Information processing method, device and system

The method enhances answer accuracy by incorporating grammatical and semantic analysis to determine similar questions and their answers, addressing the limitations of current models that focus only on input question information.

CN119090016APending Publication Date: 2024-12-06LENOVO (BEIJING) LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202411214994.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Current information processing models focus solely on input question information, leading to less accurate and user-unfriendly answer generation.

Method used

An information processing method that determines similar question information, screens target information based on answer features, and generates answers using a combination of question and similar question answers, incorporating grammatical and semantic analysis.

Benefits of technology

Improves answer accuracy by considering similar questions and their answers, aligning output with user needs through enhanced grammatical and semantic screening.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090016A_ABST
    Figure CN119090016A_ABST
Patent Text Reader

Abstract

The invention discloses an information processing method, device and system, and the method comprises the steps: obtaining first problem information; determining second question information similar to the first question information; according to answer features of second answer information corresponding to the second question information, target information corresponding to the first question information is obtained through screening in multiple pieces of alternative information obtained through retrieval based on the first question information; and generating first answer information corresponding to the first question information according to the target information.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

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

[0002] Information question answering is a common application scenario of artificial intelligence-based data processing models. In this scenario, the model can receive question information input by users, process the question information, and output answer information for answering the question information.

[0003] The current data processing model only focuses on the input question information when generating answer information. Therefore, the generated answer information is less accurate and sometimes fails to meet user needs. Summary of the invention

[0004] To this end, this application discloses the following technical solutions:

[0005] The first aspect of the present application provides an information processing method, comprising:

[0006] Obtain the first question information;

[0007] Determining second question information similar to the first question information;

[0008] According to the answer feature of the second answer information corresponding to the second question information, the target information corresponding to the first question information is screened from the multiple candidate information retrieved based on the first question information;

[0009] First answer information corresponding to the first question information is generated according to the target information.

[0010] Optionally, the second question information and the second answer information are stored in a knowledge base, and determining the second question information similar to the first question information includes:

[0011] Acquire first grammatical information of the first question information, where the first grammatical information indicates a structural relationship between sub-contents contained in the first question information in the first question information; the sub-contents are phrases, words or characters;

[0012] According to the similarity between the first grammatical information and the grammatical information corresponding to each piece of question information in the knowledge base, second question information similar to the first question information is searched and obtained.

[0013] Optionally, the answer feature of the second answer information indicates a structural relationship of sub-contents contained in the second answer information in the second answer information;

[0014] The step of selecting, according to the answer feature, a plurality of candidate information retrieved based on the first question information to obtain target information corresponding to the first question information includes:

[0015] Determining whether a structural relationship of sub-contents contained in the candidate information retrieved based on the first question information in the candidate information is consistent with a structural relationship indicated by the answer feature;

[0016] The candidate information whose structural relationship of the sub-contents contained therein and the structural relationship indicated by the answer feature satisfy the consistency condition is determined as the target information corresponding to the first question information.

[0017] Optionally, the multiple candidate information retrieved based on the first question information includes:

[0018] Extracting a plurality of key phrases having a target grammatical relationship from the first question information, wherein the target grammatical relationship represents a structural relationship between a plurality of phrases in the key phrases;

[0019] A plurality of candidate information is retrieved according to the key phrase.

[0020] Optionally, also include:

[0021] Obtaining at least one keyword of the first question information determined according to semantics;

[0022] determining key elements based on the key phrases and the keywords;

[0023] The retrieving a plurality of candidate information according to the key phrase includes:

[0024] A plurality of candidate information is retrieved based on the key elements.

[0025] Optionally, determining the key elements according to the key phrases and the keywords includes:

[0026] Determining, according to the first question information, the contribution of the key phrase and the contribution of the keyword, wherein the contribution represents the strength of the key phrase and the keyword in processing the first question information;

[0027] Key elements are obtained by screening the key phrases and the keywords according to the contribution.

[0028] Optionally, the retrieving multiple candidate information according to the key elements includes at least one of the following:

[0029] Retrieving information containing the key elements as candidate information;

[0030] The information whose corresponding feature vector matches the feature vector of the key element is retrieved as candidate information.

