CN-121996744-A - Information processing method, information processing device, computer readable storage medium and electronic equipment

Abstract

The application discloses an information processing method, an information processing device, a computer readable storage medium and electronic equipment. The method comprises the steps of obtaining target problems related to the target knowledge field, carrying out vector conversion processing on the target problems by adopting an embedded characterization model to obtain initial vectors, carrying out vector update processing on the initial vectors through a target neural network model to obtain target vectors, wherein the target neural network model is obtained based on reference knowledge training of the target knowledge field, carrying out knowledge search in the reference knowledge of the target knowledge field according to the target vectors, and determining target reply information corresponding to the target problems according to search results. The method solves the technical problem that in a question-answering scene in the related technology, under the condition that a general embedded characterization model cannot be finely tuned, the accuracy of vectors obtained by processing the territorial knowledge by adopting the general embedded characterization model is low, so that the accuracy of the territorial knowledge question-answering is low.

Inventors

• MENG ZIHAO
• LIU TAO
• ZHANG HENG

Assignees

• 阿里云计算有限公司

Dates

Publication Date

20260508

Application Date

20241105

Claims (16)

1. 1. An information processing method, characterized by comprising: Acquiring a target problem related to the field of target knowledge; Performing vector conversion processing on the target problem by adopting an embedded characterization model to obtain an initial vector; Performing vector update processing on the initial vector through a target neural network model to obtain a target vector, wherein the target neural network model is obtained based on reference knowledge training in the target knowledge field; And searching knowledge in the reference knowledge in the target knowledge field according to the target vector, and determining target reply information corresponding to the target problem according to a search result.
2. 2. The method of claim 1, wherein performing a knowledge search in reference knowledge of the target knowledge domain based on the target vector comprises: acquiring a knowledge vector of reference knowledge in the target knowledge field; Determining target reference knowledge from the reference knowledge in the target knowledge field according to the similarity between the target vector and the knowledge vector; And determining the search result according to the target reference knowledge.
3. 3. The method of claim 2, wherein obtaining a knowledge vector of reference knowledge of the target knowledge domain comprises: Performing vector conversion processing on the reference knowledge by adopting the embedded characterization model to obtain an initial knowledge vector; and carrying out vector update processing on the initial knowledge vector through the target neural network model to obtain the knowledge vector.
4. 4. The method of claim 1, wherein the search results include at least one target reference knowledge, and determining target reply information corresponding to the target problem according to the search results includes: And inputting the target problem and the at least one target reference knowledge into a large language model, and determining target reply information corresponding to the target problem through the large language model.
5. 5. The method of claim 1, wherein the target neural network model is obtained by: Acquiring an initial training sample set, wherein an initial training sample in the initial training sample set at least comprises a sample problem in the target knowledge field, first sample reference knowledge corresponding to the sample problem and second sample reference knowledge, the first sample reference knowledge comprises knowledge capable of solving the sample problem, and the second sample reference knowledge does not comprise knowledge capable of solving the sample problem; Performing vector conversion processing on the initial training sample by adopting the embedded characterization model to obtain a training sample, and constructing a training sample set based on the training sample; and training an initial neural network model based on the training sample set to obtain the target neural network model.
6. 6. The method according to claim 5, wherein the initial training samples in the initial training sample set further comprise third sample reference knowledge corresponding to the sample problem, wherein sample problems other than the sample problem are determined as target sample problems, and the third sample reference knowledge is randomly selected from the first sample reference knowledge and/or the second sample reference knowledge corresponding to the target sample problem.
7. 7. The method of claim 6, wherein training an initial neural network model based on the training set of samples, the target neural network model comprising: for training samples in the training sample set, vector updating processing is carried out on vectors corresponding to the sample problems in the training samples through the initial neural network model, so that a first vector is obtained; Vector updating processing is carried out on vectors corresponding to sample reference knowledge in the training samples through the initial neural network model, so that second vectors are obtained, wherein the sample reference knowledge comprises the first sample reference knowledge, the second sample reference knowledge and the third sample reference knowledge; calculating a loss function value of the initial neural network model according to the similarity between the first vector and the second vector; and optimizing the initial neural network model based on the loss function value so as to obtain the target neural network model under the condition that the loss function value of the optimized initial neural network model is smaller than a preset threshold value.
