Information processing method and device, computer device and storage medium

Abstract

Embodiments of the present application disclose an information processing method and device, computer equipment and a storage medium, belonging to the technical field of computers. The method comprises: obtaining first classification features based on question information and candidate answer information; obtaining evidence features of evidence information corresponding to the question information, the evidence information being used to provide evidence that the candidate answer information is answer information of the question information; adjusting the evidence features based on the correlation between the first classification features and the evidence features to obtain second classification features; and performing classification based on the first classification features and the second classification features to obtain a classification result corresponding to the candidate answer information. The method can be applied in the field of intelligent transportation. In the process of determining the correctness of the candidate answer information, the evidence information is used as evidence to prove whether the candidate answer information can be the answer information of the question information, the influence of the evidence information is considered, and therefore a more accurate classification result is obtained, and the accuracy of the determined answer information is improved.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of computer, and particularly relate to an information processing method and device, computer equipment and storage medium. BACKGROUND

[0002] Natural language processing has great significance in academic research and practical application, and the most extensive practical application of natural language processing is various forms of question answering. The current question answering process usually searches for question information, extracts a segment from a document obtained by searching as candidate answer information that can answer the question information, and finally judges the correctness of the candidate answer information.

[0003] In related technologies, a matching degree between question information and candidate answer information is directly obtained, and the correctness of the candidate answer information is determined according to the matching degree, so as to determine the final answer information. However, this way only considers the association relationship between the question information and the candidate answer information, and the information relied on is less, so that the accuracy of the determined answer information is poor. SUMMARY

[0004] Embodiments of the present application provide an information processing method, device, computer equipment and storage medium, which improves the accuracy of the determined answer information. The technical solution is as follows:

[0005] On the one hand, an information processing method is provided, and the method comprises:

[0006] obtaining a first classification feature based on question information and candidate answer information, the first classification feature at least indicating semantic information of the candidate answer information;

[0007] obtaining evidence features of evidence information corresponding to the question information, the evidence information being used to provide evidence that the candidate answer information is answer information of the question information;

[0008] adjusting the evidence features based on the association degree between the first classification features and the evidence features to obtain second classification features;

[0009] performing classification based on the first classification features and the second classification features to obtain a classification result corresponding to the candidate answer information, the classification result indicating whether the candidate answer information is answer information of the question information.

[0010] Optionally, the evidence information comprises a plurality of evidence segments,

[0011] the feature extraction layer in the second extraction network is configured to obtain evidence segment features corresponding to the plurality of evidence segments, respectively;

[0012] The attention layer in the second extraction network is used to adjust the features of the multiple evidence fragments based on the evidence weights corresponding to the multiple evidence fragments, wherein the evidence weights indicate the importance of the evidence fragments.

[0013] The attention layer is also used to splice together the adjusted features of multiple evidence fragments to obtain the evidence features.

[0014] Optionally, the second extraction network in the second processing sub-model is also used to obtain problem features of the problem information;

[0015] The third attention network in the second processing sub-model is used to match the problem features and the evidence features to obtain the second confidence level corresponding to the evidence information. The second confidence level represents the credibility of the evidence information as evidence of the problem information.

[0016] The third attention network is also used to adjust the evidence features based on the second confidence level.

[0017] Optionally, the training process of the information processing model includes:

[0018] Obtain sample question information, sample candidate answer information, sample evidence information, and sample classification results corresponding to the sample candidate answer information;

[0019] The information processing model is invoked to process the sample question information, the sample candidate answer information, and the sample evidence information to obtain the predicted classification result corresponding to the sample candidate answer information;

[0020] The information processing model is trained based on the sample classification results and the predicted classification results.

[0021] On the other hand, an information processing apparatus is provided, the apparatus comprising:

[0022] The first feature acquisition module is used to acquire a first classification feature based on question information and candidate answer information, wherein the first classification feature at least represents the semantic information of the candidate answer information;

[0023] The second feature acquisition module is used to acquire the evidence features of the evidence information corresponding to the question information, and the evidence information is used to provide evidence that the candidate answer information is the answer information of the question information.

[0024] The first feature adjustment module is used to adjust the evidence feature based on the correlation between the first classification feature and the evidence feature to obtain the second classification feature;

[0025] The classification module is configured to classify based on the first classification feature and the second classification feature to obtain a classification result corresponding to the candidate answer information, the classification result indicating whether the candidate answer information is answer information of the question information.

[0026] Optionally, the apparatus further comprises:

[0027] The similarity obtaining module is configured to obtain an answer similarity between each two of the candidate answer information.

[0028] The second feature adjusting module is configured to, for each candidate answer information, obtain a third classification feature based on the plurality of answer similarities corresponding to the candidate answer information and the first classification features corresponding to other candidate answer information except the candidate answer information in the plurality of candidate answer information.

[0029] The classification module is configured to, for each candidate answer information, classify based on the first classification feature, the second classification feature and the third classification feature corresponding to the candidate answer information to obtain the classification result.

