Search

CN-122019789-A - Intellectual property consultation session processing method and server based on AI application

CN122019789ACN 122019789 ACN122019789 ACN 122019789ACN-122019789-A

Abstract

The invention provides an intellectual property consultation session processing method and a server based on AI application, which are used for generating session triplet data by combining a preset industrial factor labeling paradigm according to single-round intellectual property consultation session text acquired in real time, associating corresponding factor expressions in stored historical consultation session records based on the session triplet data, constructing user consultation complaint association map data, matching new session triplet data generated by newly acquired subsequent consultation session text with map data, extracting a historical original consultation table associated with a historical node and a responded content expression, fusing complete missing consultation context to obtain a complete complaint data set, identifying associated node content corresponding to historical complaints with business continuity of the current complaint type expression, extracting core business information fragments in the associated node content, and combining the current complaint type expression to generate a user potential complaint result. The invention improves service pertinence and suitability.

Inventors

  • LIU YINYAN
  • WANG XIYING
  • DU KEQING

Assignees

  • 河南牧业经济学院

Dates

Publication Date
20260512
Application Date
20260127

Claims (10)

  1. 1. An intellectual property consultation session processing method based on an AI application, the method comprising: aiming at a single-round intellectual property consultation session text acquired in real time, generating session triplet data comprising user identification expression, known object expression and appeal type expression by combining a preset known industrial element annotation paradigm; Based on the session triplet data, associating corresponding element expressions in the stored historical yield-aware consultation session record, constructing user yield-aware consultation appeal association spectrum data, wherein nodes of the user yield-aware consultation appeal association spectrum data correspond to the user identification expression, the yield-aware object expression, the appeal type expression and the responded content expression, and establishing directional association links among the nodes by using industry-aware business progressive logic; Performing node matching on new session triplet data generated by newly acquired subsequent known produced consultation session text and the related map data of the user known produced consultation appeal, extracting a historical original consultation list and a responded content list which are related to the historical node based on the historical node locked by the user identification list and the known produced object list, fusing the historical original consultation list and the responded content list as background information with the current subsequent known produced consultation session text, and complementing the missing consultation context to obtain a complete appeal data set; traversing a business progressive association link in the association map data of the consultation-producing consultation of the user according to the complete appeal data set, identifying association node content corresponding to a historical appeal with business continuity expressed by the current appeal type, and extracting core business information fragments; And generating a user potential appeal result based on the core business information fragment and combining with the appeal type expression of the current follow-up consultation conversation text, and synchronizing the user potential appeal result to a to-be-processed queue of the follow-up consultation response system.
  2. 2. The method as set forth in claim 1, wherein the associating the corresponding element expressions in the stored history counseling session record based on the session triplet data to construct user counseling appeal association map data includes: Splitting a native Japanese text expression of a user identification expression, a known object expression and a appeal type expression from the session triplet data, and corresponding to the native Japanese text expression of the same type in the stored historical known consultation session record one by one to obtain an initial semantic matching table containing a corresponding matching mark; cross-checking corresponding items in the initial semantic matching table, removing cross-category error matching items, and reserving items directly related to the category expressions to obtain a checked semantic matching table containing priority ordering marks; binding the user identification expression in the verified semantic matching table with the user identification expression in the history known product consultation session record, associating the corresponding known product object expression and the appeal type expression, and obtaining a user identification associated node set containing a node preliminary pointing mark; for each known object expression in the user identification association node set, associating a corresponding appeal type expression and a historical responded content expression to obtain a known object-appeal-response association subset containing a hierarchical pointing mark; combining the user identification association node set with a plurality of known object-appeal-response association subsets, and combing the hierarchical bearing relations among the nodes to obtain an association map initial version containing complete link marks; and de-duplicating the node links of the initial version of the association graph, removing the repeated level points and node contents, and generating association graph data containing unique link marks for user to know and produce consultation and appeal.
  3. 3. The method of claim 2, wherein cross-checking the corresponding entries in the initial semantic matching table, eliminating cross-class erroneous matching entries, and retaining entries directly associated with class expressions to obtain a checked semantic matching table with prioritized labels, comprising: Extracting cross-category matching items which do not pass the verification from the initial semantic matching table, splitting the expression in the items into minimum semantic units, and corresponding to the minimum semantic units of the same category in the history knowledge-based consultation session record one by one to obtain a fine-granularity semantic matching table containing unit matching marks; Counting unit matching items in the fine granularity semantic matching table, reserving items with unit matching proportion meeting requirements, and re-adding the items to the checked semantic matching table to obtain an expanded semantic matching table containing the full item marks; Based on the user identification expression in the extended semantic matching table, associating cross-session knowledge production object expressions of the same user in the history knowledge production consultation session record, supplementing the cross-session knowledge production object expressions to the user identification association node set to obtain an extended user identification association node set containing cross-session node marks; Aiming at cross-session known object producing expressions in the expanded user identification association node set, associating corresponding cross-session appeal type expressions and responded content expressions, supplementing the cross-session known object producing-appeal-response association subsets to obtain expanded association subsets containing cross-session association marks; Combining the expanded user identification association node set with the expanded association subset, and combing the hierarchical bearing relation of the cross-session nodes to obtain an expanded association map initial version containing the cross-session link marks; And verifying the cross-session link of the initial version of the extended association graph, removing the links with time sequence conflicts, and generating association graph data containing cross-session association marks for user know-how consultation appeal.
