Text abbreviation automatic recognition method and system for intelligent customer service

Abstract

The present application relates to the field of Chinese abbreviation recognition, and more particularly to a text abbreviation automatic recognition method and system for intelligent customer service, which comprises the following steps: collecting input text and input rhythm within a preset time period to form corresponding text records; dividing sentences in the text to obtain corresponding single characters and words, marking the single characters or words after sorting and deduplication; forming corresponding text records by collecting input text and input rhythm within a preset time period, and dividing sentences to obtain single characters and words, which can completely capture the context information in customer service interaction and provide sufficient data basis for subsequent abbreviation recognition. By sorting the single characters and words in descending order of frequency of occurrence, then performing interpretation matching and deduplication, high-frequency words are processed preferentially, which effectively improves the retrieval efficiency and reduces redundant calculation.

Description

Technical Field

[0001] The present invention relates to the field of Chinese abbreviation recognition, and in particular to a method and system for automatic text abbreviation recognition for intelligent customer service. Background Art

[0002] Abbreviations generally refer to special words with shortened length and unchanged meaning formed by compressing and reordering one or more words, and the original form before compression is called the primitive of the abbreviation. The abbreviation methods of Chinese are relatively complex and diverse. Meanwhile, with the rapid development of informatization and intelligent construction in the construction industry, text information in the construction field has exploded, covering design descriptions, construction logs, supervision reports, engineering change orders, completion acceptance materials, and multi-party meeting minutes, etc. In these texts, a large number of professional abbreviations are used to improve writing and reading efficiency, for example, "BIM" refers to Building Information Modeling, "tong" replaces "concrete", "PC component" refers to "precast concrete component", and "jian shi tu" is the abbreviation of "architectural construction drawing", etc.

[0003] In addition, different participants - design institutes, construction units, supervision units and developers - often form a large number of non-standard custom abbreviations according to their respective habits and regions, such as "jie ping tu" (structural layout plan), "feng nuan tu" (ventilation and heating drawing), and "ji keng zhi hu zhuan pian" is abbreviated as "zhi hu pian", etc. These abbreviations have various forms, and the word formation methods include: initial letter abbreviation (BIM, CAD), morpheme interception (tong, jian shi), number homophony or code designation (such as "3D" in "3D printed building"), as well as the mixed abbreviations unique to the field.

[0004] Due to the complexity of the formation methods of abbreviations and the endless emergence of new words, there are few versions of Chinese abbreviation dictionaries at present, which are mainly compiled by experts based on personal knowledge, difficult to be exhaustive, and updated slowly. In addition, the primitives can be retrieved online, such as through Baidu and Google search engines, which have a high recognition rate in the recognition of common general abbreviations. However, due to the low accuracy of word frequency statistics itself, the recognition accuracy is difficult to further improve when used alone.

[0005] Chinese Patent Publication No. CN109614493B discloses a method and system for text abbreviation recognition based on supervised word vectors, including the following steps: 1. Training a supervised word vector model using an open large corpus; 2. Labeling the parts of speech of primitive words in an existing abbreviation dictionary; forming a sequence of primitive word parts of speech; 3. Calculating the parts of speech of a given text; 4. Selecting candidate primitive words and candidate abbreviations for the given text; 5. Inputting the candidate primitive words and candidate abbreviations into the synonym relationship classifier of the supervised word vector model for discrimination. It can be seen that the technical solution lacks sufficient verification of the meaning of abbreviations in the original text, resulting in insufficient recognition accuracy for text abbreviations. Summary of the Invention

[0006] To address this issue, the present invention provides a method and system for automatic recognition of text abbreviations for intelligent customer service, which overcomes the problem in the prior art where the verification of the meaning of abbreviations in the original text is insufficient, resulting in insufficient recognition accuracy of text abbreviations.

[0007] On the one hand, this invention provides a method for automatically recognizing text abbreviations in intelligent customer service, used to identify and merge long text information in customer service interactions, including: Collect the text and input rhythm within a preset time period to form a corresponding text record; The sentences in the text are segmented to obtain the corresponding single characters and words; The segmented characters and words are sorted in descending order according to their frequency of occurrence in the text; Retrieve from the storage database, search for each sorted single character and / or word to match its corresponding definition, and remove duplicate single characters and / or words with the same definition; Its characteristic is that, after sorting and deduplication, the single character or word is marked, including, When a matching definition is found in the abbreviation database, it is marked as the first abbreviation. When there is no matching definition in the abbreviation database, it is marked as a second abbreviation; For each character or word marked as the second abbreviation, the central word of the corresponding text is determined according to the position of the character or word, and the second abbreviation is extended and associated according to the formation of the central word to obtain the corresponding number of extended interpretations; The definition of an abbreviation is determined based on the number of extended definitions obtained, indicating whether the definition meets the preset extension criteria. In response to non-compliance with the preset extension criteria, the meaning of the abbreviation is redefined based on the extended meaning corresponding to a single abbreviation. In response to conforming to preset expansion criteria, verify whether each expanded definition matches the abbreviation, or... After filtering the extended definitions based on the domain, the extended definitions corresponding to the domain are selected, and it is verified whether each extended definition matches the abbreviation. The first abbreviation and the second abbreviation are derived and an abbreviation mapping is formed. The mapping table includes the abbreviation, the original word of the abbreviation, and the definition of the abbreviation. The storage library includes an abbreviation database, a word database, and a text database.

[0008] In particular, the step of determining whether the definition of an abbreviation meets the preset expansion criteria based on the number of extended definitions includes: When the number of extended definitions is less than the preset extension number threshold, the abbreviation method is determined to be morpheme abbreviation, and the morpheme abbreviation is divided into several abbreviations, and, Obtain the extended meaning corresponding to a single abbreviation, and redefine the meaning of the abbreviation based on the extended meaning corresponding to the single abbreviation; When the number of extended definitions is greater than or equal to the preset extension number threshold, the corresponding abbreviation in the text is replaced by the extended definition corresponding to the abbreviation in turn to verify whether each extended definition matches the abbreviation. Alternatively, keywords can be extracted sequentially from the extended definitions corresponding to the abbreviation. Based on the domain to which the definition containing the keyword belongs, the extended definitions are categorized by domain. Then, the extended definitions in the corresponding domains are selected to replace the corresponding abbreviations in the text in turn to verify whether each extended definition matches the abbreviation. The preset expansion standard is whether the number of expanded interpretations obtained after expanding the association for the second abbreviation reaches a preset expansion number threshold. When the number of extended definitions obtained is greater than or equal to the preset extension number threshold, the definition of the abbreviation is determined to meet the preset extension standard. When the number of extended definitions obtained is less than the preset extension number threshold, it is determined that the definition of the abbreviation does not meet the preset extension standard, and the abbreviation method of the abbreviation is determined to be morpheme abbreviation.

[0009] In particular, obtaining the extended meaning corresponding to a single abbreviation includes: For a single abbreviation, based on the composition of the central word, the abbreviation is expanded and associated to obtain several extended meanings corresponding to the single abbreviation; When an extended meaning cannot be obtained by expanding association for a single abbreviation, the original character of the abbreviation is retained and used as the extended meaning corresponding to the abbreviation.

[0010] In particular, for any abbreviation, the step of redetermining the meaning of the abbreviation based on the extended meaning corresponding to the individual abbreviation includes: Among the extended definitions corresponding to its adjacent abbreviations; The extended definition that best matches the extended definition of the current abbreviation in terms of semantic coherence is taken as the matching extended definition of the adjacent abbreviation, and the matching extended definitions of each abbreviation are concatenated in sequence to reconstruct the definition of the abbreviation.