[0030] Optionally, the classification module is configured to:

[0031] fuse the first classification feature and the third classification feature to obtain a fused classification feature;

[0032] splice the first classification feature, the third classification feature and the fused classification feature to obtain a fourth classification feature of the candidate answer information;

[0033] classify based on the second classification feature and the fourth classification feature to obtain the classification result.

[0034] Optionally, the second feature adjusting module comprises:

[0035] The normalization unit is configured to normalize the plurality of answer similarities to obtain normalized plurality of answer similarities.

[0036] The adjusting unit is configured to, based on the normalized plurality of answer similarities, perform weighted processing on the first classification features corresponding to other candidate answer information except the candidate answer information in the plurality of candidate answer information to obtain the third classification feature.

[0037] Optionally, the classification module is configured to:

[0038] weight the second classification feature based on a first confidence degree corresponding to the evidence information to obtain a weighted second classification feature, the first confidence degree indicating a credibility of the evidence information.

[0039] perform classification based on the first classification feature and the weighted second classification feature to obtain the classification result.

[0040] Optionally, the evidence information includes a plurality of evidence segments, and the device further includes a confidence degree acquisition module configured to:

[0041] acquire a segment similarity between each two of the plurality of evidence segments;

[0042] for each evidence segment, determine a segment confidence degree corresponding to the evidence segment based on a plurality of segment similarities corresponding to the evidence segment;

[0043] splice the segment confidence degrees corresponding to the plurality of evidence segments to obtain the first confidence degree.

[0044] Optionally, the evidence information includes a plurality of evidence segments, and the second feature acquisition module is configured to:

[0045] acquire an evidence segment feature corresponding to each of the plurality of evidence segments respectively;

[0046] adjust the plurality of evidence segment features respectively based on evidence weights corresponding to the plurality of evidence segments, the evidence weights indicating importance degrees of the evidence segments;

[0047] splice the adjusted plurality of evidence segment features to obtain the evidence feature.

[0048] Optionally, the second feature acquisition module is configured to:

[0049] acquire a question feature of the question information;

[0050] match the question feature and the evidence feature to obtain a second confidence degree corresponding to the evidence information, the second confidence degree representing a credibility degree of the evidence information as evidence of the question information;

[0051] adjust the evidence feature based on the second confidence degree.

[0052] Optionally, the classification module is configured to:

[0053] fuse the first classification feature and the second classification feature to obtain a fused classification feature;

[0054] splice the first classification feature, the second classification feature and the fused classification feature to obtain a fifth classification feature of the candidate answer information;

[0055] perform classification based on the fifth classification feature to obtain the classification result.

[0056] Optionally, the first feature acquisition module is configured to:

[0057] combine the question information and the candidate answer information to obtain combined information;

[0058] extract a first classification feature from the combined information.

[0059] Optionally, the apparatus further comprises an answer acquisition module configured to:

[0060] search based on the question information and other question information having the same meaning as the question information;

[0061] extract the candidate answer information from the searched information.

[0062] Optionally, the information processing model comprises a first processing sub-model, a second processing sub-model, and a classification sub-model,

[0063] the first processing sub-model is configured to acquire the first classification feature based on the question information and the candidate answer information;

[0064] the second processing sub-model is configured to acquire an evidence feature of evidence information corresponding to the question information;

[0065] the second processing sub-model is further configured to adjust the evidence feature based on a correlation degree between the first classification feature and the evidence feature to obtain the second classification feature of the candidate answer information;

[0066] the classification sub-model is configured to classify based on the first classification feature and the second classification feature to obtain a classification result corresponding to the candidate answer information.

[0067] Optionally, a first attention network in the first processing sub-model is configured to acquire an answer similarity between each two of a plurality of candidate answer information;

[0068] the first attention network is further configured to, for each candidate answer information, acquire a third classification feature based on a plurality of answer similarities corresponding to the candidate answer information and first classification features corresponding to other candidate answer information except the candidate answer information in the plurality of candidate answer information;

[0069] the classification sub-model is further configured to, for each candidate answer information, classify based on the first classification feature, the second classification feature, and the third classification feature corresponding to the candidate answer information to obtain the classification result.

[0070] Optionally, the fusion network in the classification sub-model is configured to perform weighted processing on the second classification feature according to a first confidence degree corresponding to the evidence information, to obtain a weighted second classification feature, where the first confidence degree refers to a degree of credibility of the evidence information.

[0071] The classification network in the classification sub-model is configured to perform classification based on the first classification feature and the weighted second classification feature, to obtain the classification result.

[0072] Optionally, the evidence information includes a plurality of evidence segments, and the self-attention network in the second processing sub-model is configured to:

[0073] obtain a segment similarity between each two of the plurality of evidence segments;

[0074] for each evidence segment, determine a segment confidence degree corresponding to the evidence segment based on a plurality of segment similarities corresponding to the evidence segment;

[0075] splice the segment confidence degrees corresponding to the plurality of evidence segments, to obtain the first confidence degree.

