Information processing method and device, electronic equipment and storage medium

Abstract

The invention provides an information processing method and device, electronic equipment and a storage medium, and can be applied to the field of financial science and technology businesses, and the method comprises the steps: inputting target question information into a pre-trained question and answer information processing model, so as to carry out the analysis of the target question information based on thinking chain prompt, obtaining candidate answer information and target thinking chain information; performing thinking chain information retrieval for the target question information through a preset knowledge base, and performing thinking chain verification on the target thinking chain information through thinking chain verification information obtained through retrieval; if the target thinking chain information verification fails, generating first prompt information according to the thinking chain verification information, and inputting the target question information and the first prompt information into a question and answer information processing model to perform thinking chain verification on the re-obtained target thinking chain information until the thinking chain verification is passed; and outputting an answer according to the finally obtained candidate answer information. According to the embodiment of the invention, the accuracy of the answers output by the model can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an information processing method, device, electronic device and storage medium. Background Art

[0002] In related technologies, some companies in the financial industry have introduced large language models (LLMs) to provide question-answering services for relevant practitioners or customer groups. During the operation of LLMs, the Chain-Of-Thought (CoT) method can be used to perform semantic analysis on relevant question information and generate answers to induce LLM to generate intermediate reasoning steps.

[0003] However, since the thought chain is information spontaneously induced within the LLM, this process is uncontrollable, so errors are inevitable, making the reasoning chain inconsistent with the facts and untrue, which affects the accuracy of the answers generated by the LLM based on the thought chain. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose an information processing method, device, electronic device and storage medium, aiming to improve the accuracy of the generated thought chain so as to improve the accuracy of the answers output by the model.

[0005] To achieve the above object, a first aspect of an embodiment of the present application proposes an information processing method, comprising the following steps: Inputting the target question information into a pre-trained question-answering information processing model, so that the question-answering information processing model analyzes the target question information based on the thought chain prompt, and obtains candidate answer information and target thought chain information of the target question information; Retrieving the thinking chain information for the target question information through a preset knowledge base to obtain thinking chain verification information, and performing thinking chain verification on the target thinking chain information through the thinking chain verification information; If the target thinking chain information verification fails, first prompt information is generated according to the thinking chain verification information, and the target question information and the first prompt information are input into the question-answer information processing model to perform thinking chain verification on the target thinking chain information obtained again, until the target thinking chain information obtained passes the thinking chain verification; The answer to the target question information is output according to the candidate answer information finally obtained.

[0006] In some embodiments, before outputting the answer to the target question information, the method further includes: If the currently obtained target thought chain information passes the thought chain information verification, the text information of the target question information is retrieved through the knowledge base to obtain answer verification text information, and the candidate answer information is verified through the answer verification text information; If the candidate answer information verification fails, a second prompt information is generated based on the answer verification text information, and the target question information and the second prompt information are input into the question and answer information processing model to perform thinking chain verification on the retrieved target thinking chain information until the candidate answer information obtained passes the answer verification.

[0007] In some embodiments, the thought chain verification information is one or more; The performing thought chain verification on the target thought chain information through the thought chain verification information includes: Calculating a first similarity between the obtained thought chain verification information and the target thought chain information; When the first similarity corresponding to any of the obtained thought chain verification information is less than a first preset threshold, it is determined that the verification of the target thought chain information has failed.

[0008] In some embodiments, the answer verification text information is one or more; The step of verifying the candidate answer information by using the answer verification text information includes: Calculating a second similarity between the obtained answer verification text information and the candidate answer information; When the second similarity corresponding to any of the obtained answer verification text information is less than a second preset threshold, it is determined that the candidate answer information verification has failed.

[0009] In some embodiments, the target thought chain information includes a target subject, a target object, and a target relationship between the target subject and the target object, and the thought chain verification information includes a verification subject, a verification object, and a verification relationship between the verification subject and the verification object; The generating the first prompt information according to the thought chain verification information includes: When there are multiple thought chain verification information, calculate the third similarity between the target subject and each verification subject, and determine the verification subject corresponding to the largest third similarity as the prompt subject; Calculating a fourth similarity between the target object and each of the verification objects, and determining the verification object corresponding to the largest fourth similarity as a prompt object; Calculating the fifth similarity between the target relationship and each of the verification relationships, and determining the verification relationship corresponding to the largest fifth similarity as the prompt relationship; Information is constructed according to the prompt subject, the prompt object and the prompt relationship to generate the first prompt information.

[0010] In some embodiments, before calculating the first similarity, performing thought chain verification on the target thought chain information through the thought chain verification information further includes: If the current thinking chain verification is not the first thinking chain verification for the target question information, obtaining the thinking chain verification information retrieved from the knowledge base in the last thinking chain verification; Calculate the sixth similarity between the thinking chain verification information retrieved from the knowledge base in this thinking chain verification and the thinking chain verification information retrieved from the knowledge base in the last thinking chain verification. When the sixth similarity is less than the third preset threshold, determine that the verification of the target thinking chain information has failed.

[0011] In some embodiments, before calculating the second similarity, verifying the candidate answer information by using the answer verification text information includes: If the current answer verification is not the first answer verification for the target question information, obtaining the answer verification text information retrieved from the knowledge base in the last answer verification; Calculate the seventh similarity between the answer verification text information retrieved from the knowledge base for this answer verification and the answer verification text information retrieved from the knowledge base for the last answer verification. When the seventh similarity is less than the fourth preset threshold, determine that the candidate answer information verification has failed.

[0012] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application provides an information processing device, including: A question information parsing module, used for inputting target question information into a pre-trained question-answer information processing model, so that the question-answer information processing model performs parsing on the target question information based on a thought chain prompt, and obtains candidate answer information and target thought chain information of the target question information; A thinking chain verification module is used to retrieve the thinking chain information for the target question information through a preset knowledge base to obtain thinking chain verification information, and to verify the target thinking chain information through the thinking chain verification information; a re-analysis module, for generating first prompt information according to the thought chain verification information if the verification of the target thought chain information fails, and inputting the target question information and the first prompt information into the question-answer information processing model to perform thought chain verification on the re-obtained target thought chain information, until the obtained target thought chain information passes the thought chain verification; The answer output module is used to output the candidate answer information finally obtained as the answer to the target question information.

