Search

CN-121997944-A - Case clue recognition analysis system

CN121997944ACN 121997944 ACN121997944 ACN 121997944ACN-121997944-A

Abstract

The invention relates to the technical field of computer data processing and natural language processing, in particular to a case clue identification and analysis system which comprises a data access and positioning module, a cascading semantic analysis module, an element feature extraction module, a value quantification and pricing module and a value quantification and pricing module, wherein the data access and positioning module calls a preset multisource map service interface to analyze network addresses in parallel, the target positions are determined through voting arbitration, the cascading semantic analysis module utilizes coarse-granularity identification attributes of a primary classification model to route to a vertical expert intelligent agent to generate fine granularity labels, the element feature extraction module removes interference digital features to generate target volume features and analyze user intention strength, the value quantification and pricing module maps the heterogeneous features into multidimensional scores based on a tree weight system to generate clue commercial value scores and pricing results, and the value assessment difficult problem caused by information non-standardization under a compliance consultation scene is solved.

Inventors

  • FU DAN
  • ZENG CHANGLIN
  • Qu Tianfen

Assignees

  • 成都优啊网络科技有限公司

Dates

Publication Date
20260508
Application Date
20260410

Claims (9)

  1. 1. A case thread recognition analysis system, comprising: The data access and positioning module is used for acquiring unstructured text data and network address data, calling a preset multi-source map service interface to carry out multi-channel analysis on the network address data, and determining a target geographic position based on a preset voting arbitration logic; The cascade semantic analysis module is used for carrying out coarse granularity identification on the unstructured text data by utilizing a primary classification model, determining primary classification attributes, dynamically routing the primary classification attributes to corresponding vertical expert agents based on the primary classification attributes, and generating secondary and tertiary fine granularity classification labels by the vertical expert agents; The element feature extraction module is used for carrying out numerical entity identification on the unstructured text data, generating target amount features after eliminating preset interference digital features, and synchronously analyzing the intention strength of a user; And the value quantification and pricing module is used for mapping the target geographic position, the fine granularity classification label, the target forehead characteristic and the user intention strength into multi-dimensional scores based on a preset tree weight configuration system, and generating cue commercial value scores and pricing results according to a preset multi-dimensional numerical value operation logic.
  2. 2. The case thread recognition analysis system of claim 1, wherein the data access and location module determines the target geographic location based on a preset voting arbitration logic configured to: acquiring a plurality of candidate geographic position information returned by the multi-source map service interface; Detecting consistency of the plurality of candidate geographic location information; If the plurality of candidate geographic position information is completely consistent, determining the consistent information as the target geographic position; if the candidate geographic position information with partial coincidence exists and the coincidence quantity exceeds a preset majority threshold, determining the majority information as the target geographic position; and if the conditions are not met, selecting candidate geographic position information corresponding to a preset authority source as the target geographic position, or triggering a front-end interaction instruction to acquire user confirmation information.
  3. 3. The case thread recognition analysis system of claim 1, wherein the cascading semantic analysis module dynamically routes to corresponding vertical expert agents based on the primary classification attributes, comprising: Constructing an intelligent agent group comprising a plurality of independent operation environments, wherein each intelligent agent corresponds to a preset compliance service field; Activating a target vertical expert agent matched with the primary classification attribute in the agent group in response to the generation of the primary classification attribute; and carrying out deep semantic analysis on the unstructured text data by utilizing the target vertical expert agent, and outputting a secondary classification result and a tertiary classification result which belong to the primary classification attribute.
  4. 4. The case cue recognition analysis system of claim 1, wherein the element feature extraction module generates a target value feature after eliminating a preset interference digital feature, comprising: scanning the unstructured text data by using a preset Chinese amount regularization algorithm, and extracting an original numerical value set; identifying character strings which accord with a preset mobile phone number rule and a preset license plate number rule in the unstructured text data, and defining the character strings as the interference digital characteristics; and removing the part overlapped with the interference digital characteristic from the original numerical value set, and carrying out unit unification processing on the residual numerical values to generate the target value characteristic.
  5. 5. The case thread recognition analysis system of claim 1, wherein the value quantification and pricing module maps the target geographic location, the fine-grained classification tag, to a multi-dimensional score based on a preset tree weight configuration hierarchy, comprising: calling a preset regional weight tree and a classification weight tree, wherein the regional weight tree comprises three-level nodes of provinces, cities and counties, and the classification weight tree comprises one-level, two-level and three-level classification nodes; Searching corresponding region weight scores in the region weight tree according to the target geographic position; and searching the corresponding classification weight scores in the classification weight tree according to the fine-granularity classification labels.
  6. 6. The system of claim 5, wherein the value quantification and pricing module generates the lead value scores and pricing results based on a predetermined multidimensional numerical algorithm, comprising: based on a preset numerical mapping relation, obtaining an amount score corresponding to the target forehead characteristic and an intention score corresponding to the user intention strength; Accumulating the classification weight score, the region weight score, the monetary score and the intention score to generate a comprehensive value score; And acquiring a preset global price adjustment coefficient, calculating the product of the comprehensive value integral and the global price adjustment coefficient, and defining the calculated product as the pricing result.
  7. 7. The case thread recognition analysis system of claim 4, wherein the element feature extraction module parses a user intent strength, comprising: analyzing the action verbs and time idioms in the unstructured text data by using a preset intention recognition model; If the semantic features representing the proxy agent or the instant contact are identified, judging that the semantic features are strong in intention strength; otherwise, the weak intention strength is determined.
  8. 8. The case thread recognition analysis system of claim 1, further comprising: the safety filtering module is used for cleaning the unstructured text data before the data access and positioning module is executed; Matching the unstructured text data based on a preset blacklist library and a sensitive word library; If the matching hits, the data is intercepted and the subsequent flow is terminated, and if the matching hits, the cleaned data is transmitted to the cascade semantic analysis module.
  9. 9. The case thread recognition analysis system of claim 1, further comprising: the auction ordering display module is used for receiving the pricing result and the target geographic position; writing the pricing result back to a price attribute field of a database; and responding to a front-end query request, establishing a double index based on the price attribute field and the target geographic position, and generating a cue display list according to the pricing result from high to low.