[0076] Optionally, the first extraction network in the first processing sub-model is configured to obtain the first classification feature based on the question information and the candidate answer information;

[0077] The second extraction network in the second processing sub-model is configured to obtain an evidence feature of the evidence information corresponding to the question information;

[0078] The second attention network in the second processing sub-model is configured to adjust the evidence feature based on an association degree between the first classification feature and the evidence feature, to obtain the second classification feature of the candidate answer information.

[0079] Optionally, the evidence information includes a plurality of evidence segments,

[0080] The feature extraction layer in the second extraction network is configured to obtain an evidence segment feature corresponding to each of the plurality of evidence segments;

[0081] The attention layer in the second extraction network is configured to adjust a plurality of evidence segment features based on evidence weights corresponding to the plurality of evidence segments, where the evidence weights indicate importance degrees of the evidence segments;

[0082] The attention layer is further configured to splice the adjusted plurality of evidence segment features, to obtain the evidence feature.

[0083] Optionally, the second extraction network in the second processing sub-model is further configured to obtain a question feature of the question information.

[0084] a third attention network in the second processing sub-model, configured to match the question feature and the evidence feature to obtain a second confidence corresponding to the evidence information, the second confidence representing a degree of credibility of the evidence information as evidence of the question information;

[0085] the third attention network is further configured to adjust the evidence feature based on the second confidence.

[0086] Optionally, the training process of the information processing model comprises:

[0087] obtaining sample question information, sample candidate answer information, sample evidence information, and a sample classification result corresponding to the sample candidate answer information;

[0088] calling the information processing model to process the sample question information, the sample candidate answer information, and the sample evidence information to obtain a predicted classification result corresponding to the sample candidate answer information;

[0089] training the information processing model based on the sample classification result and the predicted classification result.

[0090] In another aspect, a computer device is provided, comprising a processor and a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to implement the operations performed by the information processing method according to the above aspect.

[0091] In another aspect, a computer readable storage medium is provided, the computer readable storage medium storing at least one computer program, the at least one computer program being loaded and executed by a processor to implement the operations performed by the information processing method according to the above aspect.

[0092] In another aspect, a computer program product is provided, comprising a computer program, the computer program being executed by a processor to implement the operations performed by the information processing method according to the above aspect.

[0093] The technical scheme provided by the embodiments of the present application has at least the following beneficial effects:

[0094] The technical scheme provided in the embodiments of the present application introduces the evidence information corresponding to the question information in the process of determining the correctness of the candidate answer information, takes the evidence information as the evidence proving whether the candidate answer information can be the answer information of the question information, and thus adjusts the second classification feature obtained by the evidence feature based on the correlation degree between the first classification feature corresponding to the candidate answer information and the evidence feature, so that the second classification feature can not only reflect the semantic information of the evidence information itself, but also reflect the support degree of the evidence information to the candidate answer information, and thus when classification is performed based on the first classification feature and the second classification feature, the influence of the evidence information can be considered to obtain a more accurate classification result, and the accuracy of the determined answer information is improved. BRIEF DESCRIPTION OF DRAWINGS

[0095] In order to more clearly illustrate the technical schemes in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0096] Figure 1 is a schematic diagram of an implementation environment provided by the embodiments of the present application;

[0097] Figure 2 is a flowchart of an information processing method provided by the embodiments of the present application;

[0098] Figure 3 is a flowchart of an information processing method provided by the embodiments of the present application;

[0099] Figure 4 is a flowchart of an information processing method provided by the embodiments of the present application;

[0100] Figure 5 is a flowchart of an information processing method provided by the embodiments of the present application;

[0101] Figure 6 is a flowchart of an information processing method provided by the embodiments of the present application;

[0102] Figure 7 is a schematic diagram of an information processing model provided by the embodiments of the present application;

[0103] Figure 8 is a schematic diagram of a first processing sub-model provided by the embodiments of the present application;

[0104] Figure 9 is a schematic diagram of a second processing sub-model provided by the embodiments of the present application;

[0105] Figure 10is a schematic diagram of a classification sub-model provided by an embodiment of the present application;

[0106] Figure 11 is a flowchart of an information processing method provided by an embodiment of the present application;

[0107] Figure 12 is a flowchart of an information processing method provided by an embodiment of the present application;

[0108] Figure 13 is a schematic diagram of an information processing method provided by an embodiment of the present application;

[0109] Figure 14 is a flowchart of an information processing method provided by an embodiment of the present application;

[0110] Figure 15 is a schematic diagram of an information processing method provided by an embodiment of the present application;

[0111] Figure 16 is a structural schematic diagram of an information processing apparatus provided by an embodiment of the present application;

[0112] Figure 17 is a structural schematic diagram of a terminal provided by an embodiment of the present application;

[0113] Figure 18 is a structural schematic diagram of a server provided by an embodiment of the present application. DETAILED DESCRIPTION

[0114] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the drawings.

