Text matching method and system and related product

Abstract

The embodiment of the invention discloses a text matching method and system and a related product. The method comprises the steps of obtaining a text label dictionary; the text label dictionary comprises text labels to which part of object full names respectively and correspondingly belong; identifying a corresponding target text label to which the to-be-matched text belongs in the text label dictionary; determining a label intersection degree between the target text label and the text label of each alternative full name in a full name library; based on the label intersection degree corresponding to each alternative full name, calculating a matching degree score between each alternative full name and the to-be-matched text; and taking a preset number of alternative full names with high matching degree scores as target object full names matched with the to-be-matched text. According to the embodiment of the invention, the text tag is introduced as a key point, so that the accuracy of fuzzy matching through the to-be-matched text can be improved, the detail information of the to-be-matched text searched by association is highly related to user requirements, and the user experience is improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of information processing technology, and in particular to text matching methods, systems and related products. Background Art

[0002] Data as a Service (DaaS) is a cloud computing service model that provides a cloud-based data management solution that enables users to access and share data over the Internet without having to store and process the data locally.

[0003] In DaaS applications, the server can provide external data subscription and data consumption. For example, in enterprise service (toB, Business to Business) scenarios such as enterprise information calibration, enterprise authentication access, automatic enterprise information supplementation, and enterprise portrait analysis, users can submit the enterprise name to the server to query or call the detailed information of the enterprise.

[0004] However, in reality, the company names submitted by users are most likely company abbreviations (including non-standard names), and company abbreviations are often agreed upon based on user habits and have no standard form. Therefore, the accuracy of the query results for fuzzy queries using company abbreviations is difficult to guarantee, and it is easy to match more company information results that are not highly relevant or even irrelevant. Summary of the invention

[0005] The embodiments of the present application provide a text matching method, system and related products for improving the matching accuracy of text to be matched.

[0006] A first aspect of an embodiment of the present application provides a text matching method, comprising:

[0007] Obtain a text label dictionary; the text label dictionary contains a portion of object full names corresponding to the corresponding text labels;

[0008] Identify the target text tag to which the to-be-matched text corresponds in the text tag dictionary;

[0009] Determining the label intersection between the target text label and the text label of each candidate full name; the candidate full name includes the object full name of the part;

[0010] Based on the label intersection degree corresponding to each of the candidate full names, calculating the matching score between each of the candidate full names and the text to be matched;

[0011] A preset number of the candidate full names with high matching scores are used as the full names of the target objects that match the text to be matched.

[0012] Optionally, obtaining a text tag dictionary includes:

[0013] Extracting a portion of the full names of the objects from the full name library as samples for word segmentation and word meaning analysis to summarize an initial dictionary of text labels; the initial dictionary of text labels is used to record the text labels corresponding to each of the samples under different label types;

[0014] For each of the tag types, the text tags that belong to the tag type in the historical period are collected, and the collected text tags are added to the text tag initial dictionary to obtain a text tag dictionary with more complete text tags.

[0015] Optionally, before determining the label intersection degree between the target text label and the text label of each candidate full name, the method further includes:

[0016] Performing word segmentation on the full names of other objects in the full name library except for the part of the full names of the objects, and performing word meaning analysis on the word segmentation results;

[0017] In combination with the word meaning analysis result, the text label dictionary is used to perform text label recognition on the word segmentation result to obtain the text label corresponding to the full name of the other object.

[0018] Optionally, the calculating the matching score between each candidate full name and the text to be matched based on the label intersection corresponding to each candidate full name includes:

[0019] Obtaining text similarity between the text to be matched and each of the candidate full names;

[0020] Based on the label intersection and the text similarity respectively corresponding to each of the candidate full names, a matching score between each of the candidate full names and the text to be matched is calculated.

[0021] Optionally, after calculating the matching score between each candidate full name and the text to be matched based on the label intersection corresponding to each candidate full name, the method further includes:

[0022] For a preset number of candidate full names with high matching scores, obtaining relevant index information of the preset number of candidate full names; wherein the preset number is greater than or equal to the preset number, and the relevant index information describes the object details of the candidate full names from a perspective different from that of the text labels;

[0023] The relevant index information is used as an influencing factor to adjust the matching score of each candidate full name to obtain a new matching score; the new matching score is used to select the full name of the target object.

[0024] Optionally, after taking a preset number of candidate full names with high matching scores as the full names of the target objects that match the text to be matched, the method further includes:

[0025] Selecting at least one of the candidate full names from the preset number of candidate full names;

[0026] The at least one selected candidate full name and / or the detailed information of the at least one selected candidate full name are returned to the input end of the text to be matched.

[0027] Optionally, when the candidate full name and the text to be matched are both company name description texts, the tag type to which the text tag belongs includes:

[0028] At least one of the region to which the enterprise belongs, the industry to which the enterprise belongs, the enterprise name keyword, and the enterprise name suffix.

[0029] When the method described in the first aspect of the present application is implemented, the content described in the second aspect of the present application can be adopted to achieve it.

[0030] A second aspect of an embodiment of the present application provides a text matching system, including: an acquisition unit and a processing unit;

[0031] The acquisition unit is used to acquire a text label dictionary; the text label dictionary contains a portion of object full names corresponding to the corresponding text labels;

[0032] The processing unit is used to identify the target text label to which the to-be-matched text corresponds in the text label dictionary;

[0033] The processing unit is further used to determine the label intersection degree between the target text label and the text label of each candidate full name; the candidate full name includes the part of the object full name;

[0034] The processing unit is further used to calculate a matching score between each of the candidate full names and the text to be matched based on the label intersection corresponding to each of the candidate full names;

[0035] The processing unit is further configured to use a preset number of candidate full names with high matching scores as target object full names that match the text to be matched.

