficient. The matching compliance parameter is obtained by matching the compliance parameter to obtain the matching compliance evaluation coefficient. The matching compliance evaluation coefficient is obtained by the mutual influence and joint influence of the data in the matching compliance parameter. The text input is adjusted according to the input deviation degree evaluation coefficient, and then whether to re-extract the keyword is analyzed according to the extracted keyword compliance coefficient. Finally, whether to re-match is analyzed according to the matching compliance evaluation coefficient. This achieves more accurate text matching, makes the question-answering process more accurate, and obtains more accurate answers.

[0082] The technical solution in the above-mentioned embodiment of the present application obtains an input deviation degree parameter of the input text, analyzes whether to adjust the text input according to the input deviation degree parameter, obtains an extraction keyword parameter for extracting keywords after the text input adjustment is qualified, analyzes whether to re-extract keywords according to the extraction keyword parameter, obtains matching parameters of the matching text after the keyword extraction is qualified, and then analyzes whether to re-match according to the matching parameters, thereby achieving accurate text matching and effectively solving the problem of inaccurate keyword matching in the prior art.

[0083] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0084] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0085] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0086] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0087] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0088] 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 keyword matching method based on natural language processing, characterized in that: The following steps are involved: Obtaining an input deviation degree parameter of the input text, and analyzing whether to adjust the text input according to the input deviation degree parameter; After the text input is adjusted to be qualified, extracting keyword parameters for extracting keywords are obtained, and whether to re-extract keywords is analyzed according to the extracting keyword parameters; After the keyword extraction is qualified, the matching parameters of the matching text are obtained, and whether to re-match is analyzed based on the matching parameters.

2. The keyword matching method based on natural language processing as claimed in claim 1, characterized in that: The input deviation degree parameters include: the number of spelling errors, the number of grammatical errors, the length of input text, the length of text input, the preset number of spelling errors, the preset number of grammatical errors, the preset length of input text and the preset length of text input; The input deviation degree parameter is used to analyze the input deviation degree evaluation coefficient; The input deviation degree evaluation coefficient is used to reflect the input deviation degree of the input text.

3. The keyword matching method based on natural language processing as claimed in claim 2, characterized in that: The specific process of obtaining the input deviation degree evaluation coefficient is as follows: The number of spelling errors is calculated by ratioing the number of spelling errors to a preset number of spelling errors to obtain a spelling error coefficient; The grammatical error coefficient is obtained by performing a ratio operation on the sum of the number of grammatical errors and the preset number of grammatical errors and twice the preset number of grammatical errors; The preset input text length and the input text length are processed to obtain the input text length coefficient; The result of the ratio operation of the text input time and the preset text input time is processed to obtain a text input time coefficient; When the input text length coefficient is less than or equal to 1, the input deviation degree evaluation coefficient is obtained by processing according to the spelling error coefficient, the grammatical error coefficient and the input text length coefficient; When the input text length coefficient is greater than 1, the input deviation degree assessment coefficient is obtained by processing the input text length influencing factor, spelling error coefficient, grammatical error coefficient and text input time coefficient.

4. The keyword matching method based on natural language processing as claimed in claim 1, characterized in that: The specific process of analyzing whether to adjust the text input is as follows: Obtaining a preset input deviation degree threshold from a database, and comparing the obtained input deviation degree evaluation coefficient with the preset input deviation degree threshold obtained from the database; When the input deviation degree evaluation coefficient is less than the preset input deviation degree threshold obtained from the database, no text input adjustment is performed; When the input deviation degree evaluation coefficient is not less than the preset input deviation degree threshold obtained from the database, text input adjustment is performed, and the text input adjustment includes input error reminder and re-entry reminder. When the monitored input deviation degree evaluation coefficient is less than the input deviation degree threshold, the text input adjustment is stopped.