[0115] It can be understood that the terms "first", "second" and the like used in the present application can be used in the present document to describe various concepts, but unless specifically stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the present application, the first arrangement order can be referred to as the second arrangement order, and the second arrangement order can be referred to as the first arrangement order.

[0116] The terms "at least one", "multiple", "each", "any" and the like used in the present application include one, two or more than two, multiple includes two or more than two, each refers to each of the corresponding multiple, and any refers to any one of the multiple. For example, multiple classification features include 3 classification features, and each classification feature refers to each of the 3 classification features, and any refers to any one of the 3 classification features, which can be the first, the second or the third.

[0117] To facilitate understanding of the embodiments of the present application, the key words involved in the embodiments of the present application are explained first:

[0118] Question Answering (QA): For the question information given by the user, return concise, accurate and matching answer information to the question information through retrieval of corpus, knowledge graph or question answering knowledge base. The question answering system can be divided into structured data-based question answering system, text-based question answering system and question pair-based question answering system, for example, the structured data-based question answering system is KBQA (Knowledge Base Question Answering), the text-based question answering system is MRC (Machine Reading Comprehension), and the question pair-based question answering system is FAQ (Frequently Asked Questions) question answering system.

[0119] Reading comprehension: A technology that uses algorithms to make computer devices understand text semantics and answer related questions. Common question types include fill-in-the-blank, multiple-choice, and essay questions. In the text-based question answering system, the fragment extraction type of reading comprehension technology is usually used to extract continuous fragments from the problem-related documents as candidate answers.

[0120] Artificial Intelligence (AI) is to use digital computers or digital computer controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which tries to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0121] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, automatic driving, intelligent transportation and several other major directions.

[0122] Natural language processing (NLP) is an important direction in the field of computer science and artificial intelligence. It studies various theories and methods that can realize effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science and mathematics. Therefore, the research in this field will involve natural language, i.e. the language used in daily life, so it is closely related to the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question and answer, knowledge graph and other technologies.

[0123] Machine learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It is a specialized study of how computers simulate or implement human learning behavior to acquire new knowledge or skills, reorganize existing knowledge structure to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent. Its applications are widespread in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and teaching learning.

[0124] With the research and progress of artificial intelligence technology, artificial intelligence technology has been researched and applied in many fields, such as common smart home, smart wearable device, virtual assistant, smart speaker, smart marketing, unmanned vehicle, autonomous driving, unmanned aerial vehicle, robot, smart medical treatment, smart customer service, Internet of vehicles, autonomous driving, intelligent transportation, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0125] The scheme provided by the embodiments of the present application relates to natural language processing, machine learning and other technologies of artificial intelligence. By calling an information processing model, the question information, candidate answer information and evidence information are processed to obtain a classification result corresponding to the candidate answer information, so as to determine whether the candidate answer information can be used as the answer information of the question information.

[0126] The information processing method provided in the embodiments of the present application is executed by a computer device, which is a terminal, such as a mobile phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, a vehicle-mounted terminal, or another type of terminal, or the computer device is a server. Optionally, the server is a stand-alone physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and basic cloud computing services such as big data and artificial intelligence platforms.

[0127] Alternatively, the information processing method provided in the embodiments of the present application is executed by interaction between a terminal and a server. Figure 1 FIG. 1 is a schematic diagram of an implementation environment provided by the embodiments of the present application. Referring to FIG. 1, Figure 1 The implementation environment includes a terminal 101 and a server 102. The terminal 101 and the server 102 are connected through a wireless or wired network. The terminal 101 is installed with a target application provided by the server 102, and the terminal 101 can realize functions such as information search and data transmission through the target application. Optionally, the target application is a target application in the operating system of the terminal 101, or a target application provided by a third party. For example, the target application is a search application configured with a question and answer system, which has a question search function, and of course, the search application can also have other functions, such as a download function, a sharing function, and the like.

[0128] A user can input question information through the terminal 101, and send the question information input by the user to the server 102 through the target application. After receiving the question information, the server 102 obtains corresponding candidate answer information and evidence information based on the question information, and determines whether the candidate answer information can be used as the answer information of the question information based on the evidence information and the question information. Optionally, the server 102 can return the candidate answer information of the candidate question information to the terminal 101, and the terminal 101 displays the candidate answer information; or the server 102 directly returns the finally determined answer information to the terminal 101, and the terminal 101 displays the answer information.

[0129] Figure 2 FIG. 2 is a flowchart of an information processing method provided by the embodiments of the present application. The execution subject of the embodiments of the present application is a computer device. Referring to FIG. 2, Figure 2 The method includes the following steps:

[0130] 201. The computer device obtains first classification features based on the question information and the candidate answer information.

[0131] The computer device obtains question information and candidate answer information corresponding to the question information. In order to facilitate subsequent determination of whether the candidate answer information can be used as answer information of the question information, feature extraction is performed based on the question information and the candidate answer information to obtain first classification features, which at least represent semantic information of the candidate answer information.

