Intelligent topic selection system

Abstract

An intelligent topic selection system includes a processing unit and a storage unit electrically connected to the processing unit. The storage unit stores a topic database containing multiple self-assessment topics and a re-ranking model. The processing unit is used to: utilize retrieval-enhanced generation technology to select multiple candidate self-assessment topics from the self-assessment topics that have semantic relevance to the audit reference materials, based on audit reference materials; utilize the re-ranking model to select the self-assessment topic with the highest semantic relevance to the audit reference materials from the waiting self-assessment topics; add the selected self-assessment topic to a self-assessment form; and output the self-assessment form.

Claims

1. An intelligent topic selection system, comprising: a processing unit; and a storage unit electrically connected to the processing unit, storing a topic database containing multiple self-assessment topics, and a re-ranking model; the processing unit is configured to: utilize retrieval enhancement generation technology, based on audit reference materials, select multiple candidate self-assessment topics from the self-assessment topics that have semantic relevance to the audit reference materials; utilize the re-ranking model to select the self-assessment topic with the highest semantic relevance to the audit reference materials from the waiting self-assessment topics; add the selected self-assessment topic to a self-assessment form; and output the self-assessment form.

2. The intelligent topic selection system as described in claim 1, wherein, The reordering model is implemented using a cross-encoder architecture. The processing unit selects the self-assessment items using the reordering model by merging the audit reference materials and the self-assessment items and inputting them into the reordering model, so as to analyze the semantic relevance between each self-assessment item and the audit reference materials through the reordering model.

3. The intelligent topic selection system as described in claim 2 is suitable for use in conjunction with a language processing model; wherein, The method by which the processing unit selects the self-evaluation topic using the reordering model further includes: generating multiple relevance scores corresponding to the self-evaluation topics to be selected using the reordering model, wherein each relevance score indicates the degree of semantic relevance between the corresponding candidate self-evaluation topic and the audit reference material; and selecting the self-evaluation topic from the preferred self-evaluation topics with the highest relevance scores among the self-evaluation topics to be selected using the language processing model.

4. The intelligent topic selection system as described in claim 1, wherein, The storage unit also stores a semantic embedding model, and the question database contains multiple question feature data corresponding to the self-assessment questions. Each question feature data is presented as a semantic vector of more than 1024 dimensions. The processing unit selects the self-assessment questions from the audit reference data in the following way: using the semantic embedding model, a reference data feature vector is generated based on the audit reference data; multiple question feature data with the highest similarity to the reference data feature vector are selected from the question feature data; and the self-assessment questions corresponding to the selected multiple question feature data are respectively used as the self-assessment questions to be selected.

Images