[0013] To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0014] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0015] The present application proposes an information processing method, device, electronic device and storage medium, the method comprising: inputting target question information into a pre-trained question and answer information processing model, so that the question and answer information processing model performs analysis on the target question information based on a thinking chain prompt, and obtains candidate answer information and target thinking chain information of the target question information; performing a thinking chain information search for the target question information through a preset knowledge base, and obtains thinking chain verification information, and performs a thinking chain verification on the target thinking chain information through the thinking chain verification information; if the verification of the target thinking chain information fails, generating a first prompt information according to the thinking chain verification information, and inputting the target question information and the first prompt information into the question and answer information processing model to perform a thinking chain verification on the target thinking chain information obtained again, until the target thinking chain information obtained passes the thinking chain verification; and outputting the answer to the target question information according to the candidate answer information finally obtained. By retrieving the thinking chain information for the target question information through the knowledge base, the thinking chain verification information related to the target question information can be retrieved from the large amount of information related to the target question information stored in the knowledge base. In this way, a strong reference can be provided for the target thinking chain information output by the question and answer information processing model. Then, the thinking chain verification of the target thinking chain information can be performed to determine whether the intermediate reasoning steps generated by the question and answer information processing model for the target question information are correct. When the verification of the target thinking chain information fails, the question and answer information processing model can be input into the question and answer information processing model. The target question information and the first prompt information generated based on the thinking chain verification information can be input into the question and answer information processing model. The question and answer information processing model can enter an iterative reasoning process, so that the question and answer information processing model can adjust the intermediate reasoning steps generated for the target question information based on the first prompt information in each iterative reasoning process, so that the question and answer information processing model can infer more accurate target thinking chain information, so that the question and answer information processing model can gradually infer more accurate answer information based on more accurate target thinking chain information, which is conducive to improving the accuracy of the answers output by the question and answer information processing model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of an information processing method provided by an embodiment of the present application; Figure 2 is a flow chart of an information processing method provided by another embodiment of the present application; Figure 3 yes Figure 1 A flowchart of an embodiment of a sub-step of step 120; Figure 4 yes Figure 2 A flowchart of an embodiment of a sub-step of step 210; Figure 5yes Figure 1 A flowchart of an embodiment of a sub-step of step 130; Figure 6 yes Figure 1 A flowchart of another sub-step embodiment of step 120; Figure 7 yes Figure 2 A flowchart of another sub-step embodiment of step 210; Figure 8 is a schematic diagram of the structure of an information processing device provided in an embodiment of the present application; Fig. 9 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0018] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0020] First, some nouns involved in this application are analyzed: Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0021] Natural language processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can enable effective communication between people and computers using natural language. Natural language processing involves natural language, which is the language people use in daily life, and is closely related to linguistics research; it also involves computer science and mathematics. The pre-training model, an important technology for model training in the field of artificial intelligence, is developed from the LLM in the field of NLP. After fine-tuning, the large language model can be widely used in downstream tasks. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question answering, knowledge graph and other technologies.

[0022] In related technologies, some companies in the financial industry have introduced large language models (LLMs) to provide question-and-answer services for relevant practitioners or customer groups. During the operation of LLMs, the Chain-Of-Thought (CoT) method can be used to perform semantic analysis on relevant question information and generate answers to induce LLM to generate intermediate reasoning steps. However, since the chain of thought is information spontaneously induced and generated within the LLM, this process is not controllable, so errors are inevitable, making the reasoning chain inconsistent with the facts and untrue, which affects the accuracy of the answers generated by LLM based on the chain of thought.

[0023] In order to improve the accuracy of the generated thinking chain and to improve the accuracy of the answers output by the model, the embodiments of the present application provide an information processing method, an information processing device, an electronic device and a computer-readable storage medium. The method can first input the target question information into a pre-trained question and answer information processing model, so that the question and answer information processing model parses the target question information based on the thinking chain prompt, obtains the candidate answer information and the target thinking chain information of the target question information, then retrieves the thinking chain information for the target question information through a preset knowledge base, obtains the thinking chain verification information, and performs thinking chain verification on the target thinking chain information through the thinking chain verification information. If the verification of the target thinking chain information fails, a first prompt information is generated according to the thinking chain verification information, and the target question information and the first prompt information are input into the question and answer information processing model to perform thinking chain verification on the retrieved target thinking chain information until the obtained target thinking chain information passes the thinking chain verification; and outputs the answer to the target question information according to the candidate answer information finally obtained. By retrieving the thinking chain information for the target question information through the knowledge base, the thinking chain verification information related to the target question information can be retrieved from the large amount of information related to the target question information stored in the knowledge base. In this way, a strong reference can be provided for the target thinking chain information output by the question and answer information processing model. Then, the thinking chain verification of the target thinking chain information can be performed to determine whether the intermediate reasoning steps generated by the question and answer information processing model for the target question information are correct. When the verification of the target thinking chain information fails, the question and answer information processing model can be input into the question and answer information processing model. The target question information and the first prompt information generated based on the thinking chain verification information can be input into the question and answer information processing model. The question and answer information processing model can enter an iterative reasoning process, so that the question and answer information processing model can adjust the intermediate reasoning steps generated for the target question information based on the first prompt information in each iterative reasoning process, so that the question and answer information processing model can infer more accurate target thinking chain information, so that the question and answer information processing model can gradually infer more accurate answer information based on more accurate target thinking chain information, which is conducive to improving the accuracy of the answers output by the question and answer information processing model.

[0024] An information processing method, device, electronic device and storage medium provided in the embodiments of the present application are specifically described through the following embodiments. First, the information processing method in the embodiments of the present application is described.

[0025] The information processing method provided in the embodiment of the present application relates to the field of financial technology. The information processing method provided in the embodiment of the present application can be applied to a terminal, can be applied to a server side, or can be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the information processing method, etc., but is not limited to the above forms.

[0026] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0027] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0028] See also Figure 1 , Figure 1 The flow chart of the information processing method provided by an embodiment of the present application is shown. In the embodiment of the present application, Figure 1 The information processing method shown may include steps 110 to 140 .