[0132] In addition, since the first classification features are obtained based on the question information and the candidate answer information, in the case of considering the correlation between the question information and the candidate answer information when obtaining the first classification features, the first classification features can also represent the correlation between the question information and the candidate answer information.

[0133] 202. The computer device obtains evidence features of evidence information corresponding to the question information.

[0134] The evidence information is used to provide evidence that the candidate answer information is answer information of the question information. The computer device performs feature extraction on the evidence information to obtain the evidence features, which represent semantic information of the evidence information.

[0135] 203. The computer device adjusts the evidence features based on a correlation degree between the first classification features and the evidence features to obtain second classification features.

[0136] The correlation degree represents a correlation between the first classification features and the evidence features. The greater the correlation degree, the more similar the candidate answer information and the evidence information, which means that the possibility of the candidate answer information being answer information of the question information is greater. The smaller the correlation degree, the greater the difference between the candidate answer information and the evidence information, which means that the possibility of the candidate answer information being answer information of the question information is smaller. Then, the evidence features are adjusted based on the correlation degree to obtain the second classification features, which represent a support degree of the evidence information for the candidate answer information.

[0137] 204. The computer device performs classification based on the first classification features and the second classification features to obtain a classification result corresponding to the candidate answer information, which indicates whether the candidate answer information is answer information of the question information.

[0138] Since the first classification features and the second classification features can reflect the correlation between the candidate answer information and the question information and the correlation degree between the candidate answer information and the evidence information, classification based on the first classification features and the second classification features can obtain the classification result corresponding to the candidate answer information, thereby determining whether the candidate answer information is answer information of the question information.

[0139] The method provided in the embodiments of the present application introduces the evidence information corresponding to the question information in the process of determining the correctness of the candidate answer information, takes the evidence information as the evidence proving whether the candidate answer information can be the answer information of the question information, and thus adjusts the second classification feature obtained by the evidence feature based on the correlation degree between the first classification feature corresponding to the candidate answer information and the evidence feature, so that the second classification feature can not only reflect the semantic information of the evidence information itself, but also reflect the support degree of the evidence information to the candidate answer information, and thus when classification is performed based on the first classification feature and the second classification feature, the influence of the evidence information can be considered to obtain a more accurate classification result, and the accuracy of the determined answer information is improved.

[0140] The embodiments of the above Figure 2 only simply introduce the process of using the evidence information as the evidence to determine whether the candidate answer information can be the answer information of the question information in the process of determining the answer information. The information processing process is described in detail through the embodiments shown in the following Figure 3 .

[0141] Figure 3 is a flowchart of an information processing method provided by the embodiments of the present application. The execution subject of the embodiments of the present application is a computer device. Referring to Figure 3 , the method comprises the following steps:

[0142] 301. The computer device obtains candidate answer information and evidence information corresponding to question information.

[0143] The question information is the information input by a user, and the computer device searches for the corresponding candidate answer information and evidence information based on the question information after obtaining the question information input by the user. The question information is in the form of text, voice or other forms. Optionally, the user inputs question text through a keyboard; or the user inputs question voice, and the computer device can directly search based on the question voice or can convert the question voice into corresponding question text and then search based on the question text. Similarly, the candidate answer information and the evidence information are in the form of text, voice or other forms.

[0144] In the embodiments of the present application, the search approaches of the candidate answer information and the evidence information are different. The candidate answer information is obtained by searching in a specific question and answer system, and the evidence information is obtained by searching on a public search application.

[0145] In a possible implementation, the computer device performs searching based on the question information, and extracts candidate answer information from the searched information. The searched information can be a text, an article or other forms. Since the searched information can also include other information irrelevant to the question information, the searched information is further processed to extract the candidate answer information. For example, for an article, a single document MRC (Machine Reading Comprehension) is used to perform reading comprehension on the article, to determine a text segment in the article that is related to the question text, and the determined text segment is extracted as the candidate answer information of the question information. Of course, other ways can also be used to extract the candidate answer information from the searched information, and the application embodiments do not limit the way of extracting the candidate answer information.

[0146] Optionally, since there can be multiple ways of expressing the same question, some information can be missed when searching in only one way of expression. Therefore, in the application embodiments, other question information with the same meaning as the question information is also obtained, and searching is performed based on the question information and the other question information with the same meaning as the question information, to extract candidate answer information from the searched information. For example, the question information is “Why is it very harmful to the body to stay up late at night”, and the other question information with the same meaning as the question information can be “Why is it harmful to the body to stay up late”, or “What harm can staying up late at night cause to the body”, and the like.

[0147] Optionally, the user inputs the question information in the question and answer system, and the corresponding candidate answer information is searched by the question and answer system. Taking the medical field as an example, the question and answer system is a question and answer system specially used for searching medical knowledge.

[0148] In a possible implementation, the computer device performs searching based on the question information on a public search application, and arranges the information appearing on the first page of the search application and ranked in the first n positions as the evidence information.