[0036] A third aspect of an embodiment of the present application provides an electronic device, including: a processor and a memory;

[0037] The processor is configured to communicate with the memory and execute instructions in the memory to implement the method described in the first aspect of the embodiment of the present application or any specific implementation of the first aspect.

[0038] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, which stores computer instructions. When the computer instructions are executed by a processor, they implement the method described in the first aspect of the embodiment of the present application or any specific implementation of the first aspect.

[0039] A fifth aspect of an embodiment of the present application provides a computer program product, which includes computer instructions. When the computer instructions are executed by a processor, they implement the method described in the first aspect of the embodiment of the present application or any specific implementation of the first aspect.

[0040] It can be seen from the above technical solutions that the embodiments of the present application have at least the following advantages:

[0041] The embodiment of the present application uses the text label rather than the text itself as the entry point, determines the label intersection between the target text label and the text label of each candidate full name, and thus calculates the matching score between each candidate full name and the text to be matched. This can effectively avoid the problem that when the text to be matched input by the user is a fuzzy term in a non-standard format (such as an abbreviation), directly or only quoting the text to be matched itself to calculate the matching degree between texts will cause the matching result obtained through the text to be matched to be inaccurate. In short, the embodiment of the present application uses the text label as the key point, which can improve the accuracy of fuzzy matching through the text to be matched, and promote the detailed information of the text to be matched found by association to be highly relevant to user needs, thereby improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0043] It should be noted that, although the steps in the process diagrams (if any) involved in the embodiments are drawn in sequence as indicated by the arrows, unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.

[0044] Figure 1A schematic diagram of an application scenario of the text matching method according to an embodiment of the present application;

[0045] Figures 2 to 5 A flowchart of a text matching method according to an embodiment of the present application;

[0046] Figure 6 A structural diagram of a text matching system according to an embodiment of the present application;

[0047] Figure 7 A schematic diagram of the structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0049] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims and drawings of the present application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0050] In the following description, similar expressions such as "a specific implementation" or "a specific example" are involved, which describe a subset of all possible embodiments, but it can be understood that "a specific implementation" or "a specific example" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. In the following description, the term multiple refers to at least two. In some specific examples, the numerical value mentioned in this application reaches the threshold value (if any), which may include the case where the former is greater than the latter of the threshold value; if similar expressions such as "any" or "at least one" are mentioned, it can specifically refer to any one of the listed examples or any combination of these examples.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0052] Still taking the full name of the enterprise (such as "Perfect Software (China) Co., Ltd.") as the full name of the object and the abbreviation of the enterprise as the text to be assigned as an example, the inventors found through multiple rounds of analysis and summary that the reasons why the results of matching directly by the abbreviation of the enterprise are not accurate are roughly divided into the following situations:

[0053] 1. Using part of the full company name as the abbreviation, such as "Perfect", is easy to match by string inclusion, but the matching result range is too large to effectively narrow the target range;

[0054] 2. If you use abbreviations that are made up of different parts of the full name of a company, such as "Perfect China, China Perfect Software Company, Perfect Group, Perfect International, etc.", it is difficult to achieve good results directly through string fuzzy matching. The abbreviations that are made up of different parts of the full name of a company are too short, which can easily lead to semantic fragmentation of text segmentation and reduce the accuracy of matching.

[0055] 3. If the company's nickname or English abbreviation (such as wanmei) is used as the abbreviation, it will be difficult or even impossible to perform a direct query through string fuzzy matching.

[0056] It can be seen that the various abbreviations mentioned above are often formed according to customary conventions and have no standard form, which can easily lead to difficulties in correctly matching when performing string matching. Among them, the traditional solution for nicknames and English abbreviations is to maintain a mapping relationship between the full name and the abbreviation as a common sense library. However, if we only rely on the common sense library to solve the problem, since the company directory is large and changes over time, maintaining and updating the common sense library will become a big problem. In addition, the abbreviation is spliced ​​from different parts of the full name of the company, and it is difficult to achieve good results only through simple text matching.

[0057] Based on the above description, the applicant innovatively proposes a text matching method. The following will first describe the application scenario of the text matching method provided in the embodiment of the present application. Figure 1 .

[0058] Among them, the terminal 102 communicates with the server 101 through the network, and the data storage system 100 can store the data that the server 101 needs to process or needs. The data storage system 100 can be integrated on the server 101, or it can be placed on the cloud or other network servers. The terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers and portable wearable devices. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The server 101 can be implemented as an independent server or a server cluster consisting of multiple servers.

[0059] The full name of the object targeted by the embodiment of the present application can be a description text of different objects such as the full name of the enterprise (or the standard enterprise name), the full name of the document or the full name of the commodity. Taking the enterprise name as an example, the terminal 102 can submit the text to be assigned to the server 101 through the application programming interface (API, Application Programming Interface) opened by the server 101, such as submitting a certain enterprise abbreviation (or non-standard enterprise name) "Perfect Software Company" as the text to be matched. The server 101 can identify the target text label corresponding to the text to be matched in the text label dictionary, and determine the label intersection between the target text label and the text label of each alternative full name in the full name library, so as to calculate the matching score between each alternative full name and the text to be matched based on the label intersection corresponding to each alternative full name. For ease of explanation and understanding, the term "alternative full name" in each embodiment of the present application or the term "object full name" is used as a substitute.

[0060] Based on these matching scores, the server 101 can return a preset number of candidate full names with high matching scores, such as the object full name with the highest matching score, as the target object full name matching the text to be matched to the terminal 102. The object full name with the highest matching score may be the enterprise full name "Perfect Software (China) Co., Ltd.". Further, the server 101 can return detailed information of "Perfect Software (China) Co., Ltd." to the terminal 102, such as the company's job assignment table, administrator changes, enterprise registered capital, and historical rewards and punishments.