[0029] Step 110: input the target question information into the pre-trained question-answering information processing model, so that the question-answering information processing model analyzes the target question information based on the thought chain prompt, and obtains candidate answer information and target thought chain information of the target question information; Step 120: searching for thinking chain information for target question information through a preset knowledge base to obtain thinking chain verification information, and verifying the target thinking chain information through the thinking chain verification information; Step 130: if the target thinking chain information verification fails, generate first prompt information according to the thinking chain verification information, input the target question information and the first prompt information into the question-answer information processing model to perform thinking chain verification on the target thinking chain information obtained again, until the obtained target thinking chain information passes the thinking chain verification; Step 140: Output the answer to the target question information based on the candidate answer information finally obtained.

[0030] In one embodiment, the target question information refers to text information including questions asked by the user. The target question information may be information input by the user, or information generated based on the information input by the user and a preset question template, etc., which is not specifically limited here.

[0031] In one embodiment, the question-answering information processing model refers to a large language model that can generate an answer text corresponding to the input question information based on a thought chain method. Among them, the question-answering information processing model can be obtained by training one or more initial models such as ChatGLM, ERNIE, Qwen, CILab, etc., which are not specifically limited here. In addition, a thought chain refers to an information set that can represent a specific subject, a specific object, and the relationship between a specific subject and a specific object. Among them, the subject and the object are both entities, and the entity can be a sentence, a word, etc., which are not specifically limited here.

[0032] In one embodiment, parsing the target question information based on the thought chain prompt refers to the operation that the question and answer information processing model obtains the target thought chain information by analyzing the subject, object and the relationship between the subject and the object included in the target question information, performs text semantic analysis on the target question information according to the target thought chain information, and generates a response text (i.e., candidate answer information) according to the text semantic analysis result. Among them, the target thought chain information refers to an information set including the target subject, target object and the logical relationship between the subject and the object included in the target question information, among which the target subject refers to the entity as the subject in the target thought chain information, the target object refers to the entity as the object in the target thought chain information, the target relationship refers to the relationship between the target subject and the target object, and the target thought chain information can be in one of the specific forms such as triple form, dictionary form, sequence form, etc., which is not specifically limited here. In addition, candidate answer information refers to text information output by the question and answer information processing model that includes answers to target question information, among which the candidate answer information can be output by the question and answer information processing model based on the target question information input separately, or can be output by the target question information input by the question and answer information processing model and the first prompt information (that is, the first prompt information mentioned below, which refers to a more accurate thinking chain information than the target thinking chain information currently obtained), and there is no specific limitation here.

[0033] In one embodiment, the knowledge base (KB) refers to an information system of structured knowledge including multiple subjects, multiple objects, and thought chain information formed by the relationship between each subject and one or more objects in the field associated with the target question information. In the preset knowledge base, the relationship between a subject and an object can be one or more, which is not specifically limited here.

[0034] In one embodiment, the question-answer information processing model can be trained in the following manner: first, a plurality of training question information, a plurality of first training thinking chain information, a plurality of annotated thinking chain information, and a plurality of standard answer information are obtained from the knowledge base; then, the plurality of training question information are respectively input into the initial model, and a plurality of second training thinking chain information and a plurality of first training answer information are correspondingly obtained; then, the initial model is first parameterized according to the plurality of annotated thinking chain information, the plurality of standard answer information, the plurality of second training thinking chain information, and the plurality of first training answer information, to obtain the initial model after the first parameterization; then, the initial model after the first parameterization is continued to be adjusted. Continue training to directly meet the first preset condition; then, input multiple training question information and multiple first training thinking chain information into the initial model finally obtained after the first parameter adjustment, and obtain multiple third training thinking chain information and multiple second training answer information accordingly. Perform second parameter adjustment on the current initial model according to multiple annotated thinking chain information, multiple standard answer information, multiple third training thinking chain information and multiple second training answer information, and then continue to train the initial model after the second parameter adjustment to directly meet the second preset condition, and determine the initial model finally obtained after the second parameter adjustment as the question and answer information processing model. By obtaining training data and annotated data from the knowledge base, the quality of the training data can be improved, thereby improving the accuracy and authenticity of the model in complex reasoning tasks.

[0035] In one embodiment, a thought chain information retrieval for target question information is performed through a preset knowledge base to obtain thought chain verification information, which means that the knowledge base identifies the entity in the target prompt information, and then searches the internally stored knowledge for one or more thought chain verification information corresponding to the identified entity. Among them, the thought chain verification information refers to the thought chain information corresponding to the identified entity in the thought chain information stored in the knowledge base.

[0036] In one embodiment, in the process of retrieving thought chain information for target question information through a preset knowledge base to obtain thought chain verification information, the knowledge base can specifically first extract feature information of the target question information to obtain text features of the target question information, and then perform entity analysis based on the text features to determine the entities existing in the target question information, and then use the determined entities as search keywords to determine the thought chain verification information of the target question information in the stored thought chains.

[0037] In one embodiment, the thought chain verification refers to the operation of verifying the confidence of the target thought chain information through the thought chain verification information corresponding to the target question information. The thought chain verification information may be one or more, which is not specifically limited here.

[0038] In one embodiment, in the process of verifying the target thinking chain information through the thinking chain verification information, the semantic information of the target relationship between the target subject and the target object in the target thinking chain information can be obtained first, and then the semantic information of the verification relationship between the verification subject and the verification object in the thinking chain verification information can be obtained. Then, the semantic information of the target relationship between the target subject and the target object and the semantic information of the verification relationship between the verification subject and the verification object are compared for semantic consistency, and whether the target thinking chain information passes the thinking chain verification is determined according to the semantic consistency result. Among them, the target subject can be one of the subject and the object indicated in the target question information, and the target object can be one of the subject and the object indicated in the target question information, which is not specifically limited here. In addition, the verification subject refers to the entity as the subject in the target question information, the verification object refers to the entity as the object in the target question information, and the verification relationship refers to the relationship between the verification subject and the verification object, among which the verification subject can be one of the target subject and the target object, and the verification object can be one of the target subject and the target object, which is not specifically limited here.

