Question answering model construction method and system, question answering method and device, trial system
A technology for building methods and models, applied in the field of natural language processing reading comprehension, it can solve the problems of reading comprehension that is not suitable for multi-hop, cannot extract supporting evidence for multi-hop problems, etc., to achieve the effect of convenience and correctness
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0078] Please refer to figure 1 and figure 2 , figure 1 A schematic diagram of the composition of the question-answering model, figure 2 It is a schematic flow diagram of a method for building a question-and-answer model. Embodiment 1 of the present invention provides a method for building a question-and-answer model. The question-and-answer model includes a reasoning path retrieval model and a reading comprehension model. The method includes:
[0079] Step 1.1: Construct the retrieval reasoning path annotation dataset;
[0080] Step 1.2: Based on the retrieved inference path annotation data set, construct an inference path retrieval model for retrieving the inference path from preset information;
[0081] Step 1.3: Construct the loss function of the inference path retrieval model;
[0082] Step 1.4: Use the retrieval reasoning path annotation data set to train the reasoning path retrieval model, and obtain the trained reasoning path retrieval model;
[0083] Step 1.5: ...
Embodiment 2
[0106] Please refer to image 3 , image 3 It is a schematic diagram of the composition of the question-and-answer model building system. Embodiment 2 of the present invention provides a question-and-answer model building system. The question-and-answer model includes a reasoning path retrieval model and a reading comprehension model. The system includes:
[0107] Data set construction unit, used to construct retrieval reasoning path annotation data set;
[0108] An inference path retrieval model construction unit, configured to annotate the data set based on the retrieval inference path, and construct an inference path retrieval model for retrieving the inference path from preset information;
[0109] The loss function construction unit is used to construct the loss function of the inference path retrieval model;
[0110] The inference path retrieval model training unit is used to train the inference path retrieval model based on the retrieval inference path annotation data...
Embodiment 3
[0115] Please refer to Figure 4 , Figure 4 As a schematic flow chart of the question-and-answer method, Embodiment 3 of the present invention provides a question-and-answer method, which includes:
[0116] Step 1: Build a question answering model;
[0117] Step 2: Input the original question and information related to the original question into the question answering model;
[0118] Step 3: The question answering model outputs the answer to the original question and the reasoning path to obtain the answer from the information related to the original question;
[0119] The step 1 specifically includes:
[0120] Step 1.1: Construct the retrieval reasoning path annotation dataset;
[0121] Step 1.2: Based on the retrieved inference path annotation data set, construct an inference path retrieval model for retrieving the inference path from preset information;
[0122] Step 1.3: Construct the loss function of the inference path retrieval model;
[0123] Step 1.4: Use the re...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More