[0011] In particular, after recombining the meaning of the abbreviation, the method further includes: The reconstructed definition is used as an extended definition to replace the abbreviation at the corresponding position in the text, and the degree of matching between the replaced statement and the text before and after the abbreviation is verified to obtain the matching degree. The matching degree refers to the quantitative value of the degree to which the statement formed after expanding the definition and replacing the abbreviation is semantically consistent with the text before and after the abbreviation.

[0012] In particular, after obtaining the matching degree, it is determined whether the meaning of the reconstructed abbreviation conforms to the preset expansion standard based on the matching degree; When the abbreviation does not meet the preset extension criteria, the abbreviation is rematched with another extended meaning corresponding to its adjacent abbreviation to form a new meaning; or, the abbreviation is determined to have multiple meanings, and the meaning of the abbreviation is re-determined by combining the abbreviation with the text before and after it.

[0013] In particular, after determining whether the matching degree meets the preset extension criteria, if the matching degree does not meet the preset extension criteria, the following steps are performed: Replace the corresponding abbreviation in the text with the next extended definition corresponding to the abbreviation, and verify again whether the extended definition matches the abbreviation, or... The abbreviation is determined to be incorrect, and the incorrect abbreviation is exported.

[0014] In particular, when extracting keywords from extended definitions for domain segmentation, the selection of extended definitions corresponding to the domain includes: The number of extended definitions in each field is counted separately. The field with the most extended definitions is selected as the field to which the abbreviation belongs, and the extended definitions in that field are used for subsequent matching and verification.

[0015] For the input rhythm, the segmentation of sentences in the text to obtain corresponding single characters and words specifically includes: Based on the pause duration between adjacent input characters in the input rhythm, the character boundaries where the pause duration is greater than a preset pause threshold are identified as word boundaries to divide the single characters and words; wherein, the preset pause threshold is a pre-set value, which is 800 milliseconds in this embodiment.

[0016] This invention further provides feasible quantitative details and a weighting calculation system: The confidence score Conf of the second abbreviation is defined and obtained using a weighted calculation formula: Conf=w1·S_speed+w2·S_pause+w3·(1−S_edit) Wherein, S_speed is the input speed normalized score, S_pause is the confirmation pause normalized score, S_edit is the edit modification number normalized value, and w1, w2, and w3 are weight coefficients with w1+w2+w3=1. In this embodiment, they are taken as 0.4, 0.3, and 0.3 respectively.

[0017] The input speed normalization score S_speed is obtained by comparing the average input time per character with a preset fast input threshold. The preset fast input threshold is a pre-set value, which is 200 milliseconds / character in this embodiment. When the average input time per word is less than this threshold, S_speed takes a higher value.

[0018] When the confidence score Conf is greater than or equal to the preset high confidence threshold (0.7 in this embodiment), the confidence of the abbreviation or the priority of extended association of the second abbreviation is increased; when the input rhythm feature indicates that the user is uncertain about the word, that is, the pause duration after input is greater than the preset confirmation pause threshold (1500 milliseconds in this embodiment) or the number of modifications is greater than the preset number threshold, the matching threshold between the extended definition and the text is lowered by a preset ratio; when the input rhythm feature indicates that the user is confident that the meaning of the word is clear, that is, the input is fluent and uninterrupted and is skilled input, the most common or most authoritative definition is selected as the matching result.

[0019] To alleviate the problem of over-segmentation or under-segmentation caused by differences in typing habits among different users, the central control unit maintains a personal input habit profile for each user terminal and performs personalized calibration on the preset pause threshold, preset fast input threshold, preset confirmation pause threshold, and weight coefficient based on the user's own historical input data; the statistical method of the historical input data and the parameter determination method are described in the relevant description of the adaptive determination of preset parameter standards in the previous part of this application specification.

[0020] Furthermore, this invention clarifies that input rhythm is only one of the auxiliary signals for word segmentation and confidence determination, not the sole basis. When the text to be recognized lacks input rhythm data, such as text content pasted or copied by the user, historical dialogues imported in batches from a knowledge base or chat history, and messages forwarded between systems, the acquisition unit does not acquire input rhythm. Instead, word segmentation is performed using a statistical language model-based method: an n-gram language model or a pre-trained word segmentation model based on character annotation is used to segment the text, with the point of maximum inter-character transition probability as the word boundary, and maximum forward or backward matching is performed with an abbreviation database to prioritize the segmentation of known abbreviations. When both input rhythm and language model signals are available, the boundary where both are consistent is taken as the final word segmentation result; when they are inconsistent, the language model result takes precedence, and the input rhythm is used as a confidence correction. Therefore, this invention is applicable to real-time typing input scenarios, as well as various customer service text scenarios without input rhythm data, such as pasting, copying, and historical import.

[0021] On the other hand, the present invention also provides an automatic text abbreviation recognition system for intelligent customer service, comprising: The storage unit is used to store text data, including an abbreviation database for storing abbreviations and their definitions, a word database for storing characters and words, and a text database for storing input text. The acquisition unit, which is connected to the storage unit, is used to acquire the text and rhythm of the input within a preset time period to form a corresponding text record, and to divide the sentences in the text into single characters and words; The retrieval unit, which is connected to the storage unit and the acquisition unit respectively, is used to sort the segmented characters and words in descending order according to their frequency of occurrence in the text, remove duplicate characters and / or words with the same meaning, and then retrieve the corresponding meanings for each character or word. After sorting and deduplication, the characters or words are marked. When there is a matching meaning in the abbreviation database, it is marked as the first abbreviation; when there is no matching meaning in the abbreviation database, it is marked as the second abbreviation. An extension unit, connected to the retrieval unit, is used to determine the central word of the corresponding text based on the position of each character or word marked as the second abbreviation, and to perform extended association based on the composition of the central word to obtain a corresponding number of extended definitions; and to perform extended association on a single abbreviation divided from the second abbreviation to obtain several extended definitions corresponding to the single abbreviation. An identification unit, which is connected to the storage unit, the retrieval unit and the extension unit respectively, is used to identify the meaning of the first abbreviation and the extended meaning of the second abbreviation; A verification unit, which is connected to the identification unit and the storage unit respectively, is used to verify whether each of the extended interpretations and abbreviations matches; The central control unit, which is connected to the storage unit, the acquisition unit, the retrieval unit, the extension unit, the identification unit, and the verification unit, is used to determine whether the definition of the abbreviation conforms to a preset extension standard based on the number of obtained extended definitions. If it does not conform to the preset extension standard, the definition of the abbreviation is re-determined based on the extended definitions corresponding to a single abbreviation. If it conforms to the preset extension standard, the verification unit is controlled to verify whether each extended definition matches the abbreviation. Alternatively, the extended definitions are screened based on the domain, and the extended definitions corresponding to the domain are selected. The verification unit is then controlled to verify whether each extended definition matches the abbreviation. An output unit, connected to the central control unit, is used to derive the first abbreviation and the second abbreviation and form an abbreviation lookup table, the lookup table including the abbreviation, the original word of the abbreviation, and the definition of the abbreviation; The preset expansion standard is whether the number of expanded interpretations obtained after expanding the association for the second abbreviation reaches a preset expansion number threshold. When the number of extended definitions obtained is greater than or equal to the preset extension number threshold, the definition of the abbreviation is determined to meet the preset extension standard. When the number of extended definitions obtained is less than the preset extension number threshold, it is determined that the definition of the abbreviation does not meet the preset extension standard, and the abbreviation method of the abbreviation is determined to be morpheme abbreviation.