[0039] In one embodiment, in the process of generating the first prompt information according to the chain of thought verification information, when the chain of thought verification information retrieved is one, the text generation based on the preset prompt information text template can be performed according to the verification subject, verification object and verification relationship contained in the chain of thought verification information to obtain the first prompt information; when the chain of thought verification information retrieved is multiple, the chain of thought verification information with the highest similarity with the target chain of thought information can be determined first, and then, according to the verification subject, verification object and verification relationship contained in the chain of thought verification information corresponding to the highest similarity, the text generation based on the preset prompt information text template is performed to obtain the first prompt information. Among them, the first prompt information refers to the prompt information (prompt) that uses the verification subject, verification object and verification relationship as prompt words when the verification of the target chain of thought information fails.

[0040] In one embodiment, inputting the target question information and the first prompt information into the question-answer information processing model to perform a thinking chain verification on the target thinking chain information that is retrieved means that the target question information and the first prompt information are input into the question-answer information processing model, so that the question-answer information processing model performs a thinking chain prompt-based analysis on the target question information and the first prompt information to retrieve the candidate answer information and the target thinking chain information, and then retrieves the thinking chain information for the target question information through the knowledge base to obtain the thinking chain verification information, and performs a thinking chain verification operation on the retrieved target thinking chain information through the retrieved thinking chain verification information. It should be noted that when the target question information is first input into the question-answer information processing model, the first prompt information is not generated, that is, in the process of the candidate answer information and the target thinking chain information of the target question information that are first output by the question-answer information processing model, the analysis is not performed with the help of the first prompt information, and only in the subsequent repeated thinking chain verification process after the verification fails will the analysis be performed with the help of the first prompt information.

[0041] In one embodiment, until the target thinking chain information obtained passes the thinking chain verification means that if the regenerated target thinking chain information still fails to be verified, the first prompt information is regenerated according to the regenerated thinking chain verification information, and the regenerated first prompt information and the target question information are input into the question and answer information processing model, and the steps of generating candidate answer information and target thinking chain information and verifying the thinking chain are executed in a loop until the latest target thinking chain information obtained passes the thinking chain verification.

[0042] In one embodiment, when the target question information and the first prompt information are input into the question and answer information processing model together, as long as the target thinking chain information finally obtained passes the thinking chain verification, the above-mentioned execution loop can be jumped out, and the answer to the target question information can be output according to the candidate answer information finally obtained.

[0043] In one embodiment, if the target question information is first input into the question and answer information processing model and the target thinking chain information obtained passes the thinking chain verification, then there is no need to enter the cycle of executing candidate answer information and target thinking chain information generation and thinking chain verification, and the answer to the target question information can be directly output based on the obtained candidate answer information.

[0044] In one embodiment, in the process of outputting the answer to the target question information according to the candidate answer information finally obtained, the candidate answer information finally obtained may be output as the answer to the target question information.

[0045] In one embodiment, in the process of outputting the answer to the target question information according to the candidate answer information finally obtained, specifically, a text search can be first performed in the knowledge base according to the target question information and the target thinking chain information to obtain the comparative answer information, and then the text quality of the candidate answer information and the comparative answer information is compared based on semantic features. When the text quality of the candidate answer information is better than the comparative answer information, the candidate answer information is output as the answer to the target question information, and when the text quality of the comparative answer information is better than the candidate answer information, the candidate answer information is output as the answer to the target question information. The comparative answer information refers to the answer information stored in the knowledge base for the target subject, target object and target relationship.

[0046] See also Figure 2 , Figure 2 The flowchart of an information processing method provided by another embodiment of the present application is shown. In one embodiment, before step 140, the information processing method may further include the following steps.

[0047] Step 210: If the currently obtained target thinking chain information passes the thinking chain information verification, a text information search is performed on the target question information through the knowledge base to obtain answer verification text information, and answer verification is performed on the candidate answer information through the answer verification text information; Step 220: If the candidate answer information verification fails, generate a second prompt information based on the answer verification text information, input the target question information and the second prompt information into the question and answer information processing model to perform a thinking chain verification on the retrieved target thinking chain information until the obtained candidate answer information passes the answer verification.

[0048] In one embodiment, performing a text information search for target question information through a knowledge base to obtain answer verification text information refers to an operation of performing a text search based on a target subject and a target object of the target question information to obtain answer verification text information. The answer verification text information refers to text information in the knowledge base that includes a target subject and a target object as content entities, among which the answer verification text information can be text information of the answer to the target question information, or text information in which only text words of the target subject and the target object appear, etc., which is not specifically limited here.

[0049] In one embodiment, answer verification refers to the operation of verifying the confidence of candidate answer information through answer verification text information corresponding to the target question information. The answer verification text information may be one or more, which is not specifically limited here.

[0050] In one embodiment, in the process of verifying the answer of candidate answer information through answer verification text information, specifically, semantic understanding can be first performed based on the candidate answer information to obtain the semantic information of the candidate answer information, then, semantic understanding can be performed on the answer verification text information to obtain the semantic information of the answer verification text information, and then the semantic information of the candidate answer information and the semantic information of the answer verification text information are compared for semantic consistency, and whether the candidate answer information passes the answer verification is determined based on the semantic consistency result.

[0051] In one embodiment, in the process of generating the second prompt information according to the answer verification text information, when the answer verification text information retrieved is one, the verification subject, verification object and verification relationship in the answer verification text information can be specifically extracted first, and then, according to the verification subject, verification object and verification relationship in the answer verification text information, text generation based on the preset prompt information text template is performed to obtain the second prompt information; when the answer verification text information retrieved is multiple, the answer verification text information with the highest similarity with the candidate answer information is determined among the multiple answer verification text information, and then, the verification subject, verification object and verification relationship in the answer verification text information corresponding to the highest similarity are extracted, and then, according to the verification subject, verification object and verification relationship in the answer verification text information, text generation based on the preset prompt information text template is performed to obtain the second prompt information. Among them, the second prompt information refers to the prompt information (prompt) that uses the verification subject, verification object and verification relationship as prompt words when the answer verification fails, in which the second prompt information can be the same as the first prompt information in content, or it can be different, which is not specifically limited here.