[0031] Optionally, also include:

[0032] Obtaining a plurality of historical question-answer pairs, wherein the question-answer pairs include historical question information and historical answer information corresponding to the historical question information;

[0033] A plurality of extended question answer pairs are generated based on the plurality of historical question answer pairs to obtain a knowledge base consisting of the historical question answer pairs and the extended question answer pairs, wherein the extended question answer pairs include extended question information generated based on the plurality of historical question answer pairs and corresponding extended answer information.

[0034] A second aspect of the present application provides an information processing device, including:

[0035] An obtaining unit, used for obtaining first question information;

[0036] A determining unit, configured to determine second question information similar to the first question information;

[0037] A retrieval unit, configured to filter the target information corresponding to the first question information from the plurality of candidate information retrieved based on the first question information according to the answer feature of the second answer information corresponding to the second question information;

[0038] A generating unit is used to generate first answer information corresponding to the first question information according to the target information.

[0039] A third aspect of the present application provides an information processing system, comprising:

[0040] Processor and memory;

[0041] The memory is used to store computer programs;

[0042] The processor is used to execute the computer program to implement the information processing method provided in any one of the first aspects of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0044] Figure 1 is a flow chart of an information processing method provided by an embodiment of the present application;

[0045] Figure 2 This is a schematic diagram of a model framework provided by an embodiment of the present application;

[0046] Figure 3 is a structural diagram of an information processing device provided in an embodiment of the present application;

[0047] Figure 4 is a schematic diagram of the structure of another information processing device provided in an embodiment of the present application;

[0048] Figure 5 It is a structural diagram of an information processing system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] This application embodiment provides an information processing method, see Figure 1 , is a flow chart of the method, and the method may include the following steps.

[0051] S101, obtaining first question information.

[0052] The first question information can be input by any user.

[0053] The first question information may include any one or more of text information, image information and audio information.

[0054] S102: Determine second question information similar to the first question information.

[0055] The system executing the information processing method may determine similar second question information in a variety of ways.

[0056] An optional approach is to use a feature extraction model to process the first question information to obtain a semantic feature vector that is semantically related to the first question information, use the feature extraction model to process multiple candidate question information to obtain a semantic feature vector for each candidate question information, calculate the similarity between the semantic feature vector of the first question information and the semantic feature vector of each candidate question information, and determine the candidate question information whose corresponding similarity meets certain similarity conditions as the second question information.

[0057] The similarity condition may be that the corresponding similarity is greater than a preset similarity threshold, or that the corresponding similarities of multiple candidate question information are sorted from large to small, and the corresponding similarities are ranked in the top N positions, where N is a preset positive integer.

[0058] The candidate question information may include a plurality of question information stored in a pre-built knowledge base.

[0059] S103, according to the answer feature of the second answer information corresponding to the second question information, select the target information corresponding to the first question information from the multiple candidate information retrieved based on the first question information.

[0060] The second answer information may be answer information generated by the system and used to answer the second question information. Each question information in the knowledge base corresponds to an answer information. After determining similar second question information, the system may directly read the second answer information corresponding to the second question information from the knowledge base.

[0061] The system can determine whether each retrieved alternative information contains the answer feature of the second answer information. If an alternative information contains the answer feature of the second answer information, it is determined that the alternative information belongs to the target information. If an alternative information does not contain the answer feature of the second answer information, it is determined that the alternative information does not belong to the target information.

[0062] The candidate information may be information stored in the information base and related to the first question information, including but not limited to text, image and audio information related to the first question information. For example, the first question information may be "What is the core selling point of a certain product A?", and the retrieved candidate information may include multiple texts and images related to product A in the information base.

[0063] There may be multiple second question information similar to the first question information, and there may also be multiple corresponding second answer information. When there are multiple second answer information, if the candidate information contains the answer feature of at least one second answer information, it can be determined that the candidate information belongs to the target information; if the candidate information does not contain the answer feature of any second answer information, it can be determined that the candidate information does not belong to the target information.