  4. 4. The method of claim 3, wherein the associating the cross-session awareness object-producing expressions of the same user in the history awareness consultation session record based on the user identification expressions in the extended semantic matching table with the set of user identification associated nodes to obtain the set of extended user identification associated nodes including cross-session node labels comprises: Extracting all session items expressed by the same user identification from the extended semantic matching table, and sorting according to session occurrence time stamps to obtain a time-ordered session item set containing time-ordered marks; Carrying out semantic acceptance check on adjacent session items in the time-sequence session item set, and marking item pairs with progressive association in the representation to obtain a time-sequence acceptance item set containing an acceptance association mark; based on the time sequence receiving item set, associating corresponding known object expression, demand type expression and responded content expression, supplementing the time sequence receiving item set to the expanded user identification associated node set, and obtaining a time sequence user identification associated node set containing time sequence receiving node marks; Aiming at the time sequence receiving items in the time sequence user identification association node set, associating corresponding cross-session responded content expressions, supplementing the extended association subsets to obtain time sequence extended association subsets containing time sequence receiving association marks; combining the time-sequence user identification association node set with the time-sequence extension association subset, and combing the hierarchical links of the time-sequence receiving nodes to obtain a time-sequence association map initial version containing time-sequence link marks; And de-duplicating the time sequence link of the time sequence associated map initial version, eliminating repeated time sequence bearing points, and generating the associated map data containing time sequence associated marks for the user to know and produce consultation and appeal.
  5. 5. The method as set forth in claim 1, wherein the node matching the new session triplet data generated by the newly collected subsequent known-produced consultation session text with the user known-produced consultation appeal association map data, extracting a history original consultation table and a responded content expression associated with the history node based on the history node of the user identification expression and the known-produced object expression locking association, fusing the history original consultation table and the responded content expression as background information with the current subsequent known-produced consultation session text, complementing the missing consultation context, and obtaining a complete appeal data set, comprising: extracting the original date semantic word elements of the subsequent consultation session text of known production, extracting the original expression of the user identification expression, the known production object expression and the appeal type expression, and obtaining a new session triplet initial version containing an extraction mark; Carrying out semantic regularity on the original expression in the initial version of the new session triplet, and unifying expression formats and words to obtain standardized new session triplet data containing a regularity mark; binding the user identification expression in the standardized new session triple data with the user identification node of the related map data of the user known to produce consultation appeal, and locking all related nodes and links of the same user to obtain a link set of the related nodes of the same user containing locking marks; screening historical session nodes associated with the appeal type expression and the known object expression of the standardized new session triple data from the same user associated node link set, extracting a historical original consultation expression, a responded content expression and a link relation thereof associated with the corresponding nodes, and obtaining historical appeal associated path data containing content and link marks; extracting a historical original consultation list and a responded content representation corresponding to the known object representation in the historical appeal association path data, associating the historical original consultation list and the responded content representation with a follow-up known consultation session text as background information, and constructing a complete appeal data set initial version with continuous context; And de-duplicating the initial version of the complete appeal data set, removing repeated pre-consultation and responded content expression, and generating the complete appeal data set containing the unique content mark.
  6. 6. The method of claim 5, wherein semantically normalizing the native expression in the initial version of the new session triplet, unifying expression formats and words, and obtaining normalized new session triplet data with a normalization tag, comprises: Extracting an unordered original expression from the new session triplet initial version, and carrying out semantic mapping on the original expression and a homonymy expression in the related map data of the user known consultation appeal so as to obtain an expression mapping table containing mapping marks; Based on the expression mapping table, carrying out self-adaption normalization on the original expression in the initial version of the new session triplet, and matching the expression format in the atlas to obtain self-adaption standardized new session triplet data containing self-adaption marks; binding a user identification expression in the self-adaptive standardized new session triple data with a user identification node of the user-known consultation appeal association graph data, and locking cross-session association nodes and links of the same user to obtain a cross-session association node link set containing cross-session marks; Screening historical session nodes related to the self-adaptive standardized new session triple data in a cross-session semantic manner from the cross-session associated node link set, extracting historical original consultation expressions, responded content expressions and link relations thereof related to the corresponding nodes, and obtaining cross-session historical appeal related path data containing content and link marks; Combining the responded content expression in the cross-session history appeal correlation path data with a continuous known product consultation session text, and complementing the pre-consultation content of the cross-session deletion to obtain a complete appeal data set expansion version containing a cross-session complement mark; and performing cross-session content verification on the extended version of the complete appeal data set, removing the front consultation content with time sequence conflict, and generating the complete appeal data set containing the cross-session completion mark.