[0052] In one embodiment, inputting the target question information and the second prompt information into the question-answer information processing model to perform a thinking chain verification on the target thinking chain information that is retrieved means that the target question information and the second prompt information are input into the question-answer information processing model, so that the question-answer information processing model performs a parsing of the target question information and the second prompt information based on the thinking chain prompt, and retrieves the candidate answer information and the target thinking chain information, then retrieves the thinking chain information for the target question information through the knowledge base, obtains the thinking chain verification information, and performs a thinking chain verification on the retrieved target thinking chain information through the retrieved thinking chain verification information. When the thinking chain verification is passed, the answer verification text information for the target question information is retrieved through the knowledge base to obtain the answer verification text information, and then performs an answer verification operation on the retrieved candidate answer information through the retrieved answer verification text information. It should be noted that if the target thinking chain information retrieved according to the target question information and the second prompt information still fails to be verified, the cycle of generating and verifying the candidate answer information and the target thinking chain information will be re-entered until the generated target thinking chain information passes the thinking chain verification.

[0053] In one embodiment, until the obtained candidate answer information passes the answer verification means that if the regenerated candidate answer information still fails to be verified, the second prompt information is regenerated according to the regenerated answer verification text information, the regenerated second prompt information and the target question information are input into the question and answer information processing model, and the steps of generating the candidate answer information and the target thinking chain information, verifying the thinking chain and verifying the answer are executed in a loop until the latest candidate answer information passes the answer verification.

[0054] In one embodiment, when the target question information and the second prompt information are input into the question and answer information processing model together, as long as the final candidate answer information passes the answer verification, the above-mentioned execution loop can be jumped out, and the answer to the target question information can be output based on the final candidate answer information.

[0055] In one embodiment, if the target question information is first input into the question and answer information processing model, the target thinking chain information obtained passes the thinking chain verification, and the candidate answer information page passes the answer verification, then there is no need to enter the cycle of executing the generation, thinking chain verification, and answer verification of the candidate answer information and the target thinking chain information, and the answer to the target question information can be directly output based on the obtained candidate answer information.

[0056] On the basis of thinking chain verification, text information retrieval for the target question information is performed through the knowledge base. The answer verification text information related to the target question information can be retrieved from a large amount of information related to the target question information stored in the knowledge base. In this way, a strong reference can be provided for the candidate answer information output by the question and answer information processing model, thereby ensuring the correctness of the target thinking chain while also ensuring the correctness of the candidate answers. When the target thinking chain information verification fails, the question and answer information processing model can be input with the target question information and the second prompt information generated based on the answer verification text information into the question and answer information processing model to enter an iterative reasoning process for answer information generation. In this way, the question and answer information processing model can infer more accurate answer information based on the second prompt information in each iterative reasoning process, which is conducive to reducing the situation where the reasoning chain is correct but the answer obtained by reasoning is wrong.

[0057] See also Figure 3 , Figure 3 Shows Figure 1 In the process of a sub-step embodiment of step 120, in one embodiment, the thought chain verification information is one or more, and step 120 may include the following steps.

[0058] Step 310: Calculate the first similarity between the obtained thought chain verification information and the target thought chain information; Step 320: When the first similarity corresponding to any obtained thought chain verification information is less than the first preset threshold, it is determined that the verification of the target thought chain information has failed.

[0059] In one embodiment, in the process of calculating the first similarity between the obtained thinking chain verification information and the target thinking chain information, the thinking chain verification information can be firstly feature extracted to obtain the embedded features of the thinking chain verification information, and then the target thinking chain information can be feature extracted to obtain the embedded features of the target thinking chain information, and then the first similarity is obtained by the similarity between the embedded features of the thinking chain verification information and the embedded features of the target thinking chain information. The first similarity can be calculated by calculating cosine similarity, Jaccard Similarity, Term Frequency-Inverse Document Frequency (TF-IDF), etc., which is not specifically limited here.

[0060] In one embodiment, in the process of calculating the first similarity between the obtained thought chain verification information and the target thought chain information, the subject similarity between the verification subject and the target subject can be calculated first, then the object similarity between the verification object and the target object can be calculated, and then the relationship similarity between the verification relationship and the target relationship can be calculated, and then, a weighted operation is performed based on the subject similarity, object similarity and relationship similarity to obtain the first similarity.

[0061] In one embodiment, when the first similarity corresponding to any obtained chain of thinking verification information is less than the first preset threshold, it is determined that the verification of the target chain of thinking information has failed. Specifically, when there is only one chain of thinking verification information, if the first similarity corresponding to the chain of thinking verification information is less than the first preset threshold, it is determined that the verification of the target chain of thinking information has failed; when there are multiple chain of thinking verification information, if the first similarity corresponding to each chain of thinking verification information is less than the first preset threshold, it is determined that the verification of the target chain of thinking information has failed.

[0062] In one embodiment, when there is one thought chain verification information, if the first similarity corresponding to the thought chain verification information is not less than the first preset threshold, it is determined that the target thought chain information passes the thought chain verification; when there are multiple thought chain verification information, if the first similarities corresponding to at least n thought chain verification information are not less than the first preset threshold, it is determined that the target thought chain information passes the thought chain verification, where n is a positive integer greater than or equal to 1, and is not specifically limited here.

[0063] By calculating the first similarity between the thinking chain verification information and the target thinking chain information, the credibility of the target thinking chain information can be judged by the first similarity. Since the thinking chain verification information is obtained from the knowledge base, the first similarity can reflect the correlation between the target thinking chain information and the knowledge base. Therefore, when the first similarity is less than the first preset threshold, it can be determined that the target thinking chain information is not relevant to the information in the knowledge base covering the target subject and the target object, and then it can be accurately determined that the target thinking chain information is unreliable, which is conducive to promoting the model to re-reason to obtain more credible target thinking chain information.

[0064] See also Figure 4 , Figure 4 Shows Figure 2 In the process of a sub-step embodiment of step 210, in one embodiment, the answer verification text information is one or more, and step 210 may include the following steps.

[0065] Step 410: Calculate the second similarity between the obtained answer verification text information and the candidate answer information; Step 420: When the second similarity corresponding to any obtained answer verification text information is less than the second preset threshold, it is determined that the candidate answer information verification has failed.

[0066] In one embodiment, in the process of calculating the second similarity between the obtained answer verification text information and the candidate answer information, the answer verification text information may be firstly feature extracted to obtain the embedded features of the answer verification text information, then the candidate answer information may be feature extracted to obtain the embedded features of the candidate answer information, and then the second similarity may be obtained by the similarity between the embedded features of the answer verification text information and the embedded features of the candidate answer information. The second similarity may be obtained by calculating cosine similarity, Jaccard similarity, TF-IDF, etc., which are not specifically limited here.