[0022] In particular, the system further includes a processor and a memory, the memory storing a computer program, which, when executed by the processor, implements the steps of the method as described in any one of claims 1 to 8.

[0023] The beneficial effects of this invention are that by collecting the text and rhythm of input within a preset time period to form corresponding text records, and dividing sentences to obtain individual characters and words, it can completely capture the contextual information in customer service interactions, providing a sufficient data foundation for subsequent abbreviation recognition. By sorting individual characters and words in descending order of frequency of occurrence and then performing definition matching and deduplication, high-frequency words are prioritized, effectively improving retrieval efficiency and reducing redundant calculations.

[0024] In particular, abbreviations are categorized into primary and secondary abbreviations for differentiated processing. Primary abbreviations with matching definitions in the abbreviation database are directly marked, while secondary abbreviations without matching definitions enter an adaptive recognition process. This approach balances rapid response to standardized abbreviations with intelligent derivation capabilities for non-standardized abbreviations. By determining the central word of the corresponding text based on the position of a single character or word, and expanding associations based on the composition of the central word, the direction of expanded definition generation is constrained by the semantics of the central word, avoiding divergent interpretations and significantly improving the accuracy and relevance of expanded associations.

[0025] In particular, by setting a preset threshold for the number of extended meanings, the system determines whether an abbreviation meets the preset expansion criteria based on the number of extended meanings, thus achieving automatic classification and recognition of abbreviation methods. When the number of extended meanings is small, it is identified as a morpheme abbreviation and further broken down into individual abbreviations. By obtaining the extended meanings corresponding to each abbreviation and performing adjacent matching and splicing based on semantic coherence, the meaning of the abbreviation is reconstructed, effectively solving the technical problem that morpheme-truncation abbreviations are difficult to interpret through overall matching. When the number of extended meanings is large, context matching verification is performed by replacing the extended meanings back with the original text to ensure that the final selected meaning is consistent with the original context.

[0026] In particular, by introducing a domain screening mechanism, keywords are extracted from extended definitions, and the number of extended definitions in each domain is counted. The domain with the most extended definitions is selected as the domain to which the abbreviation belongs. This leverages domain consistency constraints to accurately screen extended definitions, effectively reducing the risk of misidentification caused by cross-domain ambiguity. During the verification process, the rationality of the definition is objectively evaluated by calculating the matching degree between the replaced statement and the preceding and following text. When the matching degree is insufficient, extended definitions are reselected or the abbreviation is determined to contain multiple meanings. This achieves a deep understanding of polysemous abbreviations and error-tolerant handling of erroneous abbreviations.

[0027] In particular, the final exported abbreviation lookup table includes abbreviations, their original characters, and their definitions, forming a complete abbreviation knowledge mapping that can be directly accessed by subsequent customer service systems, improving the semantic understanding accuracy and response efficiency of intelligent customer service in long text interaction scenarios. Furthermore, this invention expands and associates unrecognized second abbreviations in the text to obtain a corresponding number of expanded definitions, and determines whether the abbreviation's definition meets a preset expansion standard based on the number of expanded definitions, thereby solving the problem of defining abbreviations in text that cannot be recorded in a database.

[0028] In particular, when an abbreviation cannot be parsed using the core word-based structure, the central control module determines it to be a morpheme abbreviation. It then further breaks down the abbreviation that cannot be extended in terms of meaning and uses a morpheme extension association model to extend and associate it to obtain several extended meanings corresponding to a single abbreviation, thus providing a method for interpreting abbreviations that cannot be parsed using the core word-based structure.

[0029] In particular, the present invention recombines the definitions of abbreviations after splitting them, and recombines them by taking the most matching extended definitions corresponding to adjacent abbreviations as matching extended definitions, thereby accurately matching the definition of the recombined abbreviation.

[0030] In particular, the present invention ensures the determination of the abbreviation's meaning by replacing the corresponding abbreviation in the text with the extended meaning corresponding to the abbreviation, and verifying the degree of matching between the extended meaning and the text before and after the corresponding abbreviation.

[0031] In particular, the invention accurately determines whether the abbreviation's meaning matches by using the matching degree, and provides corresponding solutions when they do not match, thereby ensuring the accurate recognition of abbreviations.

[0032] In particular, when there are too many extended definitions, this invention ensures the efficiency of abbreviation recognition by dividing the extended definitions into domains and then selecting the extended definitions from one domain to verify their matching degree with the text.

[0033] In particular, the present invention determines the selection of the corresponding field by the number of extended interpretations within the field, thereby ensuring the efficiency of abbreviation recognition.

[0034] Furthermore, this invention provides a storage unit for storing text data, a collection unit for collecting and segmenting text, a collection unit for retrieving and labeling text, an extension unit for obtaining extended definitions, an identification unit for identifying extended definitions, a verification unit for verifying whether each of the extended definitions matches the abbreviation, a central control unit for determining whether the definition of the abbreviation conforms to a preset extension standard based on the number of obtained extended definitions, and an output unit for outputting the abbreviation mapping. This overcomes the problem in the prior art where the verification of the abbreviation's definition in the original text is insufficient, leading to insufficient recognition accuracy in text abbreviation recognition. Attached Figure Description

[0035] Figure 1 A flowchart of the automatic text abbreviation recognition method for the intelligent customer service of the present invention; Figure 2 This is a structural diagram of the intelligent customer service text abbreviation automatic recognition system of the present invention; Figure 3 This is a flowchart illustrating the processing and verification of abbreviations in the second embodiment of the present invention. Figure 4 This is a schematic diagram of a real estate website application scenario according to an embodiment of the present invention. Detailed Implementation

[0036] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0037] It should be noted that the data in this embodiment are all derived from the comprehensive analysis and evaluation of historical recognition data and corresponding historical recognition results of the central control unit of the present invention in the three months prior to this recognition. Before this recognition, the central control unit of the present invention determines the values ​​of various preset parameter standards for this recognition based on the processing of 101,420 instances of text containing abbreviations accumulated in the previous three months. Those skilled in the art will understand that the system of the present invention can determine the above-mentioned parameters for a single item in the following ways: selecting the value with the highest proportion according to the data distribution as the preset extended standard parameter; using weighted summation to use the obtained value as the preset extended standard parameter; substituting each historical data into a specific formula and using the value obtained by the formula as the preset extended standard parameter; or other selection methods, as long as the system of the present invention can clearly define different specific situations in the single-item judgment process through the obtained values.

[0038] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0039] Please see Figure 1 The flowchart shown is a method for automatically recognizing text abbreviations in intelligent customer service according to the present invention. This method is used to identify and merge long text information in customer service interactions, including: Collect the text and input rhythm within a preset time period to form a corresponding text record; Segmenting sentences in a text to obtain corresponding single characters and words; The segmented characters and words are sorted in descending order according to their frequency of occurrence in the text. Retrieve from the storage database, search for each sorted single character and / or word to match its corresponding definition, and remove duplicate single characters and / or words with the same definition; After sorting and deduplication, mark individual words or phrases, including: When a matching definition is found in the abbreviation database, it is marked as the first abbreviation. When there is no matching definition in the abbreviation database, it is marked as a second abbreviation; For each character or word marked as the second abbreviation, the central word of the corresponding text is determined according to the position of the character or word, and the second abbreviation is expanded and associated according to the formation of the central word to obtain the corresponding number of extended interpretations; The definition of an abbreviation is determined based on the number of extended definitions obtained, indicating whether the definition meets the preset extension criteria. In response to non-compliance with the preset extension criteria, the meaning of the abbreviation is redefined based on the extended meaning corresponding to a single abbreviation. In response to conforming to preset expansion criteria, verify whether each expanded definition matches the abbreviation, or... After filtering the extended definitions based on the domain, the extended definitions corresponding to the domain are selected, and it is verified whether each extended definition matches the abbreviation. Derive the first and second abbreviations and form an abbreviation mapping. The mapping table includes the abbreviation, the original word of the abbreviation, and the definition of the abbreviation. The storage library includes an abbreviation database, a word database, and a text database.