[0061] It should be noted that the method provided in the embodiment of the present application can be implemented jointly by the terminal device and the server as described above, or it can be implemented entirely on the server side, or it can be implemented entirely on the terminal device side. The specific method can be determined according to the actual application scenario and is not limited here.

[0062] The method of the present application will be described in further detail below.

[0063] See also Figure 2 In a first aspect, the present application provides a specific embodiment of a text matching method, which includes the following steps:

[0064] Step S21, obtaining a text label dictionary;

[0065] Still taking the company name as an example, the text label dictionary records a part of the full names of the companies and their corresponding text labels. This part of the full names of the companies can come from the company full name library. The number of full names of this part of the companies can be 5 million. These text labels can be divided according to different label types such as "industry" and "keywords". For example, under the "industry" label type, there are specific text labels such as "service, software, medical", and under the "keyword" label type, there are specific text labels such as "perfect, WeChat, investment promotion", etc. The details are not repeated here. In other words, the text label dictionary lists the text labels corresponding to the above 5 million full names of the companies, so the process of constructing this text label dictionary can be summarized as the process of text information extraction (or text labeling).

[0066] Step S22, identifying the target text tag corresponding to the to-be-matched text in the text tag dictionary;

[0067] After obtaining the text tag dictionary, we can identify which text tags the text to be matched hits in the text tag dictionary. Taking the abbreviation of a certain company "Perfect Software Company" as the text to be matched as an example, the target text tag hit by this text to be matched may be "Perfect, Software". Of course, this should be the case when the text tag dictionary itself contains the text tag "Perfect, Software".

[0068] Step S23, determining the label intersection between the target text label and the text label of each candidate full name in the full name library;

[0069] The alternative full name includes the object full name of the above part. In other words, in some examples, each of the above alternative full names can refer to each object full name in the full name library, so the alternative full name itself can be simply understood as the object full name. If the user enters a non-standard company name, then this label intersection can be understood as the degree of coverage of the label set extracted from the non-standard company name to the label set extracted from the standard company name. The more the former's labels cover in the latter's labels, the closer the non-standard company name is to the standard company name. Or it can be simply understood that the label intersection is the number of common labels in the two label sets. Of course, the user can also enter a standard company name.

[0070] For example, label intersection = score of public label 1 + ... + score of public label n, where the label score under "industry" can be pre-set to be less than or equal to the label score under "keyword". Of course, the scores of each label can be the same or different, and can be determined according to the importance or influence of the text label or its type in the text matching task, etc., which is not limited here. When the score of each public label is fixed to 1, the above label intersection can be expressed as the number of public labels in the two label sets. Of course, the score of each public label can also be weighted, such as label intersection = score of public label 1 * this label weight + ... + score of public label n * this label weight. The weights of each label can be the same or different, and can be determined according to the importance or influence of the text label or its type in the text matching task, etc., which is not limited here.

[0071] Step S24, based on the label intersection degree corresponding to each candidate full name, calculate the matching score between each candidate full name and the text to be matched;

[0072] For example, the matching score corresponding to the full name of an object = label intersection * weight value of label intersection + score value of the first item * weight value of the first item + ... + score value of the kth item * weight value of the kth item; wherein, the weight values ​​in this matching score formula can be the same or different, the weight value of the label intersection is greater than 0, and each score value can be greater than or equal to 0; the above-mentioned score value can be the text similarity between the texts themselves (see below for details) and other common information scores other than the above-mentioned text labels, which can be determined by the actual situation and are not limited here.

[0073] Step S25: taking a preset number of candidate full names with high matching scores as the full names of the target objects that match the text to be matched.

[0074] This preset number can be one or more. When it is one, the full name of the target object is the full name of the object with the highest matching score. If it is multiple, a list of object full names can be arranged according to the matching scores, and the full name of the object with the highest matching score can be at the top of the list.

[0075] In summary, the embodiment of the present application takes the text label rather than the text itself as the entry point, determines the label intersection between the target text label and the text label of each candidate full name in the full name library, and thus calculates the matching score between each candidate full name and the text to be matched. This can effectively avoid the problem that when the text to be matched input by the user is a fuzzy term in a non-standard format (such as an abbreviation), directly or only quoting the text to be matched itself to calculate the matching degree between texts will cause the matching result obtained through the text to be matched to be inaccurate. In short, the embodiment of the present application takes the text label as the key point, which can improve the accuracy of fuzzy matching through the text to be matched, and promotes the detailed information of the text to be matched found by association to be highly relevant to user needs, thereby improving the user experience.

[0076] Based on the above example descriptions, some specific possible implementation examples will be provided below. In actual applications, the implementation contents between these examples can be combined or implemented separately as needed according to the corresponding functional principles and application logic. If implemented in combination, the execution order between the combined examples can be determined according to their respective processing logics, which can be determined by the actual scenario.

[0077] If the following possible implementation examples are selected, then in the application scenario of fuzzy matching of enterprise names, the embodiment of the present application can be mainly divided into two processes, namely text information extraction (or text labeling, belonging to the early construction stage) and text matching (or fuzzy query, belonging to the later application stage). For details, please refer to Figure 3 , Figure 4 . Among them, text information extraction is mainly for the processing of standard enterprise industrial and commercial database (i.e. full name database), labeling enterprise information, such as identifying the specific text label corresponding to the full name of a certain enterprise under the label type of "industry, region", such as "software, China", etc., to facilitate subsequent text matching; text matching is mainly to extract label information from the enterprise name text information submitted by the query interface (such as the abbreviation of a certain enterprise as the text to be matched), and then perform fuzzy matching on the processed standard enterprise label information through the extracted label information, and finally return a list of enterprise names with high relevance. For details, please see the following implementation example description.