8. 8. The method of claim 5, wherein the reference knowledge of the target knowledge domain comprises a first reference knowledge belonging to a question-answer type and a second reference knowledge belonging to a non-question-answer type, and wherein obtaining the initial training sample set comprises: determining the problem in the first reference knowledge as a sample problem, and performing vector conversion processing on the sample problem by adopting the embedded characterization model to obtain a third vector; carrying out knowledge search in the second reference knowledge according to the third vector to obtain a plurality of target second reference knowledge; and determining the initial training sample according to the first reference knowledge and a plurality of target second reference knowledge corresponding to the first reference knowledge so as to obtain the initial training sample set.
9. 9. The method of claim 8, wherein determining the initial training sample from the first reference knowledge and a plurality of target second reference knowledge corresponding to the first reference knowledge comprises: Constructing a plurality of reference knowledge sets according to the plurality of target second reference knowledge, wherein the reference knowledge sets comprise part of target second reference knowledge in the plurality of target second reference knowledge; inputting the sample problem and the reference knowledge set into a large language model, and determining reply information aiming at the sample problem through the large language model to obtain first reply information corresponding to the reference knowledge set; Determining real reply information corresponding to the sample problem from the first reference knowledge, and outputting a processing result through the large language model according to the first reply information and the real reply information, wherein the processing result is used for representing whether the first reply information can solve the sample problem; And determining the initial training sample according to the processing results corresponding to the multiple reference knowledge sets.
10. 10. The method of claim 9, wherein constructing a plurality of reference knowledge sets from the plurality of target second reference knowledge sets comprises: Inputting the sample problem and the plurality of target second reference knowledge into the large language model, and determining reply information aiming at the sample problem through the large language model to obtain second reply information; Outputting a target result according to the second reply information and the real reply information through the large language model, wherein the target result is one of a first target result which represents that the second reply information can solve the sample problem and a second target result which represents that the second reply information cannot solve the sample problem; and under the condition that the target result is the first target result, constructing a plurality of reference knowledge sets according to the plurality of target second reference knowledge sets.
11. 11. The method of claim 9, wherein the processing results are one of a first processing result characterizing that the first reply message fails to solve the sample problem, a second processing result characterizing that the first reply message fails to solve the sample problem, and determining the initial training sample based on the processing results corresponding to the plurality of reference knowledge sets comprises: For a reference knowledge set of the plurality of reference knowledge sets, determining a target second reference knowledge that is absent from the reference knowledge set relative to the plurality of target second reference knowledge; determining the absent target second reference knowledge as the first sample reference knowledge under the condition that the processing result corresponding to the reference knowledge set is the first processing result; Determining the absent target second reference knowledge as the second sample reference knowledge under the condition that the processing result corresponding to the reference knowledge set is the second processing result; determining the initial training sample according to the sample problem, the first sample reference knowledge and the second sample reference knowledge.
12. 12. An information processing method, characterized by comprising: Acquiring a target problem which is uploaded by a client and relates to the target knowledge field; Performing vector conversion processing on the target problem in a cloud server by adopting an embedded characterization model to obtain an initial vector, performing vector update processing on the initial vector through a target neural network model to obtain a target vector, performing knowledge search in reference knowledge of the target knowledge field according to the target vector, and determining target reply information corresponding to the target problem according to a search result, wherein the target neural network model is obtained based on reference knowledge training of the target knowledge field; And feeding back the target reply information to the client.
13. 13. An information processing apparatus, characterized by comprising: A first acquisition unit configured to acquire a target problem related to a target knowledge field; The first processing unit is used for carrying out vector conversion processing on the target problem by adopting an embedded characterization model to obtain an initial vector; The second processing unit is used for carrying out vector updating processing on the initial vector through a target neural network model to obtain a target vector, wherein the target neural network model is obtained based on reference knowledge training in the target knowledge field; and the determining unit is used for searching knowledge in the reference knowledge of the target knowledge field according to the target vector and determining target reply information corresponding to the target problem according to a search result.
14. 14. An electronic device, comprising: A memory storing an executable program; a processor for executing the program, wherein the program executes the information processing method according to any one of claims 1 to 12 when executed.
15. 15. A computer-readable storage medium, characterized in that the computer-readable storage medium comprises a stored executable program, wherein the executable program, when run, controls a device in which the storage medium is located to perform the information processing method of any one of claims 1 to 12.
16. 16. A computer program product comprising a computer program or instructions which, when executed by a processor, implement an information processing method according to any one of claims 1 to 12.