5. The keyword matching method based on natural language processing as claimed in claim 1, characterized in that: The keyword extraction parameters include: input deviation degree evaluation coefficient, number of extracted keywords, number of input text specific words, number of synonymous words, extraction response time, standard number of extracted keywords, standard number of input text specific words, standard number of synonymous words and standard extraction response time; The extracted keyword parameters are used to analyze and obtain the extracted keyword matching degree coefficient; The extracted keyword matching degree coefficient is used to reflect the matching degree of the extracted keywords.

6. The keyword matching method based on natural language processing as claimed in claim 5, characterized in that: The specific process of obtaining the extracted keyword matching degree coefficient is as follows: The extraction coefficient is obtained by performing a ratio operation on the number of extracted keywords and the standard number of extracted keywords; The specific coefficient is obtained by performing a ratio operation on twice the standard number of specific words in the input text and the sum of the number of specific words in the input text and the standard number of specific words in the input text; The result of the ratio operation between the standard number of synonymous words and the number of synonymous words is processed to obtain the synonym coefficient; The response coefficient is obtained by performing a ratio operation between the extraction response standard time and the extraction response time; The extracted keyword compliance coefficient is obtained by processing the input deviation degree assessment coefficient, extraction coefficient, specific coefficient, synonym coefficient and response coefficient.

7. The keyword matching method based on natural language processing as claimed in claim 1, characterized in that: The specific analysis process of analyzing whether to re-extract keywords is as follows: Obtaining a preset keyword extraction threshold from a database, and comparing the obtained keyword extraction degree coefficient with the preset keyword extraction threshold obtained from the database; When the extracted keyword matching coefficient is greater than the preset keyword extraction threshold obtained from the database, it means that the keyword extraction is successful; When the extracted keyword matching coefficient is not greater than the preset keyword extraction threshold obtained from the database, the keyword is re-extracted; when the monitored extracted keyword matching coefficient is greater than the preset keyword extraction threshold obtained from the database, the re-extraction of the keyword is stopped.

8. The keyword matching method based on natural language processing as claimed in claim 1, characterized in that: The matching parameters include: input deviation evaluation coefficient, keyword extraction coefficient, matching text cosine similarity, matching text second shape similarity, matching time, input deviation threshold, keyword extraction threshold, matching text standard cosine similarity, matching text standard second shape similarity and matching standard time; The matching parameters are used to analyze and obtain a matching degree evaluation coefficient; The matching conformity evaluation coefficient is used to reflect the conformity of the matching text; The specific process of analyzing whether to re-match is as follows: Obtaining a preset matching threshold from a database, and comparing the obtained matching degree evaluation coefficient with the preset matching threshold obtained from the database; When the matching degree evaluation coefficient is greater than the matching threshold, it means that the text is matched successfully; When the matching conformity evaluation coefficient is not greater than the matching conformity threshold, the matching is performed again. When the monitored matching conformity evaluation coefficient is greater than the matching conformity threshold, the matching is stopped.

9. The keyword matching method based on natural language processing as claimed in claim 3, characterized in that: The limiting expression of the input deviation degree evaluation coefficient is: Where AT n It is represented as the input deviation evaluation coefficient of the nth text input, n=1,2,...,N, n represents the number of times the user inputs text, and N represents the total number of times the user inputs text. AAP n It is expressed as the spelling error coefficient of the nth text input, AAY n It is expressed as the grammatical error coefficient of the nth text input, AAC n Represents the input text length coefficient for the nth text input, AAD n It is represented as the text input time coefficient at the nth text input, β is represented as the input text length influence factor, and e is represented as a natural constant.

10. A keyword matching device based on natural language processing, characterized in that: include: Text input adjustment analysis module, keyword extraction analysis module and matching analysis module: The text input adjustment analysis module is used to obtain an input deviation degree parameter of the input text and analyze whether to perform text input adjustment according to the input deviation degree parameter; The keyword extraction and analysis module is used to obtain keyword extraction parameters for extracting keywords after the text input is adjusted to be qualified, and to analyze whether to re-extract keywords according to the keyword extraction parameters; The matching analysis module is used to obtain matching parameters of the matching text after the keyword extraction is qualified, and analyze whether to re-match according to the matching parameters.

Images