[0078] based on Figure 2 The present application provides another specific embodiment of a text matching method, which includes the following steps:

[0079] Step S21, obtaining a text label dictionary;

[0080] In some specific examples, the specific operation process of step S21 may include (building a text label dictionary): extracting a part of the full names of the objects from the full name library as samples for word segmentation and word meaning analysis to summarize the text label initial dictionary; the text label initial dictionary is used to record the text labels corresponding to each sample under different label types; for each label type, collect the text labels belonging to the label type in the historical period, and add the collected text labels to the text label initial dictionary to obtain a more complete text label dictionary.

[0081] In some specific examples, when the candidate full name and the text to be matched are both company name description texts, the label type to which the text label belongs includes at least one of: the company's region, the company's industry, the company's name keywords, and the company's name suffix.

[0082] like Figure 3 The schematic diagram of standard enterprise name labeling and indexing shown in the figure focuses on how to extract information (which can be attributed to the link of building a text label dictionary) and index information from standard enterprise names. First, more than 5 million standard enterprise names are randomly selected from the industrial and commercial data package (i.e., the full name library). These enterprise full names can be segmented and analyzed for parts of speech (or meaning) through the Chinese Language Processing (HanLP) tool; then, the segmented words and their parts of speech can be statistically summarized according to pre-designed rules, such as summarizing and summarizing multiple custom part-of-speech dictionaries (i.e., text label dictionaries) such as the region, industry, suffix, and enterprise organization to which the enterprise belongs.

[0083] Figure 3 A judgment step "judging whether the length of the text to be matched is greater than 3 characters" can be added in the process. This is because the applicant has found through summary that if the text to be matched submitted by the interface is greater than 3, this is generally a search for a long sentence name; but if the length of the text to be matched submitted by the interface is less than or equal to 3 characters, it may be a search for a short text name (such as a company name named after a person), so there is no need to segment the text to be matched (which can be attributed to the information extraction link), but directly use it as the segmented text for labeling and classification, saving operating resources; of course, without considering the waste of operating resources, it is also possible to segment the text with a length of less than or equal to 3 characters, and then label and classify the segmented words one by one.

[0084] like Figure 5 As shown, the following will describe in more detail how to extract information from standard enterprise names, that is, the process of building a text label dictionary.

[0085] (1) Through the analysis and summary of a large number (more than 5 million) of company names, the company names are summarized into four parts: region, keyword, industry, and name suffix. This rule is used to extract labels for each component of the company name. For example, for the company name: Perfect Software (China) Co., Ltd., the text label extracted under the region tag type is China, the text label under the keyword tag type is Perfect, the text label under the industry tag type is Software, and the text label under the name suffix tag type is Co., Ltd. Through the part-of-speech tagging function provided by HanLP, more than 5 million randomly selected company names are segmented and the part-of-speech of the company names after segmentation is analyzed; then, the parts of speech (that is, the specific text labels) of the segmented words of these company names under the tag types such as region, keyword, industry, and name suffix can be summarized respectively, and the corresponding custom part-of-speech dictionary can be compiled. The custom part-of-speech dictionary includes:

[0086] ◆Custom enterprise organization word dictionary (i.e. keyword dictionary): contains nicknames and abbreviations of large enterprises, usually maintaining the top 500 companies in the country in previous years, as well as relatively well-known listed companies.

[0087] ◆Customized regional part-of-speech dictionary: It can be used to supplement regional identification, that is, to supplement some cities and counties that cannot be identified. For example, the default regional dictionary library of HanLP (which can be regarded as a component of the initial dictionary of text tags) is not perfect, so county-level information or even township-level information can be supplemented according to the actual business scenario; the regional part-of-speech dictionary can contain the name of a country and the names of its provinces, cities and counties at the same time, so a company name can be attributed to more than two regional text tags at the same time, depending on the naming composition of the company name.

[0088] ◆Customized industry part-of-speech dictionary: It can be used to supplement industry identification, summarize and count by oneself, or organize according to public industry classifications, including standard industry classifications divided by the country and industry classifications collected and published on the Internet. The industry part-of-speech is a customized part-of-speech <named ind>, that is, a customized industry part-of-speech named ind is defined in the part-of-speech dictionary; similarly, a company name can be attributed to more than two industry text tags at the same time, depending on the naming composition of the company name. For example, if there are two industry text tags of "technology, service" at the same time, the score under such tags can be doubled. The specific setting can be based on the actual situation and is not limited here.

[0089] ◆Customize the company suffix word dictionary. Common company suffixes include: company, limited company, limited liability company, joint stock limited company, group, etc. In addition, organization suffixes can also include:

[0090] Hotels, offices, firms, schools, factories, research institutes, offices, management offices, etc. Similarly, a large number of suffixes can also be summarized as supplements.

[0091] In a word, HanLP's built-in core dictionaries such as "region, keyword, industry, name suffix" can be classified as general dictionaries and regarded as components of the initial dictionary of text tags; in addition to HanLP's built-in core dictionaries, you can customize the "region, keyword, industry, name suffix" and other types of customized part-of-speech dictionaries for supplementary purposes. For example, through more than 5 million full names of enterprises, high-frequency text tags are summarized and sorted out to supplement the initial dictionary of text tags to obtain a more complete text tag dictionary. The addition of a customized part-of-speech dictionary can overcome the limitations of the initial dictionary of text tags, making the information extraction and classification of standard and non-standard enterprise names more accurate and efficient. In addition, the text to be assigned in the embodiment of the present application can be a text description in Chinese, English, Arabic numerals, or other languages, without limitation.

[0092] Specifically, the identification rules for extracting and classifying information about company names can be as follows:

[0093] ◆Region identification: ns (place name, with customizable regional word category dictionary)

[0094] ◆Identify industry: ind (can be derived from the assistance of a custom industry word dictionary)

[0095] ◆Suffix identification: nis (organization suffix, company suffix part-of-speech dictionary assistance)

[0096] ◆Identify keywords: In addition to the above three parts of speech, stop words (such as prepositions, pronouns, conjunctions, auxiliary verbs, etc.) and w (punctuation marks), the connection results of the first n participles in the company name can be identified as keywords. This n can take any value from 4 to 6 according to experience.