[0067] In one embodiment, when the second similarity corresponding to any obtained answer verification text information is less than the second preset threshold, it is determined that the candidate answer information verification has failed. Specifically, in the case where there is only one answer verification text information, if the second similarity corresponding to the answer verification text information is less than the second preset threshold, it is determined that the candidate answer information verification has failed; in the case where there are multiple answer verification text information, if the second similarity corresponding to each answer verification text information is less than the second preset threshold, it is determined that the candidate answer information verification has failed.

[0068] In one embodiment, when there is one answer verification text information, if the second similarity corresponding to the answer verification text information is not less than the second preset threshold, it is determined that the candidate answer information has passed the answer verification; when there are multiple answer verification text information, if the second similarities corresponding to at least n answer verification text information are not less than the second preset threshold, it is determined that the candidate answer information has passed the answer verification, where n is a positive integer greater than or equal to 1, and is not specifically limited here.

[0069] By calculating the second similarity between the answer verification text information and the candidate answer information, the credibility of the candidate answer information can be judged by the second similarity. Since the answer verification text information is obtained from the knowledge base, the second similarity can reflect the relevance of the candidate answer information and the knowledge base. Therefore, when the second similarity is less than the second preset threshold, it can be determined that the candidate answer information is not relevant to the information in the knowledge base covering the target subject and the target object, and thus the candidate answer information can be accurately determined to be unreliable, which is conducive to promoting the model to re-infer the candidate answer information with higher credibility.

[0070] See also Figure 5 , Figure 5 Shows Figure 1In the process of a sub-step embodiment of step 130, in one embodiment, step 130 may include the following steps.

[0071] Step 510: When there are multiple thought chain verification information, calculate the third similarity between the target subject and each verification subject, and determine the verification subject corresponding to the maximum third similarity as the prompt subject; Step 520: Calculate the fourth similarity between the target object and each verification object, and determine the verification object corresponding to the maximum fourth similarity as the prompt object; Step 530: Calculate the fifth similarity between the target relationship and each verification relationship, and determine the verification relationship corresponding to the maximum fifth similarity as the prompt relationship; Step 540: construct information according to the prompt subject, prompt object and prompt relationship to generate first prompt information.

[0072] In one embodiment, in the process of calculating the third similarity between the target subject and each verification subject, the target subject and each verification subject may be respectively extracted to obtain embedded features of the target subject and each verification subject, and then, for each verification subject, the third similarity between the embedded features of the verification subject and the embedded features of the target subject is calculated. The third similarity may be calculated by calculating cosine similarity, Jaccard similarity, TF-IDF, etc., which are not specifically limited here.

[0073] In one embodiment, in the process of calculating the fourth similarity between the target object and each verification object, the target object and each verification object may be respectively extracted to obtain embedded features of the target object and each verification object, and then, for each verification object, the fourth similarity between the embedded features of the verification object and the embedded features of the target object is calculated. The fourth similarity may be obtained by calculating cosine similarity, Jaccard similarity, TF-IDF, etc., which are not specifically limited here.

[0074] In one embodiment, in the process of calculating the fifth similarity between the target relationship and each verification relationship, the target relationship and each verification relationship may be respectively extracted to obtain the embedded features of the target relationship and each verification relationship, and then, for each verification relationship, the fifth similarity between the embedded features of the verification relationship and the embedded features of the target relationship is calculated. The fifth similarity may be obtained by calculating cosine similarity, Jaccard similarity, TF-IDF, etc., which are not specifically limited here.

[0075] In one embodiment, information is constructed according to a prompt subject, a prompt object, and a prompt relationship, and generating the first prompt information refers to taking the prompt subject, the prompt object, and the prompt relationship as entities of the prompt information, and generating the first prompt information including the prompt subject, the prompt object, and the prompt relationship based on a preset prompt information text template.

[0076] By determining the prompt subject that is most similar to the target subject, the target object that is most similar to the target object, and the prompt relationship that is most similar to the target relationship from multiple thought chain verification information, it is possible to construct the first prompt information that is closest to the target question information, thereby improving the prompt effect of the first prompt information in the model, and then in the process of the next iterative reasoning, it is possible to increase the probability of the model inferring the correct thought chain information, reduce the probability of needing to perform iterative reasoning again, and help improve the efficiency of question-answering reasoning.

[0077] See also Figure 6 , Figure 6 Shows Figure 1 In the process of another sub-step embodiment of step 120, in one embodiment, before step 310, step 120 may also include the following steps.

[0078] Step 610: if the current thinking chain verification is not the first thinking chain verification for the target question information, obtain the thinking chain verification information retrieved from the knowledge base in the last thinking chain verification; Step 620: Calculate the sixth similarity between the thinking chain verification information retrieved from the knowledge base in this thinking chain verification and the thinking chain verification information retrieved from the knowledge base in the last thinking chain verification. When the sixth similarity is less than the third preset threshold, it is determined that the verification of the target thinking chain information has failed.

[0079] In one embodiment, the first thought chain verification for the target question information refers to the first thought chain verification performed after the target question information is input into the question-answer information processing model separately.

[0080] In one embodiment, no matter whether the number of thought chain verification information obtained in the last thought chain verification is one or more, the thought chain verification information obtained in this thought chain verification can be the same or different in number, the content can overlap or not overlap, etc., which is not specifically limited here.

[0081] In one embodiment, in the process of calculating the sixth similarity between the thinking chain verification information retrieved from the knowledge base in this thinking chain verification and the thinking chain verification information retrieved from the knowledge base in the previous thinking chain verification, the embedded features of the thinking chain verification information retrieved from the knowledge base in this thinking chain verification can be extracted first, and then the embedded features of the thinking chain verification information retrieved from the knowledge base in the previous thinking chain verification can be extracted. Then, the similarity between the embedded features of the thinking chain verification information retrieved from the knowledge base in this thinking chain verification and the embedded features of the thinking chain verification information retrieved from the knowledge base in the previous thinking chain verification can be calculated (that is, the sixth similarity). Among them, when there are multiple thinking chain verification information retrieved from the knowledge base in this thinking chain verification, the features used to calculate the sixth similarity can be obtained by fusion of the embedded features of each thinking chain verification information in this thinking chain verification. When there is only one thinking chain verification information retrieved from the knowledge base in this thinking chain verification, the features used to calculate the sixth similarity can be the embedded features of the thinking chain verification information retrieved from the knowledge base in this thinking chain verification. In addition, when there are multiple pieces of thinking chain verification information retrieved from the knowledge base in the last thinking chain verification, the features used to calculate the sixth similarity can be obtained by fusion of the embedded features of each thinking chain verification information in the last thinking chain verification; when there is only one piece of thinking chain verification information retrieved from the knowledge base in the last thinking chain verification, the features used to calculate the sixth similarity can be the embedded features of the thinking chain verification information retrieved from the knowledge base in the last thinking chain verification.

