Search

CN-121996772-A - Prompting word optimization method, system, equipment and medium based on artificial intelligence

CN121996772ACN 121996772 ACN121996772 ACN 121996772ACN-121996772-A

Abstract

The application is suitable for the technical field of artificial intelligence, and provides an artificial intelligence-based prompting word optimization method, which comprises the steps of constructing a multisource fusion knowledge graph based on structured field knowledge, common knowledge and interaction data, maintaining conversation context, receiving user inquiry, carrying out semantic completion and deep intention analysis on the user inquiry by combining the multisource fusion knowledge graph to obtain complete semantic representation, generating a model by combining conditions based on the complete semantic representation, and carrying out personalized reordering on the candidate prompting words to generate an optimized prompting word list, so that the deep meaning and the expression intention of non-literal of a user can be accurately understood, and the problems of 'insufficient semantic understanding' and 'no-literal expression is not in a policy' in the prior art are effectively solved.

Inventors

  • YAN YIQIANG

Assignees

  • 安徽三七极域网络科技有限公司

Dates

Publication Date
20260508
Application Date
20251225

Claims (10)

  1. 1. The artificial intelligence-based prompt word optimization method is characterized by comprising the following steps of: Constructing a multisource fusion knowledge graph based on structured domain knowledge, common general knowledge and interaction data; maintaining a session context, receiving a user query, and carrying out semantic completion and deep intention analysis on the user query by combining the multisource fusion knowledge graph to obtain a complete semantic representation; generating a model based on the complete semantic representation and combining conditions, and generating a plurality of candidate prompt words in parallel; And performing personalized reordering on the candidate prompt words to generate an optimized prompt word list.
  2. 2. The method of claim 1, wherein constructing a multi-source fusion knowledge-graph based on structured domain knowledge, common sense, and interaction data comprises: Collecting multi-source heterogeneous data from a vertical field database, a common sense atlas and a historical interaction log; Performing entity disambiguation and attribute alignment on the multi-source heterogeneous data, and performing multi-level relation extraction to construct an initial knowledge network containing attribute relations, classification relations and scene relations; aligning the domain entities in the vertical domain database with concepts in the common sense atlas through entity linking technology to generate background complement knowledge; mining entity pairs co-occurring at high frequency from the history interaction log and a query mode to form a semantic association rule reflecting the implicit requirements of the user; And fusing the initial knowledge network, the background complement knowledge and the semantic association rule based on a graph neural network to generate a multi-source fusion knowledge graph.
  3. 3. The method of claim 1, wherein maintaining the session context comprises: constructing a layered dialogue memory network, wherein the layered dialogue memory network comprises a short-term memory layer, a medium-term memory layer and a long-term memory layer; performing reference resolution by multi-constraint reasoning based on the hierarchical dialogue memory network and the multi-source fusion knowledge graph; Modeling and tracking the evolution of the user intention in the multi-round dialogue based on the state machine model, and outputting the current intention state and the context continuation path.
  4. 4. The method of claim 3, wherein the performing an reference resolution by multi-constraint reasoning comprises: extracting an entity list from the short-term memory layer, and searching related entities in the multi-source fusion knowledge graph to form a candidate entity set; Dynamically calculating focus hotness scores of candidate entities in the candidate entity set based on a focus tracking model; And calculating the comprehensive index score of each candidate entity according to the comprehensive grammar consistency, the knowledge graph verification result and the focus hotness score, and determining the candidate entity with the highest score as the index object.
  5. 5. The method of claim 1, wherein the receiving the user query, in conjunction with the multi-source fusion knowledge-graph, semantically completing and deep intent parsing the user query to obtain a complete semantic representation, comprises: Acquiring a session context package and a user query statement, and carrying out joint coding to generate complete query semantics; Performing entity link disambiguation, semantic completion and scenerization information injection on the user query statement based on the multi-source fusion knowledge graph to form a semantic enhanced query expression; performing literal intent analysis and slot filling on the semantic enhanced query representation through a deep analysis network to generate a deep semantic intent; performing recognition and conversion of non-literal expression on the semantic enhanced query representation to generate a standard semantic intention; generating a complete semantic representation based on the complete query semantics, the deep semantic intent, and the standard semantic intent.
  6. 6. The method of claim 1, wherein generating a model based on the complete semantic representation in combination with conditions, generates a plurality of candidate hint words in parallel, comprising: constructing a plurality of parallel prompting word generation paths, wherein each prompting word generation path corresponds to different knowledge sources and generation strategies; Building a structured condition prompt for each prompt word generating path, and encoding a condition signal of a corresponding knowledge source into a condition vector; the condition vectors from different prompt generation paths are weighted and fused through a dynamic gating fusion network to generate a condition control signal; Based on the condition control signal, driving a pre-training language model to execute controllable parallel decoding to generate a plurality of candidate prompt words; The prompt word generation path at least comprises a knowledge framework extension path based on the multi-source fusion knowledge graph and the complete semantic representation, a current intention state and context continuation path based on a session context package, an information injection path based on real-time external knowledge retrieval and a creative heuristic path based on standard semantic intention.
  7. 7. The method of claim 1, wherein the personalized reordering of the candidate hint words generates an optimized hint word list comprising: based on a personalized fitness prediction model, predicting personalized adoption probability of each candidate prompting word by combining real-time user images, current session preferences and generation metadata of the candidate prompting word; and reordering each candidate prompting word based on the personalized adoption probability to generate an optimized prompting word list.
  8. 8. An artificial intelligence based alert word optimization system comprising: the first processing module is used for constructing a multisource fusion knowledge graph based on structured domain knowledge, common knowledge and interaction data; The second processing module is used for maintaining the session context, receiving user inquiry, and combining the multisource fusion knowledge graph to perform semantic completion and deep intention analysis on the user inquiry so as to obtain complete semantic representation; the third processing module is used for generating a model based on the complete semantic representation and combining conditions to generate a plurality of candidate prompt words in parallel; and the fourth processing module is used for carrying out personalized reordering on the candidate prompt words and generating an optimized prompt word list.
  9. 9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the method according to any of claims 1 to 7 when executing the computer program.
  10. 10. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the method according to any one of claims 1 to 7.