[0064] S104: Generate first answer information corresponding to the first question information according to the target information.

[0065] After obtaining the target information, the system can input the first question information and the target information into a pre-built neural network model for generating answers, such as a large language model, and the model processes the first question information and the target information to obtain the corresponding first answer information.

[0066] The beneficial effects of this embodiment are:

[0067] When generating the first answer information, not only is the search performed based on the first question information, but the retrieved information is also screened based on the answers to the second question information that are similar to the first question information. Compared with a solution that only focuses on the first question information, generating the first answer information by combining the first question information and the answers to similar questions can improve the accuracy of the obtained answer and make the output answer more in line with user needs.

[0068] In some embodiments, the system may also determine the second question information similar to the first question information as follows:

[0069] Acquire first grammatical information of the first question information, where the first grammatical information indicates a structural relationship of sub-contents contained in the first question information in the first question information; the sub-contents are phrases, words or characters;

[0070] According to the similarity between the first grammatical information and the grammatical information corresponding to each question information in the knowledge base, the second question information similar to the first question information is searched and obtained.

[0071] See also Figure 2 In this embodiment, the system can use a recognizer to process the first question information to obtain the first grammatical information of the first question information. The recognizer can be a model with grammatical recognition capability, such as a large language model with grammatical recognition capability.

[0072] When the first question information is audio information or image information, the audio information or image information can be first converted into text information, and then the first grammatical information of the text information can be obtained by using a recognizer. When the first question information is text information, the first question information can be directly processed by a recognizer to obtain the corresponding first grammatical information.

[0073] As an example, the first question information may include “What is the algorithm task of license plate recognition?” For the first question information, the following first grammatical information may be obtained after being processed by the recognizer.

[0074] First grammatical information: {"license plate recognition": "noun phrase, indicating a task or field"; "of": "particle"; "task": "noun, indicating the goal or work of license plate recognition"; "what is": "question phrase, leading to the interrogative nature of the entire sentence"}.

[0075] Among them, "license plate recognition", "of", "task" and "what is" belong to the sub-content contained in the first question information. For example, "license plate recognition" is a phrase contained in the first question information, "of" is a character contained in the first question information, and "task" is a word contained in the first question information.

[0076] The information behind each sub-content in the first grammatical information is equivalent to the structural relationship of the sub-content, and these structural relationships reflect the part of speech and grammatical function of the corresponding sub-content in the first question information. For example, the structural relationship of the sub-content "license plate recognition" is "noun phrase, indicating a task or field", which indicates that the part of speech of the phrase "license plate recognition" in the first question information is a noun, and its grammatical function in the first question information is to indicate a specific task or field.

[0077] Each candidate question information in the knowledge base can also be processed by the identifier to obtain the grammatical information corresponding to these question information. For example, candidate question information 1 in the knowledge base can be processed by the identifier to obtain grammatical information 1, and candidate question information 2 can be processed by the identifier to obtain grammatical information 2.

[0078] Thus, the system can determine the similarity between the first grammatical information and the grammatical information corresponding to the candidate question information, and determine the second question information similar to the first question information according to the similarity between the grammatical information.

[0079] For example, the first grammatical feature vector corresponding to the first grammatical information can be extracted, the grammatical feature vector of the grammatical information corresponding to the candidate question information can be extracted, the similarity between the first grammatical feature vector and the grammatical feature vectors of the grammatical information corresponding to each question in the knowledge base can be calculated, and the question information whose similarity meets the similarity condition can be determined as the second question information.

[0080] When constructing the knowledge base, the system can use an identifier to process each question information contained in the knowledge base in advance, obtain the corresponding grammatical information, and store the question information and grammatical information accordingly in the knowledge base. In this way, when determining the second question information, the system can directly read the grammatical information already stored in the knowledge base, thereby improving the processing efficiency of the method of this embodiment.

[0081] The advantages of determining the second question information in the above manner are:

[0082] By evaluating the similarity between question information in combination with the grammatical information of the question, the determined second question information can not only have similar content to the first question information, but also have a similar structural relationship. Compared with the method of determining the second question information only based on the content of the first question information, the determination method of this embodiment can obviously obtain more similar second question information, thereby helping to improve the accuracy of the final output answer.