[0040] By collecting the input text and its rhythm within a preset time period to form corresponding text records, and segmenting sentences to obtain individual characters and words, the system can fully capture the contextual information in customer service interactions, providing a sufficient data foundation for subsequent abbreviation recognition. After sorting individual characters and words in descending order of frequency, definition matching and deduplication are performed, prioritizing high-frequency words, which effectively improves retrieval efficiency and reduces redundant computation.

[0041] In one embodiment, the step of determining whether the definition of an abbreviation conforms to a preset expansion criterion based on the number of extended definitions includes: When the number of extended definitions is less than a preset threshold, the abbreviation is determined to be a morpheme abbreviation, and the abbreviation of the morpheme abbreviation is divided into several abbreviations, and... Obtain the extended meaning corresponding to a single abbreviation, and redefine the meaning of the abbreviation based on the extended meaning corresponding to the single abbreviation; When the number of extended definitions is greater than or equal to the preset threshold for the number of extended definitions, the corresponding extended definitions in the text are replaced in turn to verify whether each extended definition matches the abbreviation. Alternatively, keywords can be extracted sequentially from the extended definitions corresponding to the abbreviation. Based on the domain to which the definition containing the keyword belongs, the extended definitions are filtered by domain. Then, the extended definitions in the corresponding domains are selected to replace the corresponding abbreviations in the text in turn to verify whether each extended definition matches the abbreviation. The preset expansion standard is whether the number of expanded meanings obtained after expanding associations for the second abbreviation reaches the preset expansion number threshold. When the number of extended definitions obtained is greater than or equal to the preset extension number threshold, the definition of the abbreviation is determined to meet the preset extension standard. When the number of extended definitions obtained is less than the preset extension number threshold, it is determined that the definition of the abbreviation does not meet the preset extension standard, and the abbreviation method of the abbreviation is determined to be morpheme abbreviation.

[0042] By replacing the corresponding abbreviation in the text with its extended definition and verifying the degree of matching between the extended definition and the text before and after the abbreviation, the definition of the abbreviation is determined.

[0043] Abbreviations are categorized into primary and secondary abbreviations for differentiated processing. Primary abbreviations with matching definitions in the abbreviation database are directly marked, while secondary abbreviations without matching definitions enter an adaptive recognition process. This approach balances rapid response to standardized abbreviations with intelligent derivation capabilities for non-standardized abbreviations. By determining the central word of the corresponding text based on the position of a single character or word, and expanding associations based on the composition of the central word, the direction of expanded definition generation is semantically constrained by the central word, avoiding divergent definitions and significantly improving the accuracy and relevance of expanded associations.

[0044] In one embodiment, obtaining the extended meaning corresponding to a single abbreviation includes: For a single abbreviation, based on the way the central word is constructed, the abbreviation is expanded and associated to obtain several extended meanings corresponding to the single abbreviation; When an extended meaning cannot be obtained by expanding association for a single abbreviation, the original character of the abbreviation is retained and used as the extended meaning corresponding to the abbreviation.

[0045] In one embodiment, the step of redefining the meaning of an abbreviation based on the extended meaning corresponding to the individual abbreviation includes: Among the extended definitions corresponding to its adjacent abbreviations; The extended definition that best matches the extended definition of the current abbreviation in terms of semantic coherence is taken as the matching extended definition of the adjacent abbreviation, and the matching extended definitions of each abbreviation are concatenated in sequence to reconstruct the definition of the abbreviation.

[0046] By setting a preset threshold for the number of extended meanings, the system determines whether an abbreviation meets the preset expansion criteria based on the number of extended meanings, thus achieving automatic classification and recognition of abbreviation methods. When the number of extended meanings is small, it is identified as a morpheme abbreviation and further broken down into individual abbreviations. By obtaining the extended meanings corresponding to each abbreviation and performing adjacent matching and splicing based on semantic coherence, the meaning of the abbreviation is reconstructed, effectively solving the technical problem that morpheme-truncation abbreviations are difficult to interpret through overall matching. When the number of extended meanings is large, context matching verification is performed by replacing the extended meanings back with the original text to ensure that the final selected meaning is consistent with the original context.

[0047] Please see Figure 2 As shown, it is a structural diagram of the intelligent customer service text abbreviation automatic recognition system of the present invention, including: The storage unit is used to store text data, including an abbreviation database for storing abbreviations and their definitions, a word database for storing characters and words, and a text database for storing input text. The acquisition unit, which is connected to the storage unit, is used to acquire the input text and the input rhythm within a preset time to form a corresponding text record, and to divide the sentences in the text into single characters and words; The retrieval unit, which is connected to the storage unit and the acquisition unit respectively, is used to sort the segmented characters and words in descending order according to their frequency of occurrence in the text, remove duplicate characters and / or words with the same meaning, and then retrieve the corresponding meanings one by one. After sorting and deduplication, the characters or words are marked. When there is a matching meaning in the abbreviation database, it is marked as the first abbreviation; when there is no matching meaning in the abbreviation database, it is marked as the second abbreviation. An extension unit, connected to the retrieval unit, is used to determine the central word of the corresponding text based on the position of each character or word marked as the second abbreviation, and to perform extended association based on the composition of the central word to obtain a corresponding number of extended definitions; and to perform extended association on a single abbreviation divided from the second abbreviation to obtain several extended definitions corresponding to that single abbreviation. The identification unit, which is connected to the storage unit, the retrieval unit and the extension unit respectively, is used to identify the definition of the first abbreviation and the extended definition of the second abbreviation; The verification unit, which is connected to the recognition unit and the storage unit respectively, is used to verify whether each extended definition matches the abbreviation; The central control unit, which is connected to the storage unit, acquisition unit, retrieval unit, extension unit, identification unit, and verification unit, is used to determine whether the definition of the abbreviation conforms to the preset extension standard based on the number of extended definitions obtained. If it does not conform to the preset extension standard, the definition of the abbreviation is re-determined based on the extended definition corresponding to a single abbreviation. If it conforms to the preset extension standard, the verification unit is controlled to verify whether each extended definition matches the abbreviation. Alternatively, the extended definitions are screened based on the domain, and the extended definitions corresponding to the domain are selected. The verification unit is then controlled to verify whether each extended definition matches the abbreviation. The output unit, which is connected to the central control unit, is used to derive the first abbreviation and the second abbreviation and form an abbreviation lookup table. The lookup table includes the abbreviation, the original word of the abbreviation, and the definition of the abbreviation. The system also includes a processor and memory, with the memory storing computer programs.

[0048] The preset expansion standard is whether the number of expanded meanings obtained after expanding associations for the second abbreviation reaches the preset expansion number threshold. When the number of extended definitions obtained is greater than or equal to the preset extension number threshold, the definition of the abbreviation is determined to meet the preset extension standard. When the number of extended definitions obtained is less than the preset extension number threshold, it is determined that the definition of the abbreviation does not meet the preset extension standard, and the abbreviation method of the abbreviation is determined to be morpheme abbreviation.