Description

Prompting word optimization method, system, equipment and medium based on artificial intelligence Technical Field The application belongs to the technical field of artificial intelligence, and particularly relates to a prompting word optimization method, a prompting word optimization system, prompting word optimization equipment and prompting word optimization media based on artificial intelligence. Background The prompt word is used as a core medium for the interaction of the user with the search engine, the intelligent assistant and the content generation tool, and the quality of the prompt word directly influences the efficiency and experience of information acquisition. The traditional method mainly relies on a statistical language model and a predefined rule engine, and can provide basic relevance suggestions for users in a simple and clear scene by analyzing shallow features such as query word frequency, co-occurrence relation and the like. For example, when a user enters an explicit keyword, the system may generate a conventional associative or complement hint based on historical search data. However, as application scenarios become increasingly complex and user demands continue to deepen, such conventional techniques expose inherent limitations at multiple core levels. First, in terms of semantic understanding depth, it is mainly limited to processing literal meaning, difficult to resolve complex semantic structures, ambiguities and implicit demands of users, which are based on mechanisms of surface vocabulary matching, and lack of understanding when encountering knowledge that depends on deep logic and context. Second, in terms of interactive continuity, existing systems typically treat each user query as an independent event, lacking efficient maintenance and utilization of multi-turn dialog contexts. Furthermore, in terms of knowledge utilization breadth and timeliness, most systems rely on a closed and static local knowledge base, and cannot dynamically access and fuse an external general knowledge base or real-time updated vertical field data, so that the generated prompt word has insufficient information quantity or old content. Finally, in the aspect of language diversity understanding, the prior art is not in the way for non-literal expression forms such as metaphors, exaggeration and the like which are frequently used in daily communication of people. For example, in the face of the user expressing "hungry to swallow a bull", the system cannot insight into the real requirement of "find a large number of high cost performance restaurants" behind it, and then generate irrelevant prompt words, disabling interaction. Therefore, a prompt word optimization technology is needed to fundamentally cope with challenges in complex interaction scenarios, and to realize accurate, coherent, intelligent and insight-rich prompt word generation and service. Disclosure of Invention The embodiment of the application provides a prompting word optimization method, a prompting word optimization system, prompting word optimization equipment and prompting word optimization media based on artificial intelligence, which can solve one of the problems in the prior art. In a first aspect, an embodiment of the present application provides an artificial intelligence based alert word optimization method, including: Constructing a multisource fusion knowledge graph based on structured domain knowledge, common general knowledge and interaction data; maintaining a session context, receiving a user query, and carrying out semantic completion and deep intention analysis on the user query by combining the multisource fusion knowledge graph to obtain a complete semantic representation; generating a model based on the complete semantic representation and combining conditions, and generating a plurality of candidate prompt words in parallel; And performing personalized reordering on the candidate prompt words to generate an optimized prompt word list. Further, the constructing a multisource fusion knowledge graph based on the structured domain knowledge, the common general knowledge and the interaction data includes: Collecting multi-source heterogeneous data from a vertical field database, a common sense atlas and a historical interaction log; Performing entity disambiguation and attribute alignment on the multi-source heterogeneous data, and performing multi-level relation extraction to construct an initial knowledge network containing attribute relations, classification relations and scene relations; aligning the domain entities in the vertical domain database with concepts in the common sense atlas through entity linking technology to generate background complement knowledge; mining entity pairs co-occurring at high frequency from the history interaction log and a query mode to form a semantic association rule reflecting the implicit requirements of the user; And fusing the initial knowledge network, the background complement k