[0083] In some embodiments, the answer feature of the second answer information may indicate a structural relationship among the sub-contents contained in the second answer information.

[0084] As an example, assume that the second question information determined by the system is "What is the core algorithm task of recognizing invoice text?".

[0085] The grammatical information corresponding to the second question information can be: {“Recognize invoice text”: “noun phrase, indicating a task or field”; “of”: “auxiliary word”; “core algorithm task”: “noun phrase, indicating the problem to be solved”; “what is”: “interrogative phrase, guiding the interrogative nature of the entire sentence”}.

[0086] The second answer information corresponding to the second question information may be “OCR image recognition”, and the answer feature of the second answer information may be the structural relationship of the sub-content of “image recognition” in the second answer information. The answer feature may be expressed as {“specific task”: “image recognition”}.

[0087] Among them, "image recognition" is a phrase included in the second question information, and "specific task" is equivalent to the structural relationship of the sub-content. The structural relationship is used to explain that the role of the "image recognition" phrase in the second question information is to represent a specific task.

[0088] The system can utilize the recognizer to process the second answer information, obtain grammatical information of the second answer information, and obtain answer features of the second answer information from the grammatical information of the second answer information.

[0089] The method of obtaining the answer feature of the second answer information from the grammatical information may be:

[0090] The grammatical information of the second answer information is compared with the grammatical information of the second question information to determine the key sub-content of the second answer information, and the key sub-content and its corresponding structural relationship are determined as the answer feature of the second answer information. The key sub-content may be a sub-content in the second answer information that has the same or similar structural relationship with the sub-content of the second question information.

[0091] Combined with the above example, the grammatical information of the second answer information "OCR image recognition" can be {"OCR": "an algorithm for image recognition tasks"; "image recognition": "specific tasks"}. After comparing the grammatical information of the second answer information with the grammatical information of the second question information, it is determined that the sub-content "image recognition" of the second answer information and the "recognizing invoice text" of the second question information have a similar structural relationship, so it is determined that "image recognition" belongs to the key sub-content of the second answer information, and then the answer feature of the second answer information is determined to be {"specific task": "image recognition"}.

[0092] Based on the above answer features, the system can input the answer features and alternative information into Figure 2In the filter shown, the candidate information is filtered by the filter in the following manner to obtain the target information:

[0093] Determining whether the structural relationship of the sub-content contained in the candidate information retrieved based on the first question information is consistent with the structural relationship indicated by the answer feature;

[0094] The candidate information whose structural relationship of the sub-contents contained therein and the structural relationship indicated by the answer feature meet the consistency condition is determined as the target information corresponding to the first question information.

[0095] The filter can be a large language model or other neural network model with corresponding processing power.

[0096] For each alternative information, the filter can process the alternative information, obtain the grammatical information corresponding to the alternative information, and then compare the structural relationship indicated by the answer feature with the structural relationship of the sub-contents in the grammatical information. If the structural relationship of one or more sub-contents in the alternative information and the structural relationship indicated by the answer feature meet the consistency condition, it is determined that the alternative information belongs to the target information. If the structural relationship of each sub-content in the alternative information and the structural relationship indicated by the answer feature do not meet the consistency condition, it is determined that the alternative information does not belong to the target information.

[0097] The above consistency condition can be that the structural relationship of the sub-content contained in the alternative information is the same as the structural relationship indicated by the answer feature; or it can be that the similarity between the structural relationship of the sub-content contained in the alternative information and the structural relationship indicated by the answer feature is greater than a certain threshold.

[0098] When there are multiple answer features, if the structural relationship of the sub-content contained in an alternative information and the structural relationship indicated by at least one answer feature meet the consistency condition, it can be determined that the alternative information belongs to the target information. If the structural relationship of the sub-content contained in an alternative information and the structural relationship indicated by each answer feature do not meet the consistency condition, it can be determined that the alternative information does not belong to the target information.