[0097] (2) Build a custom part-of-speech dictionary, which is mainly used to enhance the part-of-speech recognition function of HanLP and improve the accuracy of information extraction. The custom part-of-speech dictionary can be configured to the custom dictionary directory of HanLP (HanLP installation directory / data / dictionary / custom) and extract information from the entire enterprise data.

[0098] Step S22, identifying the target text tag corresponding to the to-be-matched text in the text tag dictionary;

[0099] Specifically, based on the text tag dictionary constructed based on the above-mentioned more than 5 million company names, information can be extracted from the submitted search text (i.e., the text to be matched). For example, the submitted company abbreviation can be labeled, that is, the text tags belonging to this company abbreviation can be identified as "China, Shenzhen, software, company", etc.

[0100] In some specific examples, before step S23, the method of the embodiment of the present application may also include (performing text label recognition on the existing full name of all objects): performing word segmentation on the full names of other objects in the full name library except for the full names of the objects mentioned above, and performing semantic analysis on the segmentation results; combining the semantic analysis results, using a text label dictionary to perform text label recognition on the segmentation results to obtain text labels corresponding to the full names of other objects.

[0101] As explained above, after constructing a text label dictionary based on the above-mentioned more than 5 million company names, the HanLP core part-of-speech dictionary (optional) and multiple custom part-of-speech dictionaries can be combined to extract and classify the full names of all companies in the industrial and commercial data package, that is, to extract the region, keywords (abbreviations), industries, suffixes and other labeled information from all standard company names in the library. Of course, standard company names that have been extracted and classified do not need to be repeatedly extracted and classified to avoid wasting resources. Finally, the labeled information of each standard company name and the standard company name can be indexed into a distributed search engine (ES, Elasticsearch) for downstream text matching. At this point, it can be regarded as the completion of the information extraction and indexing of all standard company names in the full name library.

[0102] In some specific examples, other auxiliary search information of each standard enterprise name can also be indexed into ES to assist ES to accurately and quickly find the full name of the enterprise corresponding to the submitted enterprise abbreviation through the combination of text tags and other auxiliary information. The above-mentioned other auxiliary search information can be at least one of the enterprise identification id, enterprise address, establishment date, enterprise capital, listing code, listing mark, etc. In addition to ES, the above-mentioned search engine can also be a search engine such as Lucene or Sphinx, which is not limited to the embodiments of this application. The above-mentioned HanLP is a powerful natural language processing (NLP, Natural Language Processing) toolkit, which is mainly used for various NLP tasks such as Chinese word segmentation, lexical analysis, and syntactic analysis; of course, the jieba word segmentation tool jieba can be used to replace or assist the above-mentioned HanLP tool.

[0103] Step S23, determining the label intersection between the target text label and the text label of each candidate full name in the full name library;

[0104] Step S24, based on the label intersection degree corresponding to each candidate full name, calculate the matching score between each candidate full name and the text to be matched;

[0105] In some specific examples, other scoring factors besides label intersection can be introduced to calculate the matching score, so the specific operation of step S24 may include: obtaining the text similarity between the text to be matched and each alternative full name; based on the label intersection and text similarity corresponding to each alternative full name, calculating the matching score between each alternative full name and the text to be matched.

[0106] As described above, the matching score corresponding to the full name of an object = label intersection * weight value of label intersection + score value of the first item * weight value of the first item + ... + score value of the kth item * weight value of the kth item; wherein k is greater than or equal to 1, and the weight values ​​in this matching score formula can be the same or different, and a certain score value can be the text similarity between the full name of the object participating in the matching and the text to be matched. This text similarity can specifically be the shortest edit distance and / or the number of common words between the two. This shortest edit distance can refer to the minimum number of single-character edits (insertion, deletion or replacement) required to convert one string into another. Of course, in addition to this, the above-mentioned score value can also be the registered capital of the enterprise participating in the matching, the listing status (such as the listing score value can be recorded as 1, otherwise it can be recorded as 0) and other scoring items, which are not limited here.

[0107] In some specific examples, the matching score (or the original total score) obtained above can be further adjusted to make the final matching result more accurate. Therefore, after step S24, the method of the embodiment of the present application can also include: for a preset number of candidate full names with high matching scores, obtaining relevant index information of a preset number of candidate full names; wherein the preset number is greater than or equal to the preset number, and the relevant index information describes the object details of the candidate full name from an angle different from the text label; using the relevant index information as an influencing factor, adjusting the matching score of each candidate full name to obtain a new matching score; the new matching score is used to select the full name of the target object.

[0108] The above-mentioned influencing factors can be at least one of other auxiliary search (ES) index information such as enterprise ID, establishment date, enterprise capital, listing code, listing mark, whether the enterprise is valid, extraction status, text type, etc., which can be selected according to actual conditions and are not limited here.

[0109] Continue Figure 5As shown in the above description, (3) the extracted enterprise name information can be labeled and indexed into ES for text information retrieval. In addition to the label information, in order to improve the accuracy of text matching, a custom scoring influencing factor can be added, so that on the basis of the original total score, the score intervention is performed through the size of the enterprise's registered capital, establishment date, listing mark and other influencing factors to change the relevance of the search result document; accordingly, the ES custom function parameters can be adjusted to adjust the influence of the above influencing factors on the original total score, such as adjusting the weight of the influencing factor in the scoring function as needed.