Description

Information processing method, information processing device, computer readable storage medium and electronic equipment Technical Field The present application relates to the field of artificial intelligence, and in particular, to an information processing method, an information processing apparatus, a computer readable storage medium, and an electronic device. Background In recent years, with the continuous evolution of artificial intelligence technology, the application of intelligent question-answering technology is quite hot. One typical scenario for intelligent questions and answers is a territorial knowledge question and answer. That is, an intelligent question-answering system is constructed that can understand the proprietary knowledge of an enterprise or organization in its proprietary domain and answer related questions. Under the scene, when the intelligent question-answering system receives a question of a user, the intelligent question-answering system firstly uses an embedded representation (embedding) model to understand and represent the question text of the user, converts the question text into a digital vector which can be understood and calculated by a computer, and then gives an answer to the question of the user based on the digital vector and proprietary knowledge of the proprietary field. At present, a general embedded characterization model obtained based on public data training is generally adopted in the related technology, and for a model user (such as a management mechanism of an intelligent question-answering system), the embedded characterization model is generally black-boxed, a user cannot contact model parameters and cannot fine tune the model, so that the accuracy of vectors obtained when the general embedded characterization model is adopted to process domain knowledge is low, and the problem of low accuracy of the domain knowledge questions and answers exists. In view of the above problems, no effective solution has been proposed at present. Disclosure of Invention The embodiment of the application provides an information processing method, an information processing device, a computer readable storage medium and electronic equipment, which are used for at least solving the technical problem that in a question-answering scene of the related technology, under the condition that a general embedded characterization model cannot be finely tuned, the accuracy of a vector obtained by processing the domain knowledge by adopting the general embedded characterization model is low, so that the accuracy of the domain knowledge question-answering is low. According to one aspect of the embodiment of the application, an information processing method is provided, which comprises the steps of obtaining a target problem related to the target knowledge field, carrying out vector conversion processing on the target problem by adopting an embedded characterization model to obtain an initial vector, carrying out vector update processing on the initial vector through a target neural network model to obtain a target vector, wherein the target neural network model is obtained by training based on reference knowledge of the target knowledge field, carrying out knowledge search in the reference knowledge of the target knowledge field according to the target vector, and determining target reply information corresponding to the target problem according to a search result. Further, the knowledge searching in the reference knowledge of the target knowledge field according to the target vector comprises the steps of obtaining a knowledge vector of the reference knowledge of the target knowledge field, determining target reference knowledge from the reference knowledge of the target knowledge field according to the similarity between the target vector and the knowledge vector, and determining a search result according to the target reference knowledge. Further, obtaining the knowledge vector of the reference knowledge in the field of target knowledge comprises the steps of carrying out vector conversion processing on the reference knowledge by adopting an embedded characterization model to obtain an initial knowledge vector, and carrying out vector update processing on the initial knowledge vector by using a target neural network model to obtain the knowledge vector. Further, the search result comprises at least one target reference knowledge, and determining target reply information corresponding to the target problem according to the search result comprises the steps of inputting the target problem and the at least one target reference knowledge into a large language model, and determining the target reply information corresponding to the target problem through the large language model. The target neural network model is further obtained by obtaining an initial training sample set, wherein the initial training samples in the initial training sample set at least comprise sample problems in the target kn