[0049] Furthermore, by setting up a storage unit for storing text data, a collection unit for collecting and segmenting text, a collection unit for retrieving and labeling text, an extension unit for obtaining extended definitions, an identification unit for identifying extended definitions, a verification unit for verifying whether each of the extended definitions matches the abbreviation, a central control unit for determining whether the definition of the abbreviation conforms to a preset extension standard based on the number of obtained extended definitions, and an output unit for outputting the abbreviation mapping, the problem of insufficient verification of the definition of abbreviation in the original text in the prior art, resulting in insufficient recognition accuracy of text abbreviation recognition, is overcome.

[0050] Please see Figure 3 As shown, this is a flowchart of the processing and verification of the second abbreviation in an embodiment of the present invention. After recombining the interpretation of the abbreviation, the method further includes: The reconstructed definition is used as an extended definition to replace the abbreviation in the corresponding position in the text. The degree of matching between the replaced statement and the text before and after the abbreviation is verified to obtain the matching degree. Matching degree refers to the quantitative value of the degree to which the sentence formed after expanding the definition and replacing the abbreviation is semantically consistent with the text before and after the abbreviation.

[0051] In one embodiment, after obtaining the matching degree, the meaning of the reconstructed abbreviation is determined based on the matching degree to whether it conforms to the preset expansion standard. When the abbreviation does not meet the preset expansion criteria, it is rematched with another extended definition corresponding to its adjacent abbreviation to form a definition again; or, it is determined that the abbreviation contains multiple meanings, and the meaning of the abbreviation is re-determined by combining the abbreviation with the text before and after it.

[0052] By introducing a domain-based screening mechanism, keywords are extracted from extended definitions, and the number of extended definitions in each domain is counted. The domain with the most extended definitions is selected as the domain to which the abbreviation belongs. This leverages domain consistency constraints to accurately screen extended definitions, effectively reducing the risk of misidentification caused by cross-domain ambiguity. During the verification process, the rationality of the definition is objectively evaluated by calculating the matching degree between the replaced statement and the preceding and following text. When the matching degree is insufficient, an extended definition is reselected or the abbreviation is determined to contain multiple meanings. This achieves a deep understanding of polysemous abbreviations and error-tolerant handling of erroneous abbreviations.

[0053] The following uses the real estate sector as an example, and provides specific embodiments in conjunction with the accompanying drawings to further illustrate the implementation details of the present invention.

[0054] Example of financial customer service: When a user inputs "Is this investment product fixed income or net asset value (NAV)?", "fixed income" and "NAV" are marked as the second abbreviation. For "fixed income", step S2 determines the core word as "income", step S3 matches "morpheme extraction", that is, "fixed" is extracted from "fixed" and "income" is extracted from "returns", and step S4 expands to obtain definitions such as "fixed income". For "NAV", step S3 matches "modifier plus core word", step S4 obtains definitions such as "unit NAV" and "NAV type", and step S5 combines financial domain tags to confirm that "NAV" refers to "NAV type product". Those skilled in the art can extend this method to any customer service text scenario with abbreviations, such as medical and telecommunications.

[0055] E-commerce customer service example: A user inputs "Is this down jacket eligible for a discount?" The data collection unit identifies "discount" as the second abbreviation. Based on steps S1 and S2, it is determined that "discount" is located in the predicate position, with "discount" as the central word. Step S3 matches "parallel abbreviations," where "full" means "full amount" and "discount" means "reduction." Step S4 expands the association to obtain extended meanings such as "discount upon reaching a certain amount" and "discount offer." Step S5 combines e-commerce domain tags to confirm its meaning as "a discount is offered after meeting a specified amount." This process uses the exact same algorithm framework as the real estate domain, differing only in the pattern library and dictionary content.

[0056] To illustrate the universality of the present method, in addition to the real estate field, several application examples in other fields are supplemented below to show that the core word extraction and extended association method can be generalized to various intelligent customer service business scenarios: S5. Unified processing of multi-ambiguity and cross-domain: when the core word identified in step S2 has multiple senses, or the extended definitions generated in step S4 belong to multiple fields, disambiguation is performed according to the following unified logic: extract the context window of the second abbreviation, that is, N words before and after, and calculate the domain attribution probability of each candidate definition in combination with the domain labels marked in the word database; select the definition set with the highest domain attribution probability for subsequent verification; when the probabilities of multiple domains are similar, that is, the difference is less than the preset ambiguity threshold, retain the multi-domain candidates and further screen them in the verification stage in combination with the matching degree. The above disambiguation logic adopts the same set of algorithm framework for all business fields, and only the domain labels and dictionary contents are configured with different fields, so as to realize unified processing of cross-industry vocabularies.

[0057] S4. Extended definition generation: according to the core word determined in step S2 and the composition manner obtained in step S3, call the word database to perform extended association on the second abbreviation, or on the single split abbreviated character in the case of morpheme abbreviation, to generate a plurality of candidate extended definitions; the extended association can be implemented based on word co-occurrence statistics of the word database, synonym and hypernym-hyponym dictionary query, and mask filling of a pre-trained language model.

[0058] S3. Core word composition manner matching: the system maintains a cross-domain pattern library of abbreviation composition manners, which describes the composition of abbreviations with general linguistic rules, and at least includes: morpheme interception type, that is, retaining part of the morphemes of the original word, such as "tong" (concrete) intercepted from "hunningtu" (concrete), and "jianshi" (construction drawing) intercepted from "jianzhu shigong" (construction); initial letter or pinyin abbreviation type, such as "BIM" and "PC"; modifier plus core word type, such as "quanming nan liangfang" (full-brightness south-facing two-bedroom) composed of a modifier and the core word "fang" (room); number or symbol designation type, such as "3D" and "shuangnan" (two south-facing orientations); parallel abbreviation type, that is, taking the first character of each parallel component, such as "Jing-Jin-Ji" (Beijing-Tianjin-Hebei). Match the second abbreviation with the above patterns one by one at the character granularity to obtain its composition manner and core word position.

[0059] S2. Core word candidate determination: according to the syntactic role and head node obtained in step S1, core word candidates are determined according to the following domain-independent general rules: (a) When the second abbreviation is the object of a predicate verb, the core word candidate is the core noun of the object; if there are other words in the sentence that have a parallel dependency relationship with it, the core word candidate is the hypernym category word to which the parallel word belongs; (b) When the second abbreviation is an attribute modifying a certain core noun, the core word candidate is the core noun it modifies; (c) When the second abbreviation is a syntactic core, such as a one-word sentence or title-style text, the candidate headword is the abbreviation itself. The above rules are based on general syntactic theory and are applicable to texts in any field.

[0060] S1. Syntactic role recognition: Perform word segmentation and part-of-speech tagging on the sentence containing the second abbreviation, and determine the syntactic role of the second abbreviation in the sentence (such as subject, predicate, object, attributive, parallel component) and its dependent head node based on dependency syntax parsing. The present invention directly invokes existing dependency syntax parsing tools for implementation, and this step belongs to a conventional natural language processing means, on which the present invention makes no improvement.

[0061] To further illustrate that the present invention has universality for any business field, a general implementation manner of headword extraction and extended association is given below. This manner does not rely on specific domain knowledge, and can be applied to various customer service text scenarios with abbreviations such as e-commerce, finance, medical care, and telecommunications. The domain-related differences thereof are only reflected in the configuration of the word database and domain tags.