[0110] (4) The ultimate goal of enterprise information retrieval is to obtain detailed information about the enterprise. Therefore, after ES text retrieval, the list of enterprises with high relevance can be returned based on the score calculation. Finally, the target enterprise name can be screened from the returned list, or the complete information of the target enterprise name (or enterprise details information) can be further extracted. The complete enterprise information can be stored in the distributed document storage database MongoDB after standardized processing (such as labeling), including the enterprise's job assignment table, administrator changes, enterprise registered capital, historical rewards and punishments and other industrial and commercial information; specifically, the complete enterprise information can be efficiently extracted from MongoDB by the target enterprise name or enterprise ID.

[0111] (5) After all data preparation is completed and the outside world needs to consume data, the most direct and secure way may be to provide an API call. That is, the embodiment of the present application may build a microservice to provide an API query, so that users can query the list of enterprise names by submitting the enterprise abbreviation (including non-standard enterprise names). Of course, the interface parameter submitted may also be the full name of the enterprise.

[0112] (6) Extract information from API parameters (such as company abbreviations) through the HanLP custom part-of-speech dictionary, and assemble the ESAPI query logic based on the pre-set label matching weights. The label matching weight parameters can be set according to the query effect after a long period of fine-tuning. For example, according to the importance, the weight assigned to each text label can be: the weight of the keyword class ≥ the weight of the industry class ≥ the weight of the region class ≥ the weight of the suffix class. Compared with the label information set extracted from the standard company name, the label information extracted from the non-standard company name may be less, and the score of the missing text label can be recorded as 0. In the matching score calculation stage, ES can comprehensively score the coverage and label weight of the label set extracted from the standard company name based on the label set extracted from the non-standard company name. In addition, the score intervention can be carried out based on influencing factors such as the size of the company's registered capital to improve the relevance of the query results. Among them, an accurate matching score "kth score value" can be added. If the non-standard name parameter passed by the API can accurately match the standard enterprise name (the full name of the enterprise may be passed), then this kth score value can be directly multiplied, thereby widening the score gap with other similar names and making it most relevant.

[0113] Step S25: taking a preset number of candidate full names with high matching scores as the full names of the target objects that match the text to be matched.

[0114] In some specific examples, after step S25, the method of the embodiment of the present application may also include: screening out at least one alternative full name from a preset number of alternative full names; returning the screened out at least one alternative full name, and / or detailed information of the screened out at least one alternative full name to the input end of the text to be matched.

[0115] Continue Figure 5 As shown in the above description, (7) after the ES fuzzy matching query is performed on the text to be assigned submitted through the API, the ES cluster returns a fuzzy query result list, which includes the full name of the enterprise and its respective matching score results. Of course, the matching score results may not be returned, or the label information of the enterprise name may be added and returned according to the actual application, without limitation.

[0116] (8) If the application scenario is only for enterprise name calibration or enterprise name list retrieval, the fuzzy query list returned by ES can be returned directly to the calling end through the API interface.

[0117] (9) If the application scenario requires selecting a target enterprise based on the results of fuzzy matching to obtain detailed information to complete the enterprise data, the full name of the target enterprise can be filtered out from the fuzzy query result list, and a precise query task can be initiated in MongoDB based on the full name of the target enterprise to call the desired detailed information of the target enterprise.

[0118] (10) The API returns detailed information obtained through accurate query of MongoDB.

[0119] like Figure 4 The fuzzy matching flowchart of enterprise names shown in the figure will focus on summarizing how to perform fuzzy matching through enterprise name information and how to improve the accuracy of matching. First, the data call task can be initiated by the API, and the enterprise name information is passed in. The backend needs to determine whether the enterprise name is passed in; secondly, the HanLP core part-of-speech dictionary and multiple custom part-of-speech dictionaries are combined to extract information from the enterprise name passed in by the API, and the region, keywords (or abbreviations), industry, suffix and other labeled information are extracted. Since the enterprise name information passed in by the API may be incomplete, there may be missing information in the label extraction, such as missing industry, region, etc.; finally, the label information in ES is matched according to the extracted label information, and a preliminary score is made based on the label coverage, text matching and label weight between the submitted enterprise abbreviation and the full name of each enterprise in the database to obtain the matching score (that is, the original total score).

[0120] To further improve the accuracy of text matching, you can add custom scoring influencing factors. On the basis of the original total score, you can intervene in the score through influencing factors such as the amount of registered capital of the enterprise to change the relevance of the search result document, and adjust the ES custom function parameters to adjust the influence of registered capital to improve the relevance of the query results. In addition, you can add an accurate matching score: if the non-standard name parameter passed by the API can accurately match the standard enterprise name, the score will be directly multiplied, widening the score gap with other similar names, so that it has the highest relevance in the query result list.

[0121] In summary, when processing company name matching, the embodiments of the present application can be mainly divided into two parts, namely text information extraction (text labeling) and fuzzy query (text matching). Among them, information extraction: It is aimed at the full company names in the full name library and the company abbreviations submitted by users, which can be attributed to the early construction stage. Specifically, the company full name can be disassembled through text segmentation, and then the possible segmentation forms (which can be spliced ​​and combined) are classified by part of speech recognition. Fuzzy matching: It can belong to the later application stage, which is mainly based on the shortest edit distance algorithm to calculate the text similarity, and calculate the 4 types of label scores between the full name of each company and the submitted company abbreviation, which can be a total of 1+4 scoring items, and then the best matching result, that is, the most suitable target company full name, is found by screening the highest matching score.

[0122] In addition, some additional scoring rules and scoring item weights can be added to change the final scoring results through specific influencing factors, such as: companies with higher registered capital may be larger in scale, companies with longer registration time may have better reputation and brand image, and listed companies may have greater exposure, etc. Generally, the scoring process can be completed using the score calculation function of ES.

[0123] like Figure 5 As shown in the timing diagram, the design of the embodiment of the present application can be summarized as follows: construction of a custom part-of-speech dictionary (for the full names of companies in the full name library), information extraction and indexing of all standard company names in the library using the constructed custom part-of-speech dictionary, information extraction of submitted search texts (such as the abbreviation of a company), and customization of ES text search scoring rules. Such a series of operations helps to improve the accuracy of text matching results.