[0099] In combination with the above example, when the answer feature is {"specific task": "image recognition"}, if the structural relationship of a sub-content in the alternative information includes "specific task" or "indicates a task or field", it can be determined that the alternative information meets the consistency condition and belongs to the target information. If the structural relationship of the sub-content in the alternative information does not include the above-mentioned "specific task" and "indicates a task or field", it can be determined that the alternative information does not meet the consistency condition and does not belong to the target information.

[0100] The benefits of filtering target information in the manner of this embodiment are:

[0101] Through the above-mentioned screening method, the target information can be screened in combination with the content and grammar of the second answer information, so as to screen out the target information whose content and grammar are similar to the second answer information. This helps to screen out the target information that has a strong effect on answering the second question information and the first question information that is similar to the second question information, thereby obtaining the first answer information that better meets the user's needs.

[0102] In some embodiments, see Figure 2 , the system can retrieve the alternative information as follows:

[0103] Extracting multiple key phrases with target grammatical relations from the first question information, where the target grammatical relations represent structural relations between multiple phrases in the key phrases;

[0104] Multiple candidate information is retrieved based on the key phrase.

[0105] The target grammatical relation may include one or more pre-set grammatical relations. The target grammatical relation may be specified by the user, or may be determined by the system after analyzing the question information and answer information stored in the knowledge base.

[0106] For example, the system can compare the grammatical information of the question information in the knowledge base and the answer features of the answer information. After comparison, it is found that in most question information, the phrases with consistent or similar structural relationships referred to by the structural relationships and answer features are all phrases with attributive relationship or subject-predicate relationship. Then the system can determine that the target grammatical relationship is the attributive relationship and the subject-predicate relationship.

[0107] The system may process the first question information with an identifier, obtain the structural relationship of multiple phrases contained in the first question information, and then determine the phrases having the structural relationship represented by the target grammatical relationship as key phrases.

[0108] In combination with the above example, assuming that the target grammatical relationship includes an attributive relationship and a subject-predicate relationship, and the first question information is "What is the algorithmic task of license plate recognition?", the system can determine the phrases with an attributive relationship and the phrases with a subject-predicate relationship in the first question information as key phrases. For example, the phrases "recognition algorithm" and "algorithm task" with an attributive relationship, and the phrase "license plate recognition" with a subject-predicate relationship are determined as key phrases. That is, the key phrases of the first question information can include "recognition algorithm", "algorithm task" and "license plate recognition".

[0109] After obtaining the key phrase, the system can directly search based on the key phrase to obtain candidate information corresponding to the first question information, for example, retrieve information containing the key phrase as candidate information, or retrieve information with a similarity to the key phrase greater than a certain threshold as candidate information.

[0110] In some optional embodiments, see Figure 2 After the system obtains the key phrase, it can also determine the key elements of the first question information according to the key phrase as follows:

[0111] Obtaining at least one keyword of the first question information determined according to semantics;

[0112] Identify key elements based on key phrases and keywords.

[0113] The method of determining keywords can be found in the related art and will not be described in detail in this embodiment.

[0114] As an example, assuming that the first question information is “What is the algorithm task of license plate recognition?”, the keywords determined from the first question information may include “license plate”, “recognition”, “algorithm” and “task”.

[0115] After obtaining the keywords and key phrases, the system can determine the key elements as follows:

[0116] Determine the contribution of the key phrase and the keyword according to the first question information, wherein the contribution represents the strength of the key phrase and the keyword in processing the first question information;

[0117] The key elements are filtered out from the key phrases and keywords according to their contribution.

[0118] The system can input the first question information, keywords and key phrases into a pre-built neural network model (such as a large language model, Large Language Model, LLM) together, and the model determines the contribution of each keyword and each key phrase.

[0119] For each keyword or key phrase, the greater the corresponding contribution, the stronger the role of the keyword or key phrase in answering the first question information, and the smaller the corresponding contribution, the weaker the role of the keyword or key phrase in answering the first question information.

[0120] After obtaining the contribution of key phrases and keywords, they can be sorted from high to low according to the contribution, and then the first M key phrases or keywords are selected as the key elements of the first question information. M is a preset positive integer.