Description

Case clue recognition analysis system Technical Field The invention relates to the technical field of computer data processing and natural language processing, in particular to a case clue recognition analysis system. Background In the application scenario of the internet compliance consultation service, the platform interfaces massive case cue data by means of an intelligent distribution system, and the system generally needs to combine unstructured text description and user network address data to analyze the attribution and business value of the case in real time; Aiming at the recognition and pricing of case clues, the existing scheme generally adopts a single-channel processing architecture, namely, a single map server interface is relied on to analyze network addresses, a general natural language processing model is utilized to carry out keyword matching or shallow classification on consultation texts, and an extraction result is directly used as a basis for clue value evaluation; while this approach has some feasibility in a standardized simple consultation, due to its excessive reliance on a single data source and lack of deep logic constraints in the drop-type domain, when encountering cross-regional or nonstandard descriptions, a single map service often causes positioning drift due to data update lag, causing jurisdiction decision errors; in addition, the general model is difficult to accurately distinguish fine-granularity case routing in the compliance field, and a simple numerical extraction algorithm is extremely easy to misidentify living scene noise such as a mobile phone number, a license plate number and the like as high-scale features, so that a value evaluation result is seriously and virtually high; therefore, how to establish an analysis mechanism with multi-source calibration and anti-interference capability, effectively eliminates digital noise, and improves case positioning accuracy and reliability of commercial value evaluation becomes a technical problem to be solved. Disclosure of Invention In order to solve the technical problems, the invention provides a case clue recognition and analysis system, which specifically comprises the following steps: the data access and positioning module is used for acquiring unstructured text data and network address data, calling a preset multi-source map service interface to carry out multi-channel analysis on the network address data, and determining a target geographic position based on a preset voting arbitration logic; The cascade semantic analysis module is used for carrying out coarse granularity recognition on unstructured text data by utilizing a primary classification model, determining primary classification attributes, dynamically routing to corresponding vertical expert intelligent agents based on the primary classification attributes, and generating secondary and tertiary fine granularity classification labels by the vertical expert intelligent agents; The system comprises an element feature extraction module, a value quantification and pricing module and a factor feature calculation module, wherein the element feature extraction module is used for carrying out numerical entity identification on unstructured text data, generating target amount features after eliminating preset interference digital features, and synchronously analyzing user intention strength, and the value quantification and pricing module is used for mapping target geographic positions, fine granularity classification labels, the target amount features and the user intention strength into multidimensional scores based on a preset tree weight configuration system and generating cue commercial value scores and pricing results according to preset multidimensional numerical operation logic. Preferably, the data access and positioning module determines a target geographic position based on a preset voting arbitration logic and is configured to acquire a plurality of candidate geographic position information returned by the multi-source map service interface; If the partial consistent candidate geographic position information exists and the consistent quantity exceeds a preset multi-dispatch threshold value, the multi-dispatch information is determined to be the target geographic position; If the conditions are not met, selecting candidate geographic position information corresponding to a preset authority source as a target geographic position, or triggering a front-end interaction instruction to acquire user confirmation information. Preferably, the cascade semantic analysis module dynamically routes to corresponding vertical expert intelligent agents based on the first-level classification attribute, and comprises the steps of constructing an intelligent agent group comprising a plurality of independent operation environments, wherein each intelligent agent corresponds to a preset compliance service field respectively; In response to the generation of the first class