[0124] Figure 5 The timing diagram shown can also be described as the following process, where processes 11 to 15 can be regarded as the process of DaaS data subscription consumption:

[0125] 1. Data developers randomly extract more than 5 million company names from the industrial and commercial database and perform word segmentation through HanLP. They statistically summarize a large number of word parts and build multiple custom word part dictionaries. 2. Data developers initiate ETL tasks so that the backend can process standard corporate data. ETL is the abbreviation of Extract-Transform-Load, which is used to describe the process of extracting, transforming, and loading data from the source to the destination. It is the lifeline of the data warehouse. 3. Through HanLP custom word parts, the company names are extracted to obtain regions, keywords, industries, suffixes, and other auxiliary search information and indexed into Elastic search; 5. Data developers develop interfaces to query enterprise information through enterprise abbreviations; 8. Return a matching list queried through enterprise abbreviations; 4. Load the processed standard enterprise data into MongoDB; 6. HanLP extracts information from API parameter enterprise abbreviations and assembles ES query logic based on pre-adjusted different tag matching weight parameters; 7. Sort in reverse order by ES tag matching score + custom score value, and return a list of fuzzy query results; 9. Filter target enterprises for precise query; 10. Return detailed information of target enterprises; 11. Data developers create services in API gateway; 12. Add interfaces to API gateway services; 13. Service publishing; 14. Users subscribe to data services; 15. Users consume data

[0126] In short, the embodiments of the present application: 1. Use HanLP to perform text segmentation and part-of-speech analysis on company names; 2. Through recognition rules + word segmentation and part-of-speech induction, compile a custom part-of-speech dictionary; 3. Extract information and label information on company names; 4. ES basic query scoring + ES custom scoring to achieve comprehensive scoring; 5. In order to optimize the accuracy in the later stage, more scoring influencing factors can be added for comprehensive scoring, thereby improving the accuracy of corporate business data services in DaaS applications through fuzzy matching of company names. For example, the accuracy rate has been greatly improved especially for non-standard company name query scenarios.

[0127] Compared to Figure 2 The examples shown illustrate that the above-mentioned additional or detailed example contents or possible implementation methods do not necessarily have to be executed during the specific implementation. If more than two examples or possible implementation methods are added, these examples or possible implementation methods can be implemented in combination or separately. If implemented in combination, the execution order between the combined examples can be determined according to their respective processing logic, which can be determined according to the actual scenario.

[0128] See also Figure 6 , the second aspect of the present application provides a specific embodiment of a text matching system, the system comprising: an acquisition unit 601 and a processing unit 602;

[0129] The acquisition unit 601 is used to acquire a text label dictionary; the text label dictionary contains a portion of object full names corresponding to the corresponding text labels;

[0130] Processing unit 602, used to identify the target text tag corresponding to the to-be-matched text in the text tag dictionary;

[0131] The processing unit 602 is further used to determine the label intersection degree between the target text label and the text label of each candidate full name in the full name library; the candidate full name includes the above-mentioned part of the object full name;

[0132] The processing unit 602 is further configured to calculate a matching score between each candidate full name and the text to be matched based on the label intersection corresponding to each candidate full name;

[0133] The processing unit 602 is further configured to use a preset number of candidate full names with high matching scores as the full names of the target objects that match the text to be matched.

[0134] In some examples, processing unit 602 is used to:

[0135] A portion of the object full names are extracted from the full name library as samples for word segmentation and word meaning analysis to summarize the initial dictionary of text labels; the initial dictionary of text labels is used to record the text labels corresponding to each sample under different label types;

[0136] For each tag type, the text tags that belong to the tag type in the historical period are collected, and the collected text tags are added to the initial text tag dictionary to obtain a text tag dictionary with more complete text tags.

[0137] In some examples, the processing unit 602 is further configured to:

[0138] Perform word segmentation on the full names of other objects in the full name library except for the full names of the objects described above, and perform word meaning analysis on the word segmentation results;

[0139] Combined with the word meaning analysis results, the text label dictionary is used to perform text label recognition on the word segmentation results to obtain the text labels corresponding to the full names of other objects.

[0140] In some examples, processing unit 602 is used to:

[0141] Obtaining the text similarity between the text to be matched and each candidate full name;

[0142] Based on the label intersection and text similarity corresponding to each candidate full name, the matching score between each candidate full name and the text to be matched is calculated.

[0143] In some examples, the processing unit 602 is further configured to:

[0144] For a preset number of candidate full names with high matching scores, obtain relevant index information of the preset number of candidate full names; wherein the preset number is greater than or equal to the preset number, and the relevant index information describes the object details of the candidate full names from a perspective different from the text label;

[0145] The relevant index information is used as an influencing factor to adjust the matching score of each candidate full name to obtain a new matching score; the new matching score is used to select the full name of the target object.

[0146] In some examples, the processing unit 602 is further configured to:

[0147] Selecting at least one candidate full name from a preset number of candidate full names;

[0148] The at least one selected candidate full name and / or the detailed information of the at least one selected candidate full name are returned to the input end of the text to be matched.

[0149] In some examples, when the candidate full name and the text to be matched are both company name description texts, the label type to which the text label belongs includes at least one of: the company's region, the company's industry, the company's name keywords, and the company's name suffix.

[0150] In summary, the embodiments of the present application mainly solve the problem that the fuzzy query results of corporate business data services through corporate abbreviations in DaaS applications are not accurate. The company name is extracted and labeled to establish an index relationship, and the influencing factors of auxiliary search and the scoring weights of label matching are increased to comprehensively score, so that the company name query result list Top-n (n can be 10) is likely to contain the required target company full name.