[0121] In combination with the above example, assuming that the contribution of the above-mentioned key phrases "recognition algorithm", "algorithm task" and "license plate recognition" of the first question information, as well as the contribution of the keywords "license plate", "recognition", "algorithm" and "task" are determined, the top four words or phrases with the largest contribution are "license plate recognition", "algorithm task", "license plate" and "recognition" respectively, then it can be determined that the key elements of the first question information include "license plate recognition", "algorithm task", "license plate" and "recognition".

[0122] After obtaining the key elements, the system can retrieve multiple alternative information based on the key elements.

[0123] The beneficial effects of this embodiment are:

[0124] On the one hand, by combining the semantics represented by the keywords and the grammatical relationship represented by the key phrases to retrieve alternative information, it is possible to retrieve alternative information that is semantically similar to the first question information, as well as retrieve alternative information that is grammatically similar to the first question information. Compared with searching based only on semantics, combining keywords and key phrases can obtain more comprehensive alternative information, thereby helping to obtain more comprehensive first answer information.

[0125] On the other hand, by combining the contribution of keywords and key phrases to determine the key elements, it is possible to control the correlation between the retrieved alternative information and the first question information, avoid retrieving a large amount of redundant information due to too many elements involved in the retrieval, and thus improve the retrieval efficiency.

[0126] After obtaining the key elements, the system can retrieve multiple candidate information in at least one of the following retrieval methods:

[0127] Retrieval method 1: retrieve information containing key elements as candidate information.

[0128] In the first retrieval method, the system can identify whether each piece of information in the information database contains key elements. If a piece of information contains one or more key elements, it can be determined that the information belongs to alternative information. If a piece of information does not contain any key elements, it can be determined that the information does not belong to alternative information.

[0129] The second retrieval method is to retrieve the information whose corresponding feature vector matches the feature vector of the key element as the candidate information.

[0130] In the second retrieval method, the system can use the feature extraction model to extract the semantic feature vectors of the key elements, extract the semantic feature vectors of each piece of information in the information database, and then calculate the similarity between the two, and determine the information whose corresponding similarity meets the similarity conditions as candidate information.

[0131] Optionally, the system can also obtain alternative information in the following retrieval method three:

[0132] Directly obtain the semantic feature vector of the first question information and the semantic feature vector of each piece of information in the information library, then determine the similarity between the semantic feature vector of the first question information and the semantic feature vector of each piece of information in the information library, and determine the information whose corresponding similarity meets the similarity condition as candidate information.

[0133] Optionally, the system may also combine the above-mentioned multiple search methods, for example, determine the information retrieved by search method one, search method two and search method three as candidate information.

[0134] In this embodiment, the system can construct a knowledge base in the following manner:

[0135] Obtaining a plurality of historical question-answer pairs, the question-answer pairs comprising historical question information and historical answer information corresponding to the historical question information;

[0136] Multiple extended question answer pairs are generated based on multiple historical question answer pairs to obtain a knowledge base consisting of historical question answer pairs and extended question answer pairs. The extended question answer pairs include extended question information generated based on multiple historical question answer pairs and corresponding extended answer information.

[0137] Historical question information may include each question information input into the system before obtaining the first question information. Historical answer information may include answer information generated by the system based on the historical question information, and may also include answer information manually edited by the user and input together with the historical question information.

[0138] The extended question-answer pairs can be generated in the following ways:

[0139] The system selects a historical question-answer pair, expands the historical question information therein to obtain an extended question information, expands the historical answer information therein to obtain an extended answer information, and determines the extended question information and the extended answer information as an extended question-answer pair.

[0140] Alternatively, the extended question-answer pair can be generated as follows:

[0141] The system selects a historical question-answer pair, expands the historical question information therein to obtain an extended question information, and then generates extended answer information for replying to the extended question information based on the extended question information, and determines the extended question information and the extended answer information as an extended question-answer pair.

[0142] The system can use a large language model or other models that can expand question information and answer information to generate extended question-answer pairs.