[0149] It should be noted that the computer device can search for multiple candidate answer information corresponding to the question information, and the application embodiments only take any candidate answer information in the multiple candidate answer information as an example for description.

[0150] 302. The computer device obtains first classification features based on the question information and the candidate answer information.

[0151] In an embodiment of the present application, the first classification feature obtained based on the question information and the candidate answer information at least represents semantic information of the candidate answer information. The semantic information of the candidate answer information refers to the meaning of the candidate answer information. The first classification feature can be represented in the form of a vector, a matrix or other forms.

[0152] In a possible implementation, the computer device combines the question information and the candidate answer information to obtain combined information, and extracts features from the combined information to obtain the first classification feature. The first classification feature represents semantic information of the candidate answer information, semantic information of the question information, and an association relationship between the question information and the candidate answer information.

[0153] For example, the combination of the question information and the candidate answer information is: [CLS] question information [SEP] candidate answer information [SEP], or [CLS] candidate answer information [SEP] question information [SEP]. [SEP] is a separator symbol, which facilitates distinguishing the question information and the candidate answer information in subsequent processing according to the separator symbol. [CLS] is a special marker symbol, which can collect the overall semantic information of the question information and the candidate answer information.

[0154] 303. The computer device obtains evidence features of the evidence information.

[0155] In an embodiment of the present application, the computer device extracts features from the evidence information to obtain evidence features of the evidence information. The evidence features represent semantic information of the evidence information. The evidence features can be represented in the form of a vector, a matrix or other forms.

[0156] In a possible implementation, the evidence information includes a plurality of evidence segments. Some of the plurality of evidence segments have a higher importance degree, and the other evidence segments have a lower importance degree. When obtaining the evidence features of the evidence information, the evidence segments with a higher importance degree are considered more. Therefore, the computer device respectively obtains evidence segment features corresponding to the plurality of evidence segments, respectively adjusts the plurality of evidence segment features based on evidence weights corresponding to the plurality of evidence segments, the evidence weights indicating the importance degrees of the evidence segments, and splices the adjusted plurality of evidence segment features to obtain the evidence features. Adjusting the evidence segment features based on the evidence weights can highlight the evidence segments with a higher importance degree, so that the spliced evidence features contain more features of the evidence segments with a higher importance degree.

[0157] Optionally, the following formula one is used to obtain the evidence weight corresponding to each evidence segment:

[0158]

[0159] wherein, scoref , softmax(·) is an activation function, and W is a learnable matrix, E m , represents the evidence fragment feature of the evidence fragment m, , represents the transposed matrix of the evidence fragment feature of the evidence fragment m, and a1 is a reference parameter. Wherein, W l×h , l represents the number of evidence fragments, h represents the dimension of the evidence fragment feature, and R l×h , represents an l x h matrix.

[0160] Alternatively, the above formula one can be replaced by m .

[0161] The obtained evidence fragment is adjusted by using the following formula two:

[0162]

[0163] , wherein, , represents the adjusted evidence fragment feature, score f , represents the evidence weight of the evidence fragment m, E m , represents the evidence fragment feature before adjustment.

[0164] 304, The computer device matches the question feature and the evidence feature of the question information to obtain a second confidence degree corresponding to the evidence information.

[0165] Since the evidence information is also obtained from the network, it cannot be guaranteed that the evidence information is completely accurate, so before adjusting the first classification feature based on the evidence feature, it is also necessary to determine the credibility of the evidence information as the evidence of the question information.

[0166] , wherein the question feature is obtained by feature extraction on the question information, and the second confidence degree represents the credibility of the evidence information as the evidence of the question information.

[0167] Alternatively, the following formula three is used to obtain the second confidence degree:

[0168]

[0169] , wherein score e , represents the second confidence degree, softmax(·) is an activation function, q represents the question feature, , represents the evidence feature, , represents the transposed matrix of the evidence feature, and a2 represents the reference parameter.

[0170] Alternatively, the above formula three can be replaced by ​​

[0171] 305、The computer device adjusts the evidence feature based on the second confidence, to obtain an adjusted evidence feature.

[0172] Since the second confidence can identify the credibility of the evidence information as the evidence of the question information, the greater the second confidence is, the more credible the evidence information is, and at this time, the adjusted evidence feature based on the second confidence can further strengthen the semantic information of the evidence information; the smaller the second confidence is, the less credible the evidence information is, and at this time, the adjusted evidence feature based on the second confidence can weaken the semantic information of the evidence information. Thus, when subsequent processing is performed based on the evidence feature, the evidence feature can more accurately play the role of evidence.

[0173] Optionally, the evidence feature is adjusted by using the following Formula Four:

[0174]

[0175] wherein, represents the adjusted evidence feature, score e represents the second confidence, represents the transpose matrix of the evidence feature before adjustment.

[0176] Alternatively, the score in the above Formula Four can be replaced by

[0177] It should be noted that steps 204-205 are optional, and in another embodiment, after the evidence feature is obtained through the above step 203, step 206 can be directly executed.