[0151] In the embodiment of the present application, the operations performed by each unit of the text matching system are similar to the operations described in the first aspect or any specific method embodiment of the first aspect, and will not be described in detail here. Of course, the specific implementation process of each operation in the first aspect of the present application can also refer to the relevant description of the second aspect.

[0152] See also Figure 7 The electronic device 700 of the embodiment of the present application may include one or more processors (such as central processing units (CPU)) 701 and a memory 705, in which one or more applications or data are stored.

[0153] The memory 705 may be a volatile storage or a persistent storage. The program stored in the memory 705 may include one or more modules, each of which may include a series of instruction operations in the electronic device. Furthermore, the processor 701 may be configured to communicate with the memory 705 and execute a series of instruction operations in the memory 705 on the electronic device 700.

[0154] The electronic device 700 may also include one or more power supplies 702, one or more wired or wireless network interfaces 703, one or more input and output interfaces 704, and / or one or more operating systems, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

[0155] The processor 701 can execute the operations performed by the aforementioned first aspect or any specific method embodiment of the first aspect, and the details will not be repeated here.

[0156] The present application provides a computer-readable storage medium, comprising instructions, which, when executed on a computer, enable the computer to execute the method described in the first aspect or any specific implementation of the first aspect.

[0157] The present application provides a computer program product comprising instructions or a computer program. When the computer program product is run on a computer, the computer is enabled to execute the method described in the first aspect or any specific implementation of the first aspect.

[0158] It is understood that in various embodiments of the present application, the sequence number of each step does not mean the order of execution, and the execution order of each step should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application. The operation content added or refined in each example scheme of the above method, system or device (if any) does not necessarily have to be executed in the specific implementation. If more than two operations are added, these operations can be implemented in combination or separately, depending on the actual scenario.

[0159] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system (if any) and device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0160] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system or device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0161] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0162] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0163] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product (or computer program product) is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a business server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk and other media that can store program code.

Claims

1. A text matching method, characterized in that: include: Obtain a text label dictionary; the text label dictionary contains a portion of object full names corresponding to the corresponding text labels; Identify the target text tag to which the to-be-matched text corresponds in the text tag dictionary; Determining the label intersection between the target text label and the text label of each candidate full name; the candidate full name includes the object full name of the part; Based on the label intersection degree corresponding to each of the candidate full names, calculating the matching score between each of the candidate full names and the text to be matched; A preset number of the candidate full names with high matching scores are used as the full names of the target objects that match the text to be matched.

2. The text matching method according to claim 1, characterized in that: The obtaining of the text label dictionary includes: Extracting a portion of the full names of the objects from the full name library as samples for word segmentation and word meaning analysis to summarize an initial dictionary of text labels; the initial dictionary of text labels is used to record the text labels corresponding to each of the samples under different label types; For each of the tag types, the text tags that belong to the tag type in the historical period are collected, and the collected text tags are added to the text tag initial dictionary to obtain a text tag dictionary with more complete text tags.

3. The text matching method according to claim 1, characterized in that: Before determining the label intersection degree between the target text label and the text label of each candidate full name, the method further includes: Performing word segmentation on the full names of other objects in the full name library except for the part of the full names of the objects, and performing word meaning analysis on the word segmentation results; In combination with the word meaning analysis result, the text label dictionary is used to perform text label recognition on the word segmentation result to obtain the text label corresponding to the full name of the other object.

4. The text matching method according to claim 1, characterized in that: The calculating the matching score between each candidate full name and the text to be matched based on the label intersection corresponding to each candidate full name includes: Obtaining text similarity between the text to be matched and each of the candidate full names; Based on the label intersection and the text similarity respectively corresponding to each of the candidate full names, a matching score between each of the candidate full names and the text to be matched is calculated.

5. The text matching method according to claim 1, characterized in that: After calculating the matching score between each candidate full name and the text to be matched based on the label intersection corresponding to each candidate full name, the method further includes: For a preset number of candidate full names with high matching scores, obtaining relevant index information of the preset number of candidate full names; wherein the preset number is greater than or equal to the preset number, and the relevant index information describes the object details of the candidate full names from a perspective different from that of the text labels; The relevant index information is used as an influencing factor to adjust the matching score of each candidate full name to obtain a new matching score; the new matching score is used to select the full name of the target object.

6. The text matching method according to claim 1 or 5, characterized in that: After taking the preset number of candidate full names with high matching scores as the full names of the target objects that match the text to be matched, the method further includes: Selecting at least one of the candidate full names from the preset number of candidate full names; The at least one selected candidate full name and / or the detailed information of the at least one selected candidate full name are returned to the input end of the text to be matched.

7. The text matching method according to any one of claims 1 to 5, characterized in that: When the candidate full name and the text to be matched are both company name description texts, the tag type to which the text tag belongs includes: At least one of the region to which the enterprise belongs, the industry to which the enterprise belongs, the enterprise name keyword, and the enterprise name suffix.

8. A text matching system, characterized in that: include: Acquisition unit and processing unit; The acquisition unit is used to acquire a text label dictionary; the text label dictionary contains a portion of object full names corresponding to the corresponding text labels; The processing unit is used to identify the target text label to which the to-be-matched text corresponds in the text label dictionary; The processing unit is further used to determine the label intersection degree between the target text label and the text label of each candidate full name; the candidate full name includes the part of the object full name; The processing unit is further used to calculate a matching score between each of the candidate full names and the text to be matched based on the label intersection corresponding to each of the candidate full names; The processing unit is further configured to use a preset number of candidate full names with high matching scores as target object full names that match the text to be matched.

9. An electronic device, characterized in that: include: Processor and memory; The processor is configured to communicate with the memory and execute instructions in the memory to implement the method according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

11. A computer program product, characterized in that The computer program product comprises computer instructions, which implement the method according to any one of claims 1 to 7 when executed by a processor.

Images