[0143] Optionally, the system may verify the generated extended question-answer pairs, and delete extended question-answer pairs containing grammatical errors and duplicate extended question-answer pairs.

[0144] The present application also provides an information processing device. Figure 3 , the device may include the following units.

[0145] An obtaining unit 301 is used to obtain first question information;

[0146] A determining unit 302, configured to determine second question information similar to the first question information;

[0147] A retrieval unit 303 is used to filter out target information corresponding to the first question information from a plurality of candidate information retrieved based on the first question information according to the answer feature of the second answer information corresponding to the second question information;

[0148] The generating unit 304 is configured to generate first answer information corresponding to the first question information according to the target information.

[0149] Optionally, the second question information and the second answer information are stored in a knowledge base, and when the determining unit 302 determines the second question information similar to the first question information, it can be used to:

[0150] Acquire first grammatical information of the first question information, where the first grammatical information indicates a structural relationship of sub-contents contained in the first question information in the first question information; the sub-contents are phrases, words or characters;

[0151] According to the similarity between the first grammatical information and the grammatical information corresponding to each question information in the knowledge base, the second question information similar to the first question information is searched and obtained.

[0152] Optionally, the answer feature of the second answer information indicates the structural relationship of the sub-contents contained in the second answer information in the second answer information;

[0153] When the retrieval unit 303 obtains the target information corresponding to the first question information from the multiple candidate information obtained by searching the first question information according to the answer feature, it can be used to:

[0154] Determining whether the structural relationship of the sub-content contained in the candidate information retrieved based on the first question information is consistent with the structural relationship indicated by the answer feature;

[0155] The candidate information whose structural relationship of the sub-contents contained therein and the structural relationship indicated by the answer feature meet the consistency condition is determined as the target information corresponding to the first question information.

[0156] Optionally, when the retrieval unit 303 retrieves multiple candidate information based on the first question information, it can be used to:

[0157] Extracting multiple key phrases with target grammatical relations from the first question information, where the target grammatical relations represent structural relations between multiple phrases in the key phrases;

[0158] Multiple candidate information is retrieved based on the key phrase.

[0159] Optionally, the retrieval unit 303 is further configured to:

[0160] Obtaining at least one keyword of the first question information determined according to semantics;

[0161] Identify key elements based on key phrases and keywords;

[0162] When the retrieval unit 303 retrieves multiple candidate information according to the key phrase, it can be used to:

[0163] Multiple alternative information is retrieved based on key factors.

[0164] Optionally, when the retrieval unit 303 determines the key elements according to the key phrases and keywords, it can be used to:

[0165] Determine the contribution of the key phrase and the keyword according to the first question information, wherein the contribution represents the strength of the key phrase and the keyword in processing the first question information;

[0166] The key elements are filtered out from the key phrases and keywords according to their contribution.

[0167] Optionally, when the retrieval unit 303 retrieves multiple candidate information according to the key elements, at least one of the following can be performed:

[0168] Retrieve information containing key elements as candidate information;

[0169] The information whose corresponding feature vector matches the feature vector of the key element is retrieved as candidate information.

[0170] Optional, see Figure 4 The device further comprises a construction unit 401, which is used to construct a knowledge base in the following manner:

[0171] Obtaining a plurality of historical question-answer pairs, the question-answer pairs comprising historical question information and historical answer information corresponding to the historical question information;

[0172] Multiple extended question answer pairs are generated based on multiple historical question answer pairs to obtain a knowledge base consisting of historical question answer pairs and extended question answer pairs. The extended question answer pairs include extended question information generated based on multiple historical question answer pairs and corresponding extended answer information.

[0173] The working principle of the information processing device provided in this embodiment can refer to the relevant steps of the information processing method provided in any embodiment of the present application, and will not be repeated here.

[0174] The present application also provides an information processing system. Figure 5 , the system may include:

[0175] Processor 501 and memory 502;

[0176] The memory 502 is used to store computer programs;

[0177] The processor 501 is used to execute a computer program to implement the information processing method provided in any embodiment of the present application.