[0082] In one embodiment, when the sixth similarity is not less than the third preset threshold, step 310 is executed.

[0083] In the process of answering a target question, when there are multiple chain verifications, by comparing the sixth similarity between the chain verification information retrieved from the knowledge base by this chain verification and the chain verification information retrieved from the knowledge base by the last chain verification, it can be determined whether the gap between the last chain reasoning process and the current chain reasoning process is too large. When the sixth similarity is less than the third preset threshold, it proves that one of the last chain reasoning process and the current chain reasoning process is unreliable, so that the model can be iterated again, and then the probability of the model inferring the correct chain information can be improved through iteration.

[0084] See also Figure 7 , Figure 7 Shows Figure 2 In the process of another sub-step embodiment of step 210, in one embodiment, before step 410, step 210 may include the following steps.

[0085] Step 710: If the current answer verification is not the first answer verification for the target question information, obtain the answer verification text information retrieved from the knowledge base in the last answer verification; Step 720: Calculate the seventh similarity between the answer verification text information retrieved from the knowledge base for this answer verification and the answer verification text information retrieved from the knowledge base for the last answer verification. When the seventh similarity is less than the fourth preset threshold, determine that the verification of the candidate answer information has failed.

[0086] In one embodiment, the first answer verification for the target question information refers to the answer verification performed after the target question information is input into the question and answer information processing model alone and passes the thinking chain verification for the first time. Specifically, when the target question information is input alone, the target thinking chain information output by the question and answer information processing model directly passes the thinking chain verification. At this time, the answer verification will begin. Then, this answer verification is the first answer verification for the target question information; when the target question information is input alone, it is necessary to execute one or more candidate answer information and target thinking chain information generation, as well as thinking chain verification cycles, until the thinking chain verification passes, jump out of the cycle, and start answer verification for the latest candidate answer information obtained. At this time, this answer verification is the first answer verification for the target question information.

[0087] In one embodiment, regardless of whether the answer verification text information obtained in the last answer verification was one or more, the answer verification text information obtained in this answer verification can be the same or different in number from the answer verification text information obtained in the last answer verification, and the content may overlap or not overlap, etc., which is not specifically limited here.

[0088] In one embodiment, in the process of calculating the seventh similarity between the answer verification text information retrieved from the knowledge base for this answer verification and the answer verification text information retrieved from the knowledge base for the previous answer verification, the embedded features of the answer verification text information retrieved from the knowledge base for this answer verification can be first extracted, then the embedded features of the answer verification text information retrieved from the knowledge base for the previous answer verification can be extracted, and then the similarity between the embedded features of the answer verification text information retrieved from the knowledge base for this answer verification and the embedded features of the answer verification text information retrieved from the knowledge base for the previous answer verification (i.e., the seventh similarity) can be calculated. Wherein, when there are multiple pieces of answer verification text information retrieved from the knowledge base for this answer verification, the features used to calculate the seventh similarity can be obtained by fusion of the embedded features of each answer verification text information in this answer verification, and when there is only one piece of answer verification text information retrieved from the knowledge base for this answer verification, the features used to calculate the seventh similarity can be the embedded features of the answer verification text information retrieved from the knowledge base for this answer verification. In addition, when there are multiple answer verification text information retrieved from the knowledge base in the last answer verification, the features used to calculate the seventh similarity can be obtained by fusion of the embedded features of each answer verification text information in the last answer verification; when there is only one answer verification text information retrieved from the knowledge base in the last answer verification, the features used to calculate the seventh similarity can be the embedded features of the answer verification text information retrieved from the knowledge base in the last answer verification.

[0089] In one embodiment, when the seventh similarity is not less than the fourth preset threshold, step 410 is executed.

[0090] In the process of answering a target question information, when the answer verification occurs multiple times, by comparing the answer verification text information retrieved from the knowledge base for this answer verification with the seventh similarity between the answer verification information retrieved from the knowledge base for the last answer verification, it can be determined whether the gap between the last answer reasoning process and the current answer reasoning process is too large. When the seventh similarity is less than the fourth preset threshold, it proves that one of the last answer reasoning process and the current answer reasoning process is unreliable, so that the model can be iterated again, and the probability of the model inferring the correct answer information can be improved through iteration.

[0091] An application example of the information processing method provided by the present application is given below. Assuming that the present method needs to be applied in a financial question-and-answer system, the financial question-and-answer system can first obtain a financial information question-and-answer request sent by a user, the financial information question-and-answer request carrying financial question information, then input the financial question information into a pre-trained financial question-and-answer information processing model, so that the financial question-and-answer information processing model parses the financial question information based on a thinking chain prompt to obtain candidate financial answer information and target financial thinking chain information of the financial question information, then retrieve financial thinking chain information for the financial question information through a preset financial knowledge base to obtain financial thinking chain verification information, perform thinking chain verification on the target financial thinking chain information through the financial thinking chain verification information, if the verification of the target financial thinking chain information fails, generate first prompt information according to the financial thinking chain verification information, input the financial question information and the first prompt information into the financial question-and-answer information processing model to perform thinking chain verification on the target financial thinking chain information obtained again, until the target financial thinking chain information obtained passes the thinking chain verification, and output the answer to the financial question information according to the candidate financial answer information finally obtained.