[0178] 306、The computer device adjusts the evidence feature based on the second confidence, to obtain an adjusted evidence feature.

[0179] wherein, the association degree represents the correlation between the first classification feature and the evidence feature, the greater the association degree is, the more similar the candidate answer information is to the evidence information, which means that the greater the possibility that the candidate answer information can be the answer information of the question information is; the smaller the association degree is, the greater the difference between the candidate answer information and the evidence information is, which means that the smaller the possibility that the candidate answer information can be the answer information of the question information is.

[0180] ​Then, the evidence feature is adjusted based on the correlation degree to obtain a second classification feature, which represents semantic information of the evidence information and a support degree of the evidence information to the candidate answer information. The greater the correlation degree is, the more obvious the semantic information related to the candidate answer information in the second classification feature is, and the more capable the candidate answer information is of proving that the candidate answer information is related to the evidence information in the subsequent classification process. The smaller the correlation degree is, the less obvious the semantic information related to the candidate answer information in the second classification feature is, and the less capable the candidate is of proving that the candidate answer information is related to the evidence information in the classification process.

[0181] Optionally, the following Formula Five is used to calculate the correlation degree between the first classification feature and the evidence feature:

[0182]

[0183] wherein score a represents the correlation degree, softmax(·) is an activation function, represents the first classification feature corresponding to the candidate answer information i, and represents the candidate answer information i, represents a transpose matrix of the evidence feature, and a3 represents a reference parameter.

[0184] Alternatively, the score in the above Formula Five can be replaced by

[0185] The following Formula Six is used to adjust the first classification feature:

[0186]

[0187] wherein score represents the second classification feature corresponding to the candidate answer information i, and score a represents the correlation degree, represents a transpose matrix of the evidence feature.

[0188] Alternatively, the score in the above Formula Six can be replaced by

[0189] 307. The computer device performs classification based on the first classification feature and the second classification feature to obtain a classification result corresponding to the candidate answer information.

[0190] ​​The classification result indicates whether the candidate answer information is the answer information of the question information. Optionally, the classification result is represented by 0 and 1. When the classification result is 1, it indicates that the candidate answer information is the answer information of the question information. When the classification result is 0, it indicates that the candidate answer information is not the answer information of the question information. Alternatively, the classification result is represented by a probability. When the probability is greater than a threshold value, it indicates that the candidate answer information is the answer information of the question information. When the probability is less than the threshold value, it indicates that the candidate answer information is not the answer information of the question information. For example, the threshold value is 0.7, 0.8, or other numerical values.

[0191] In a possible implementation, the computer device splices the first classification feature and the second classification feature to obtain a spliced classification feature, and classifies the spliced classification feature to obtain the classification result.

[0192] In another possible implementation, the computer device fuses the first classification feature and the second classification feature to obtain a fused classification feature, splices the first classification feature, the second classification feature, and the fused classification feature to obtain a fifth classification feature of the candidate answer information, and classifies based on the fifth classification feature to obtain the classification result. The fused classification feature can strengthen the semantic information corresponding to the same part of information in the candidate answer information as the evidence information. When subsequent classification is performed based on the fifth classification feature, the semantic information of the part of information in the candidate answer information that can be supported by the evidence information can be highlighted, so that the obtained classification result is more accurate.

[0193] In another possible implementation, the computer device performs weighted processing on the second classification feature according to a first confidence degree corresponding to the evidence information to obtain a weighted second classification feature, and classifies based on the first classification feature and the weighted second classification feature to obtain the classification result. The first confidence degree refers to the credibility of the evidence information, that is, only the credibility of the evidence information itself is considered. The greater the first confidence degree, the more credible the evidence information. The smaller the first confidence degree, the less credible the evidence information. The weighted second classification feature can strengthen the credible evidence information in the second classification feature because the credibility of the evidence information is considered. When subsequent classification is performed based on the weighted second classification feature, the semantic information of the part of information in the evidence information that is credible can be highlighted, so that the obtained classification result is more accurate.

[0194] Optionally, the first confidence is determined according to the frequency of the evidence segments in the evidence information. However, since two evidence segments may represent the same meaning although they are not expressed in the same way, in order to avoid regarding the two evidence segments as different evidence segments, in the embodiment of the present application, whether the two evidence segments can be regarded as the same evidence segment is determined by calculating the segment similarity between the two evidence segments. The computer device obtains the segment similarity between each two evidence segments in the plurality of evidence segments; for each evidence segment, the segment confidence corresponding to the evidence segment is determined based on the plurality of segment similarities corresponding to the evidence segment; and the segment confidences corresponding to the plurality of evidence segments are spliced to obtain the first confidence.

[0195] Optionally, the plurality of segment similarities are obtained by using the following Formula Seven:

[0196]

[0197] wherein score c represents the plurality of segment similarities, represents the evidence feature, represents the transposed matrix of the evidence feature, and a4 represents a reference parameter. Since the evidence feature is obtained by splicing the plurality of evidence segment features, the plurality of segment similarities can be obtained by directly multiplying and .