  7. 7. The method of claim 6, wherein adaptively regularizing the native expression in the initial version of the new session triplet based on the expression mapping table, matching expression formats in a map, obtaining adaptively normalized new session triplet data with adaptive markers, comprises: Extracting a session time stamp from the continuous consultation-producing session text, and obtaining a time sequence matching table containing time sequence matching marks corresponding to the node time stamp in the correlation map data of the user consultation-producing requirement; Adding a time sequence mark to the self-adaptive standardized new session triplet data based on the time sequence matching table to obtain time sequence standardized new session triplet data containing the time sequence mark; Combining the user identification expression in the time sequence standardized new session triple data with a time sequence mark, and locking the related nodes and links of the same user and time sequence before the current session in the related map data of the user known consultation appeal to obtain a time sequence filtering related node link set containing a time sequence filtering mark; Screening historical session nodes related to the time sequence type expression and the time sequence of the time sequence standardized new session triplet data from the time sequence filtering related node link set, extracting a historical original consultation expression, a responded content expression and a link relation thereof related to the corresponding nodes, and obtaining time sequence filtering historical appeal related path data containing content and link marks; extracting a historical original consultation list and a responded content representation corresponding to the known object representation in the time sequence filtering historical appeal association path data, associating the historical original consultation list and the responded content representation with a subsequent known product consultation session text as background information, and constructing a complete appeal data set time sequence version with continuous time sequence context; and performing time sequence consistency check on the time sequence version of the complete appeal data set, removing the pre-consultation content of time sequence dislocation, and generating the complete appeal data set containing the time sequence consistency complement mark.
  8. 8. The method of claim 1, wherein traversing the business progressive association links in the user-known consultation-producing association graph data based on the complete claim data set, identifying association node content corresponding to historical appeal with business continuity expressed by current claim type, and extracting core business information fragments therein, comprises: Extracting the front consultation content of the current appeal type expression and the complement from the complete appeal data set, and obtaining appeal continuation reference data containing continuation marks corresponding to the node content in the user-known consultation appeal association map data; Resolving the current appeal type expression in the appeal continuation reference data into a minimum semantic unit, and corresponding to the semantic unit expressed by the appeal type node in the user-aware consultation appeal association map data to obtain a semantic correspondence table containing a matching mark; Screening a appeal type node expression associated with the current appeal type expression semantic unit from the semantic correspondence table, extracting a corresponding link and a responded content expression, and obtaining continuation associated path data containing a continuation mark; extracting a historical original consultation expression, a known object evolution expression and a responded content expression which are related with the current demand type expression existing service continuation from the continuous association path data to obtain a core service information fragment initial set containing a continuous mark; Content correlation screening is carried out on the initial set of the core service information fragments, and the responded content expression irrelevant to the current requirement is removed, so that a screened core service information fragment set containing correlation marks is obtained; And integrating the filtered core service information fragment sets, and unifying expression formats to generate the core service information fragments containing the unique content marks.
  9. 9. The method as claimed in claim 8, wherein said decomposing the current complaint type expression in the complaint continuation reference data into the minimum semantic unit corresponds to the semantic unit expressed by the complaint type node in the user-aware consultation complaint correlation map data to obtain the semantic correspondence table with the matching mark, comprising: Extracting semantic units which are not marked for bearing association from the semantic correspondence table, and carrying out fine-granularity bearing verification on the semantic units expressed by the appeal type nodes in the user-known consultation appeal association map data to obtain a fine-granularity semantic correspondence table containing fine-granularity matching marks; based on the fine granularity semantic correspondence table, adding a fine granularity adapting associated appeal type node expression to the continuation associated path data to obtain expanded continuation associated path data containing fine granularity marks; extracting the responded content expression of fine granularity bearing association from the extended continuity association path data, and supplementing the content expression to the initial set of the core service information fragments to obtain the initial set of the extended core service information fragments containing fine granularity marks; performing content correlation secondary screening on the initial set of the expanded core service information fragments, and removing the content which is not sufficiently associated with the current demand fine granularity to obtain a secondary screened core service information fragment set containing a secondary screening mark; performing semantic consistency carding on the core service information fragment set after the secondary screening, and arranging contents according to a service receiving sequence to obtain a coherent core service information fragment set containing a coherent mark; And de-duplicating the coherent core service information fragment set, removing repeated expression content, and generating a core service information fragment containing fine granularity bearing marks.
  10. 10. A server for a server, which comprises a server and a server, characterized by comprising the following steps: A processor; and a memory, wherein the memory has stored therein computer readable code which, when executed by the processor, causes the processor to perform the method of any of claims 1-9.