[0092] See also Figure 8 The embodiment of the present application further provides an information processing device, which can implement the information processing method of the first aspect above. The information processing device 800 includes: The question information parsing module 810 may be used to input the target question information into the pre-trained question-answering information processing model, so that the question-answering information processing model performs parsing on the target question information based on the thought chain prompt, and obtains candidate answer information and target thought chain information of the target question information; The thinking chain verification module 820 can be used to retrieve the thinking chain information for the target question information through the preset knowledge base, obtain the thinking chain verification information, and verify the thinking chain of the target thinking chain information through the thinking chain verification information; The re-analysis module 830 may be used to generate first prompt information according to the thinking chain verification information if the target thinking chain information verification fails, and input the target question information and the first prompt information into the question-answer information processing model to perform thinking chain verification on the re-obtained target thinking chain information, until the obtained target thinking chain information passes the thinking chain verification; The answer output module 840 may be used to output the candidate answer information finally obtained as the answer to the target question information.

[0093] The specific implementation of the information processing device is basically the same as the specific implementation of the above-mentioned information processing method, and will not be repeated here.

[0094] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above information processing method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0095] See also Fig. 9 , Fig. 9 The hardware structure of an electronic device of another embodiment is illustrated. The electronic device 900 includes: The processor 901 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application; The memory 902 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store an operating system and other application programs. When the technical solution provided in the embodiments of this specification is implemented by software or firmware, the relevant program code is stored in the memory 902, and the processor 901 calls and executes the information processing method of the embodiments of this application; Input / output interface 903, used to implement information input and output; Communication interface 904, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.); A bus 905 that transmits information between various components of the device (e.g., the processor 901, the memory 902, the input / output interface 903, and the communication interface 904); The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .

[0096] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program, and the computer program implements the above-mentioned information processing method when executed by a processor.

[0097] The memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0098] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0099] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0100] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0101] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0102] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0103] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0104] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0105] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0106] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0107] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0108] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. An information processing method, characterized in that: The following steps are involved: Inputting the target question information into a pre-trained question-answering information processing model, so that the question-answering information processing model analyzes the target question information based on the thought chain prompt, and obtains candidate answer information and target thought chain information of the target question information; Retrieving the thinking chain information for the target question information through a preset knowledge base to obtain thinking chain verification information, and performing thinking chain verification on the target thinking chain information through the thinking chain verification information; If the target thinking chain information verification fails, first prompt information is generated according to the thinking chain verification information, and the target question information and the first prompt information are input into the question-answer information processing model to perform thinking chain verification on the target thinking chain information obtained again, until the target thinking chain information obtained passes the thinking chain verification; The answer to the target question information is output according to the candidate answer information finally obtained.

2. The method according to claim 1, characterized in that Before outputting the answer to the target question information, the method further includes: If the currently obtained target thought chain information passes the thought chain information verification, the text information of the target question information is retrieved through the knowledge base to obtain answer verification text information, and the candidate answer information is verified through the answer verification text information; If the candidate answer information verification fails, a second prompt information is generated based on the answer verification text information, and the target question information and the second prompt information are input into the question and answer information processing model to perform thinking chain verification on the retrieved target thinking chain information until the candidate answer information obtained passes the answer verification.

3. The method according to claim 1, characterized in that The thought chain verification information is one or more; The performing thought chain verification on the target thought chain information through the thought chain verification information includes: Calculating a first similarity between the obtained thought chain verification information and the target thought chain information; When the first similarity corresponding to any of the obtained thought chain verification information is less than a first preset threshold, it is determined that the verification of the target thought chain information has failed.

4. The method according to claim 2, characterized in that: The answer verification text information is one or more; The step of verifying the candidate answer information by using the answer verification text information includes: Calculating a second similarity between the obtained answer verification text information and the candidate answer information; When the second similarity corresponding to any of the obtained answer verification text information is less than a second preset threshold, it is determined that the candidate answer information verification has failed.

5. The method according to claim 1, characterized in that The target thought chain information includes a target subject, a target object, and a target relationship between the target subject and the target object; the thought chain verification information includes a verification subject, a verification object, and a verification relationship between the verification subject and the verification object; The generating the first prompt information according to the thought chain verification information includes: When there are multiple thought chain verification information, calculate the third similarity between the target subject and each verification subject, and determine the verification subject corresponding to the largest third similarity as the prompt subject; Calculating a fourth similarity between the target object and each of the verification objects, and determining the verification object corresponding to the largest fourth similarity as a prompt object; Calculating the fifth similarity between the target relationship and each of the verification relationships, and determining the verification relationship corresponding to the largest fifth similarity as the prompt relationship; Information is constructed according to the prompt subject, the prompt object and the prompt relationship to generate the first prompt information.

6. The method according to claim 3, characterized in that: Before calculating the first similarity, performing thought chain verification on the target thought chain information through the thought chain verification information further includes: If the current thinking chain verification is not the first thinking chain verification for the target question information, obtaining the thinking chain verification information retrieved from the knowledge base in the last thinking chain verification; Calculate the sixth similarity between the thinking chain verification information retrieved from the knowledge base in this thinking chain verification and the thinking chain verification information retrieved from the knowledge base in the last thinking chain verification. When the sixth similarity is less than the third preset threshold, determine that the verification of the target thinking chain information has failed.

7. The method according to claim 4, characterized in that Before calculating the second similarity, the step of verifying the candidate answer information by using the answer verification text information includes: If the current answer verification is not the first answer verification for the target question information, obtaining the answer verification text information retrieved from the knowledge base in the last answer verification; Calculate the seventh similarity between the answer verification text information retrieved from the knowledge base for this answer verification and the answer verification text information retrieved from the knowledge base for the last answer verification. When the seventh similarity is less than the fourth preset threshold, determine that the candidate answer information verification has failed.

8. An information processing device, characterized in that: include: A question information parsing module, used for inputting target question information into a pre-trained question-answer information processing model, so that the question-answer information processing model performs parsing on the target question information based on a thought chain prompt, and obtains candidate answer information and target thought chain information of the target question information; A thinking chain verification module is used to retrieve the thinking chain information for the target question information through a preset knowledge base to obtain thinking chain verification information, and to verify the target thinking chain information through the thinking chain verification information; a re-analysis module, for generating first prompt information according to the thought chain verification information if the verification of the target thought chain information fails, and inputting the target question information and the first prompt information into the question-answer information processing model to perform thought chain verification on the re-obtained target thought chain information, until the obtained target thought chain information passes the thought chain verification; The answer output module is used to output the candidate answer information finally obtained as the answer to the target question information.

9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the information processing method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the information processing method according to any one of claims 1 to 7 is implemented.

Images