[0062] General Implementation Manner of Headword Extraction and Extended Association Embodiment 1: This embodiment is applied to the online customer service scenario of a real estate website, please refer to Figure 4 which is a schematic diagram of the real estate website application scenario according to the embodiment of the present invention: A potential home buyer consults the customer service about housing resources through a dialog box, and inputs the following text: "Please help me check whether there is a fully lit south-facing two-bedroom apartment that meets the 'full five and unique' requirement for sale in this community." The collection unit collects the text input by the user within a preset duration, and divides the text into single characters and words according to the pause intervals and sending rhythm during the user's typing process, so as to obtain units such as "please help", "me check", "a moment", "this", "community", "whether there is", "full five and unique", "of", "fully lit south-facing two-bedroom apartment", and "for sale".

[0063] The retrieval unit sorts the divided single characters and words in descending order according to the frequency of occurrence in the text, where both "full five and unique" and "fully lit south-facing two-bedroom apartment" appear only once, and belong to low-frequency specific phrases.

[0064] At this time, after deduplicating the units with the same definition, the retrieval unit retrieves the abbreviation database in the storage unit one by one for matching.

[0065] Upon retrieval, "full five and unique" has a corresponding definition in the abbreviation database that "the house property right is full five years and it is the only house under the owner's name", so it is marked as the first abbreviation.

[0066] However, "fully lit south-facing two-bedroom apartment" has no corresponding record in the abbreviation database, so it is marked as the second abbreviation.

[0067] For "Quan Ming Nan Liang Fang" marked as the second abbreviation, the expansion unit determines that the core word is "Fang (house)" according to its object position in the text.

[0068] The common composition pattern of this core word in the real estate field is "orientation + house type + house" or "feature + orientation + quantity + house".

[0069] The expansion unit performs expansion association on "Quan Ming Nan Liang Fang" according to this composition pattern, and generates a total of five expanded definitions: "Quan Ming Nan Xiang Liang Shi Fang", "Quan Ming Nan Xiang Liang Ju Fang", "Quan Ming Chao Nan Liang Fang", "Quan Ming Shuang Nan Liang Fang", and "Quan Ming Nan Wo Liang Fang".

[0070] After acquiring these expanded definitions, the central control unit compares the number of expanded definitions with a preset expansion quantity threshold. In this embodiment, the preset expansion quantity threshold is three. The five currently obtained expanded definitions are greater than the threshold, so it is determined that the definitions of this abbreviation meet the preset expansion standard.

[0071] Since the number of expanded definitions is relatively large, the central control unit decides to perform domain screening on the expanded definitions first before executing verification. The identification unit extracts keywords from each expanded definition, such as "southward", "two rooms", "two bedrooms", "facing south", "south bedroom", etc., and counts the affiliated domain of each expanded definition according to the domain categories marked in the word database of the storage unit.

[0072] So far, the statistical result shows that four of the five expanded definitions belong to the real estate field, and one may belong to the general orientation description field.

[0073] The central control unit selects the real estate field with the largest number of expanded definitions as the domain to which this abbreviation belongs, and retains the four expanded definitions under the real estate field, namely "Quan Ming Nan Xiang Liang Shi Fang", "Quan Ming Nan Xiang Liang Ju Fang", "Quan Ming Chao Nan Liang Fang", and "Quan Ming Shuang Nan Liang Fang".

[0074] The verification unit sequentially replaces the four expanded definitions back to the corresponding positions in the original text, and the replaced sentences are respectively "Is there a Quan Ming Nan Xiang Liang Shi Fang that is full five years and unique for sale", "Is there a Quan Ming Nan Xiang Liang Ju Fang that is full five years and unique for sale", etc. The verification unit calculates the matching degree between each replaced sentence and the text before and after the abbreviation, that is, the quantitative value of semantic coherence.

[0075] Among them, "Quan Ming Nan Xiang Liang Shi Fang" and "Quan Ming Nan Xiang Liang Ju Fang" connect naturally with the context, with the highest and similar matching degrees. The central control unit selects "Quan Ming Nan Xiang Liang Shi Fang", which has the highest matching degree and uses terms more in line with the habits of the real estate industry, as the final definition.

[0076] The output unit exports the first abbreviation "man wu wei yi" and its definition, as well as the second abbreviation "quan ming nan liang fang" and its final definition "fully-lit south-facing two-bedroom apartment" together to form an abbreviation comparison table. The comparison table clearly records the abbreviation "quan ming nan liang fang", the original full term of the abbreviation "fully-lit south-facing two-bedroom apartment" and the definition, and this mapping is subsequently used in the automatic semantic understanding module of the customer service system, which improves the recognition speed of abbreviations in subsequent similar property brochure descriptions.

[0077] In one embodiment, after determining whether the preset expansion criterion is met based on the matching degree, when the preset expansion criterion is not met, the following steps are executed: Replace the corresponding abbreviation in the text with the next expanded definition corresponding to the abbreviation, and verify again whether the expanded definition matches the abbreviation, or Determine that the abbreviation is erroneous, and export the erroneous abbreviation.

[0078] When the definition of an abbreviation cannot be parsed through the central word-based composition method, the central control module determines that it is a morpheme abbreviation, so as to further split the abbreviation that cannot be expanded for definition, and perform expansion association on it using a morpheme expansion association model to obtain several expanded definitions corresponding to a single abbreviated character, which provides a method for defining abbreviations that cannot be parsed through the central word-based composition method.

[0079] The finally exported abbreviation comparison table includes abbreviations, original full terms of the abbreviations and definitions of the abbreviations, forming a complete abbreviation knowledge mapping that can be directly called by the subsequent customer service system, which improves the semantic understanding accuracy and response efficiency of the intelligent customer service in long text interaction scenarios. Further, the present invention performs expansion association on the unrecognized second abbreviation in the text to obtain a corresponding number of expanded definitions, and determines whether the definition of the abbreviation meets the preset expansion criterion according to the number of expanded definitions, thereby solving the problem of defining abbreviations that are not recorded in the database in the text.

[0080] The matching degree is used to accurately determine whether the abbreviation definition matches, and when there is no match, the present invention provides a corresponding solution, thereby ensuring accurate recognition of abbreviations.

[0081] When the number of expanded definitions is too large, the expanded definitions are divided into domains, and the expanded definitions of one domain are selected preferentially to verify the matching degree with the text item by item, thereby ensuring the efficiency of abbreviation recognition.

[0082] In one embodiment, when extracting keywords from the expanded definitions for domain screening, selecting the expanded definitions of the corresponding domain includes: Count the number of expanded definitions under each domain respectively, select the domain with the largest number of expanded definitions as the domain to which the abbreviation belongs, and use the expanded definitions under this domain for subsequent matching verification.

[0083] The abbreviations are split for interpretation and then recombined, and meanwhile, the most matching expanded interpretation corresponding to the adjacent abbreviated character of a single abbreviated character is taken as a matching expanded interpretation for recombination, so as to accurately match the interpretation of the recombined abbreviation.

[0084] Embodiment 2: This embodiment is based on Embodiment 1, and also occurs in the customer service conversation of a real estate website. After Embodiment 1 is completed, the text input by the user is: "Excuse me, are the penthouse duplex and the ground-floor duplex of the garden villa still on sale?".

[0085] The acquisition unit acquires the text and the input rhythm, and divides words according to natural sentence breaks and pauses to obtain units such as "Excuse me", "this property", "still have", "penthouse duplex", "and", "ground-floor duplex of the garden villa", "on sale?" and the like.