Description

Intellectual property consultation session processing method and server based on AI application Technical Field The present invention relates to the field of data processing, and more particularly, to an intellectual property consultation session processing method and server based on AI application. Background With the continuous penetration of artificial intelligence technology in the field of intellectual property service, the intellectual property consultation session processing technology based on AI gradually becomes an important tool for improving the efficiency of the consultation service of the knowledge. In the prior art, the conventional processing mode is to execute general semantic recognition or keyword extraction operation on the real-time acquired known consultation session text to capture the consultation appeal explicitly expressed by the user, and at the same time, when the historical session record of the user is associated, the processing mode is to take the scattered historical session fragments as the supplementary reference of the current consultation by taking the time sequence of session occurrence or surface keyword matching as the screening basis, and then combine the dominant appeal acquired in real time to generate the basic data of the service response. However, in such a processing manner, the information processing of the learning and producing session text is focused only on the content explicitly expressed by the user, so that the processing logic and the association rule special for the intellectual property service are difficult to be matched, the deep association support of the service logic dimension is also lacking in the retrieval of the history session data, the complementary background information cannot be accurately matched with the service context of the current consultation, the learning and producing server can only passively respond to the explicitly presented appeal of the user, the implicit service requirement of the user cannot be perceived in advance, and the accurate consultation service response adapting to the learning industry logic is difficult to be realized. Disclosure of Invention In view of this, the embodiment of the invention provides at least an intellectual property consultation session processing method and a server based on AI application. According to one aspect of the embodiment of the invention, an intellectual property consultation session processing method based on AI application is provided, and the method comprises the steps of generating session triplet data comprising user identification expression, object expression and demand type expression by combining a preset industrial element annotation paradigm aiming at a single-round intellectual property consultation session text acquired in real time; based on the conversation triple data, the corresponding element expression in the stored historical known product consultation conversation record is related to construct the related graph data of the user known product consultation appeal, the node of the related graph data of the user known product consultation appeal corresponds to the user identification expression, the known product object expression, the appeal type expression and the responded content expression, a directional related link is established among the nodes by the known industry progressive logic, the new conversation triple data generated by the newly collected subsequent known product consultation conversation text is matched with the related graph data of the user known product consultation appeal, the historical original consultation expression and the responded content expression which are related to the historical node are extracted based on the historical node of the locking relation of the user identification expression and the known product object expression, the historical original consultation expression and the responded content expression are used as background information to be fused with the current subsequent known product consultation conversation text, the complete missing consultation context is completed, the complete appeal data set is obtained, the related link of the business progressive in the related graph data of the user is traversed according to the complete appeal data set, the related node corresponding to the current appeal type of the historical demand in existence business continuity is identified, the extracted core content corresponding to the current appeal type of the current appeal type is based on the current consultation section text, and generating a user potential appeal result, and synchronizing the user potential appeal result to a waiting queue of the known-product consultation response system. According to another aspect of an embodiment of the present invention, there is provided a server comprising a processor, and a memory, wherein the memory has stored therein computer readable code which, when executed by the processor