[0178] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0179] For the convenience of description, the above system or device is described by dividing it into various modules or units according to its functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0180] It can be known from the description of the above implementation methods that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly contributed to the prior art in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0181] Finally, it should be noted that, in this article, relational terms such as first, second, third and fourth are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0182] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. An information processing method, comprising: Obtain the first question information; Determining second question information similar to the first question information; According to the answer feature of the second answer information corresponding to the second question information, the target information corresponding to the first question information is screened out from the plurality of candidate information retrieved based on the first question information; First answer information corresponding to the first question information is generated according to the target information.

2. The method according to claim 1, wherein the second question information and the second answer information are stored in a knowledge base, and the determining of the second question information similar to the first question information comprises: Acquire first grammatical information of the first question information, where the first grammatical information indicates a structural relationship of sub-contents contained in the first question information in the first question information; The sub-content is a phrase, a word or a character; According to the similarity between the first grammatical information and the grammatical information corresponding to each piece of question information in the knowledge base, second question information similar to the first question information is searched and obtained.

3. The method according to claim 1 or 2, wherein the answer feature of the second answer information indicates the structural relationship of the sub-contents contained in the second answer information in the second answer information; The step of selecting, according to the answer feature, a plurality of candidate information retrieved based on the first question information to obtain target information corresponding to the first question information includes: Determining whether a structural relationship of sub-contents contained in the candidate information retrieved based on the first question information in the candidate information is consistent with a structural relationship indicated by the answer feature; The candidate information whose structural relationship of the sub-contents contained therein and the structural relationship indicated by the answer feature satisfy the consistency condition is determined as the target information corresponding to the first question information.

4. The method according to claim 1, wherein the plurality of candidate information retrieved based on the first question information comprises: Extracting a plurality of key phrases having a target grammatical relationship from the first question information, wherein the target grammatical relationship represents a structural relationship between a plurality of phrases in the key phrases; A plurality of candidate information is retrieved according to the key phrase.

5. The method according to claim 4, further comprising: Obtaining at least one keyword of the first question information determined according to semantics; determining key elements based on the key phrases and the keywords; The retrieving a plurality of candidate information according to the key phrase includes: A plurality of candidate information is retrieved based on the key elements.

6. The method according to claim 5, wherein determining key elements according to the key phrases and the keywords comprises: Determining, according to the first question information, the contribution of the key phrase and the contribution of the keyword, wherein the contribution represents the strength of the key phrase and the keyword in processing the first question information; Key elements are obtained by screening the key phrases and the keywords according to the contribution.

7. The method according to claim 5, wherein the retrieving a plurality of candidate information according to the key elements comprises at least one of the following: Retrieving information containing the key elements as candidate information; The information whose corresponding feature vector matches the feature vector of the key element is retrieved as candidate information.

8. The method according to claim 1, further comprising: Obtaining a plurality of historical question-answer pairs, wherein the question-answer pairs include historical question information and historical answer information corresponding to the historical question information; A plurality of extended question answer pairs are generated based on the plurality of historical question answer pairs to obtain a knowledge base consisting of the historical question answer pairs and the extended question answer pairs, wherein the extended question answer pairs include extended question information generated based on the plurality of historical question answer pairs and corresponding extended answer information.

9. An information processing device, comprising: An obtaining unit, used for obtaining first question information; A determining unit, configured to determine second question information similar to the first question information; A retrieval unit, configured to filter the target information corresponding to the first question information from the plurality of candidate information retrieved based on the first question information according to the answer feature of the second answer information corresponding to the second question information; A generating unit is used to generate first answer information corresponding to the first question information according to the target information.

10. An information processing system comprising: Processor and memory; The memory is used to store computer programs; The processor is used to execute the computer program to implement the information processing method according to any one of claims 1 to 8.

Citation Information

Patent Citations

  • Method and device for intelligently prompting answers

    CN107436916A

  • Information search method and device, terminal equipment and computer readable storage medium

    CN114416944A

  • Intelligent question and answer method, equipment and device and storage medium

    CN115730047A

  • Intelligent question answering system

    CN116910216A

  • Answer screening method and device, computer equipment and storage medium

    CN118332008A