[0086] After the retrieval unit sorts by frequency and deduplicates units with the same interpretation, it matches them with the abbreviation database one by one. The database stores the common real estate term "ground-floor duplex of the garden villa", whose interpretation is "ground-floor duplex unit of a garden villa", so it is marked as the first abbreviation.

[0087] However, "penthouse duplex" does not hit any entry in the abbreviation database, and is marked as the second abbreviation.

[0088] The expansion unit processes "penthouse duplex", and determines that it belongs to the same architectural house type category as "ground-floor duplex of the garden villa" according to its position as a coordinate object in the sentence, and further determines that the core word is "duplex", which is the abbreviation of "duplex unit". The composition mode of the core word is expressed as "position + house type".

[0089] The expansion unit attempts to perform overall expansion association on "penthouse duplex" according to this composition mode, but only obtains two expanded interpretations of "top-floor duplex" and "top composite". The central control unit compares the number of expanded interpretations with the preset expansion number threshold of 3, and the number 2 is less than the threshold, so it is determined that the interpretation of this abbreviation does not meet the preset expansion standard, and it is also determined that its abbreviation mode is morpheme abbreviation.

[0090] The expansion unit then splits "penthouse duplex" into single abbreviated characters "top" and "duplex". For the abbreviated character "top", according to the composition mode of the core word, three expanded interpretations of "top floor", "top part" and "top apartment" are obtained through expansion association; for the abbreviated character "duplex", three expanded interpretations of "duplex", "composite" and "duplex floor" are obtained through expansion association.

[0091] When expanding a single abbreviated character, one interpretation "composite" of "duplex" has a weak correlation with the architectural house type, but the expansion unit still retains it for subsequent screening.

[0092] For any abbreviation, the central control unit searches for the combination that best matches the expanded definition of the current abbreviation in terms of semantic coherence among the expanded definitions corresponding to adjacent abbreviations. Specifically, based on the expanded definition of "Ding (Top)", it is sequentially spliced with the expanded definitions of "Fu (Duplex)". In the semantic coherence evaluation, the spliced combination "Top Duplex" of "Top Floor" and "Duplex Style" obtains the highest semantic coherence score and becomes the matched expanded definition. Similarly, the combination of "Top Floor Apartment" and "Duplex Style" also has a certain degree of matching, but it is lower than the former.

[0093] Finally, the central control unit sequentially splices the matched expanded definitions of each abbreviation to re-form the definition of this abbreviation as "Top Duplex".

[0094] The verification unit replaces "Ding Fu" in the original text with the re-formed definition "Top Duplex" to form a replacement sentence "Excuse me, are there still top duplexes and garden villa ground-floor duplex units for sale in this property", verifies the matching degree of this sentence with the text before and after the abbreviation, and obtains the matching degree.

[0095] The matching degree reaches the preset standard, and the definition is confirmed and adopted.

[0096] The output unit finally exports the abbreviation comparison table for this session, including the first abbreviation "Garden Villa Ground-floor Duplex Unit" and its definition "Garden villa ground-floor duplex apartment type", as well as the original words "Top Duplex" and its definition for the second abbreviation "Ding Fu", for subsequent customer service intent recognition and knowledge base retrieval.

[0097] For the input rhythm, the specific quantitative rules and weight calculation system for obtaining corresponding single characters and words by dividing the sentences in the text, as well as the backup word segmentation means when input rhythm data is lacking, all refer to the previous quantitative description of input rhythm in the specification of this application, which will not be repeated here; the application thereof will be described below in combination with specific input behavior examples.

[0098] Taking the first embodiment as an example: User A has obvious pauses before and after "Ding Fu" (1 second before, 2 seconds after), indicating that "Ding Fu" is an independent input unit and should be retrieved as a whole word; This verifies the rationality of the collection unit dividing "Ding Fu" into an independent word, avoiding being incorrectly split into "Ding" and "Fu" and then mixed and combined with other characters; Fast input + long pause before sending: User A inputs "Ding Fu" extremely fast (completed within 2 seconds), but pauses for 2 seconds after input before continuing. This implies that the user is proficient in using the term "Ding Fu" (it may be an industry conventional abbreviation), and the pause is to confirm whether the term is accurate or wait to see if the system can understand it; Accordingly, the system increases the confidence level of "Ding Fu" as an abbreviation, and preferentially performs exact matching in the abbreviation database instead of directly treating it as a wrong character; When multiple second abbreviation candidates exist, the input rhythm can help determine the processing priority: Abbreviation candidates with the characteristics of "rapid input + pause" are usually industry abbreviations that users intend to use, and should be given priority for expanded association; Candidates entered at normal speed without obvious pauses may only be temporarily omitted and have lower priority. If a user pauses for a long time after entering an abbreviation, it indicates that the user is unsure about the meaning of the term. When verifying extended definitions, the system can appropriately lower the matching threshold and retain more candidate definitions for confirmation in subsequent interactions. If a user enters an abbreviation fluently and without pause, it indicates that the user is confident that the meaning of the word is clear. In this case, the system should select the most common and authoritative definition as the matching result. If a word is marked as a second abbreviation and the input rhythm involves repeated modifications and multiple pauses, the user may have doubts about the accuracy of the abbreviation. If the system still cannot obtain a matching definition after expanding the association, it can increase the weight of judging it as an "incorrect abbreviation" based on the rhythm feature, and promptly export a prompt to customer service for manual confirmation.

[0099] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0100] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for automatic recognition of text abbreviations in intelligent customer service, used to identify and merge long text information in customer service interactions, including: Collect the text and input rhythm within a preset time period to form a corresponding text record; The sentences in the text are segmented to obtain the corresponding single characters and words; The segmented characters and words are sorted in descending order according to their frequency of occurrence in the text; Retrieve from the storage database, search for each sorted single character and / or word to match its corresponding definition, and remove duplicate single characters and / or words with the same definition; Its characteristic is that, after sorting and deduplication, the single character or word is marked, including, When a matching definition is found in the abbreviation database, it is marked as the first abbreviation. When there is no matching definition in the abbreviation database, it is marked as a second abbreviation; For each character or word marked as the second abbreviation, the central word of the corresponding text is determined according to the position of the character or word, and the second abbreviation is extended and associated according to the formation of the central word to obtain the corresponding number of extended interpretations; The definition of an abbreviation is determined based on the number of extended definitions obtained, indicating whether the definition meets the preset extension criteria. In response to non-compliance with the preset extension criteria, the meaning of the abbreviation is redefined based on the extended meaning corresponding to a single abbreviation. In response to conforming to preset expansion criteria, verify whether each expanded definition matches the abbreviation, or... After filtering the extended definitions based on the domain, the extended definitions corresponding to the domain are selected, and it is verified whether each extended definition matches the abbreviation. The preset expansion quantity threshold is a pre-set value; the preset expansion standard is whether the number of expanded meanings obtained after the second abbreviation is expanded and associated reaches the preset expansion quantity threshold; when the number of expanded meanings obtained is greater than or equal to the preset expansion quantity threshold, it is determined that the meaning of the abbreviation meets the preset expansion standard; when the number of expanded meanings obtained is less than the preset expansion quantity threshold, it is determined that the meaning of the abbreviation does not meet the preset expansion standard and the abbreviation method of the abbreviation is morpheme abbreviation. The first abbreviation and the second abbreviation are exported and an abbreviation lookup table is formed. The abbreviation lookup table includes the following data items: abbreviation, original word of the abbreviation, and definition of the abbreviation. The storage library includes an abbreviation database, a word database, and a text database.

2. The method for automatic recognition of text abbreviations according to claim 1, characterized in that, The steps for determining whether the definition of an abbreviation meets the preset expansion criteria based on the number of extended definitions include: When the number of extended definitions is less than the preset extension number threshold, the abbreviation method is determined to be morpheme abbreviation, and the morpheme abbreviation is divided into several abbreviations, and, Obtain the extended meaning corresponding to a single abbreviation, and redefine the meaning of the abbreviation based on the extended meaning corresponding to the single abbreviation; When the number of extended definitions is greater than or equal to the preset extension number threshold, the corresponding abbreviation in the text is replaced by the extended definition corresponding to the abbreviation in turn to verify whether each extended definition matches the abbreviation. Alternatively, keywords can be extracted sequentially from the extended definitions corresponding to the abbreviation. Based on the domain to which the definition containing the keyword belongs, the extended definitions are categorized by domain. Then, the extended definitions in the corresponding domains are selected to replace the corresponding abbreviations in the text in turn to verify whether each extended definition matches the abbreviation. The preset expansion standard is whether the number of expanded interpretations obtained after expanding the association for the second abbreviation reaches a preset expansion number threshold. When the number of extended definitions obtained is greater than or equal to the preset extension number threshold, the definition of the abbreviation is determined to meet the preset extension standard. When the number of extended definitions obtained is less than the preset extension number threshold, it is determined that the definition of the abbreviation does not meet the preset extension standard, and the abbreviation method of the abbreviation is determined to be morpheme abbreviation.

3. The automatic text abbreviation recognition method according to claim 2, characterized in that, The process of obtaining the extended meaning corresponding to a single abbreviation includes: For a single abbreviation, based on the composition of the central word, the abbreviation is expanded and associated to obtain several extended meanings corresponding to the single abbreviation; When an extended meaning cannot be obtained by expanding association of a single abbreviation, the original character of the abbreviation is retained and used as the extended meaning corresponding to the abbreviation. The extended meanings of each abbreviation are matched with each other in terms of semantic coherence and then spliced ​​together in order to reconstruct the meaning of the abbreviation.

4. The method for automatic recognition of text abbreviations according to claim 3, characterized in that, For any abbreviation, the step of redetermining the meaning of the abbreviation based on the extended meaning corresponding to the individual abbreviation includes: Among the extended definitions corresponding to its adjacent abbreviations; The extended definition that best matches the extended definition of the current abbreviation in terms of semantic coherence is taken as the matching extended definition of the adjacent abbreviation, and the matching extended definitions of each abbreviation are concatenated in sequence to reconstruct the definition of the abbreviation.

5. The automatic text abbreviation recognition method according to claim 3 or 4, characterized in that, After recombining the meaning of the abbreviation, the method further includes: The reconstructed definition is used as an extended definition to replace the abbreviation at the corresponding position in the text to generate a replacement statement. The degree of matching between the replacement statement and the text before and after the abbreviation is verified to obtain the matching degree. The matching degree refers to the quantitative value of the degree to which the statement formed after expanding the definition and replacing the abbreviation is semantically consistent with the text before and after the abbreviation.

6. The method for automatic recognition of text abbreviations according to claim 5, characterized in that, After obtaining the matching degree, it is determined whether the meaning of the reconstructed abbreviation conforms to the preset expansion standard based on the matching degree; When the abbreviation does not meet the preset extension criteria, the abbreviation is rematched with another extended meaning corresponding to its adjacent abbreviation to form a new meaning; or, the abbreviation is determined to have multiple meanings, and the meaning of the abbreviation is re-determined by combining the abbreviation with the text before and after it.

7. The method for automatic recognition of text abbreviations in intelligent customer service according to claim 5, characterized in that, After determining whether the matching degree meets the preset extension criteria, if it does not meet the preset extension criteria, the following steps are performed: Replace the corresponding abbreviation in the text with the next extended definition corresponding to the abbreviation, and verify again whether the extended definition matches the abbreviation, or... The abbreviation is determined to be incorrect, and the incorrect abbreviation is exported.

8. The method for automatic recognition of text abbreviations according to claim 6 or 7, characterized in that, When extracting keywords from extended definitions for domain segmentation, the selection of extended definitions for the corresponding domain includes: The number of extended definitions in each field is counted separately. The field with the most extended definitions is selected as the field to which the abbreviation belongs, and the extended definitions in that field are used for subsequent matching and verification.

9. An automatic text abbreviation recognition system for intelligent customer service, which applies the method described in any one of claims 1-8, characterized in that, include: The storage unit is used to store text data, including an abbreviation database for storing abbreviations and their definitions, a word database for storing characters and words, and a text database for storing input text. The acquisition unit, which is connected to the storage unit, is used to acquire the text and rhythm of the input within a preset time period to form a corresponding text record, and to divide the sentences in the text into single characters and words; The retrieval unit, which is connected to the storage unit and the acquisition unit respectively, is used to sort the segmented characters and words in descending order according to their frequency of occurrence in the text, remove duplicate characters and / or words with the same meaning, and then retrieve the corresponding meanings for each character or word. After sorting and deduplication, the characters or words are marked. When there is a matching meaning in the abbreviation database, it is marked as the first abbreviation; when there is no matching meaning in the abbreviation database, it is marked as the second abbreviation. An extension unit, connected to the retrieval unit, is used to determine the central word of the corresponding text based on the position of each character or word marked as the second abbreviation, and to perform extended association based on the composition of the central word to obtain a corresponding number of extended definitions; and to perform extended association on a single abbreviation divided from the second abbreviation to obtain several extended definitions corresponding to the single abbreviation. An identification unit, which is connected to the storage unit, the retrieval unit and the extension unit respectively, is used to identify the meaning of the first abbreviation and the extended meaning of the second abbreviation; A verification unit, which is connected to the identification unit and the storage unit respectively, is used to verify whether each of the extended interpretations and abbreviations matches; The central control unit, which is connected to the storage unit, the acquisition unit, the retrieval unit, the extension unit, the identification unit, and the verification unit, is used to determine whether the definition of the abbreviation conforms to a preset extension standard based on the number of obtained extended definitions. If it does not conform to the preset extension standard, the definition of the abbreviation is re-determined based on the extended definitions corresponding to a single abbreviation. If it conforms to the preset extension standard, the verification unit is controlled to verify whether each extended definition matches the abbreviation. Alternatively, the extended definitions are screened based on the domain, and the extended definitions corresponding to the domain are selected. The verification unit is then controlled to verify whether each extended definition matches the abbreviation. An output unit, connected to the central control unit, is used to derive the first abbreviation and the second abbreviation and form an abbreviation lookup table, the lookup table including the abbreviation, the original word of the abbreviation, and the definition of the abbreviation; The preset expansion standard is whether the number of expanded interpretations obtained after expanding the association for the second abbreviation reaches a preset expansion number threshold. When the number of extended definitions obtained is greater than or equal to the preset extension number threshold, the definition of the abbreviation is determined to meet the preset extension standard. When the number of extended definitions obtained is less than the preset extension number threshold, it is determined that the definition of the abbreviation does not meet the preset extension standard, and the abbreviation method of the abbreviation is determined to be morpheme abbreviation.

10. The automatic text abbreviation recognition system according to claim 9, characterized in that, It also includes a processor and a memory, the memory storing a computer program, wherein when the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 8.

Images