[0198] Alternatively, Formula Seven can be expressed as:

[0199]

[0200] wherein score mg represents the segment similarity between the evidence segment m and the evidence segment g, represents the evidence segment feature of the evidence segment m, represents the evidence segment feature of the evidence segment g.

[0201] Optionally, the plurality of segment similarities are averaged to obtain the corresponding segment confidence, and the segment confidences corresponding to the plurality of evidence segments are spliced to obtain the first confidence.

[0202] For example, the first confidence is obtained by using the following Formula Eight:

[0203] Formula Eight: score confidence = sigmoid(mean(score c ))

[0204] wherein score confidence represents the first confidence, sigmoid(·) is an activation function, and mean(scorec ) represents the average of the similarity of multiple segments. Since score c The average is taken without further stitching because the segment similarity is included.

[0205] The method provided in the embodiments of the present application introduces the evidence information corresponding to the question information in the process of determining the correctness of the candidate answer information, and takes the evidence information as the evidence proving whether the candidate answer information can be the answer information of the question information, so that the second classification feature obtained by adjusting the evidence feature based on the association degree between the first classification feature corresponding to the candidate answer information and the evidence feature can not only reflect the semantic information of the evidence information itself, but also reflect the support degree of the evidence information to the candidate answer information, so that when classification is performed based on the first classification feature and the second classification feature, the influence of the evidence information can be considered to obtain a more accurate classification result, and the accuracy of the determined answer information is improved.

[0206] In addition, in the embodiments of the present application, the evidence feature is adjusted based on the second confidence, in the case that the evidence information is credible, the semantic information of the evidence information can be further strengthened, and in the case that the evidence information is not credible, the semantic information of the evidence information can be weakened, so that when the evidence feature is processed, the evidence feature can more accurately play the role of evidence.

[0207] The above Figure 3 The embodiments shown in the above Figure 4 In the case that the question information has multiple candidate answer information, for each candidate answer information, the influence of other candidate answer information on the candidate answer information is considered to classify the candidate answer information.

[0208] Figure 4 is a flowchart of an information processing method provided by an embodiment of the present application. The execution subject of the embodiment of the present application is a computer device. Referring to Figure 4 The method comprises the following steps:

[0209] 401. The computer device obtains multiple candidate answer information corresponding to the question information.

[0210] The implementation manner of obtaining the candidate answer information in step 401 is the same as that of obtaining the candidate answer information in step 301, and will not be described here.

[0211] 402. The computer device obtains multiple first classification features based on the question information and the multiple candidate answer information.

[0212] The implementation of obtaining the first classification features in step 402 is the same as the implementation of obtaining the first classification features in step 302, which will not be repeated here.

[0213] 403. The computer device obtains the answer similarity between each two candidate answer information in the plurality of candidate answer information.

[0214] In the embodiment of the present application, the computer device matches the first classification features corresponding to each two candidate answer information to obtain the answer similarity between the each two candidate answer information. The higher the answer similarity between the two candidate answer information is, the more the one candidate answer information can support the other candidate answer information.

[0215] Optionally, for each two candidate answer information, the following formula nine is used to obtain the answer similarity:

[0216]

[0217] wherein, i represents the candidate answer information i, j represents the candidate answer information j, s i,j represents the answer similarity between the candidate answer information i and the candidate answer information j, represents the first classification features corresponding to the candidate answer information i, represents the transpose matrix of the first classification features corresponding to the candidate answer information i, represents the first classification features corresponding to the candidate answer information j.

[0218] In the formula nine, when i=j, the answer similarity is 0, and in other cases, the answer similarity is calculated according to

[0219] Alternatively, the in the formula nine can be replaced by Alternatively, the in the formula nine can be replaced by represents the transpose matrix of the first classification features corresponding to the candidate answer information j.

[0220] 404. The computer device obtains the third classification features based on the plurality of answer similarities corresponding to any candidate answer information and the first classification features corresponding to the other candidate answer information except the candidate answer information in the plurality of candidate answer information.

[0221] ​For any candidate answer information, the influence of other candidate answer information on the candidate answer information is considered, the more similar the other candidate answer information is to the candidate answer information, the more the candidate answer information can be supported by the other candidate answer information, and subsequently, when determining the classification result of the candidate answer information, the candidate answer information can be classified according to the other candidate answer information. Wherein, the third classification feature is obtained based on the first classification feature corresponding to the other candidate answer information and the plurality of answer similarities, and the third classification feature contains the same semantic information in the other candidate answer information as the candidate answer information.

[0222] In a possible implementation, the computer device normalizes the plurality of answer similarities to obtain normalized plurality of answer similarities; and performs weighted processing on the first classification features corresponding to the other candidate answer information than the candidate answer information in the plurality of candidate answer information based on the normalized plurality of answer similarities to obtain the third classification features.

[0223] Optionally, the plurality of answer similarities are normalized by using the following formula ten:

[0224]

[0225] Wherein, α i,j represents the normalized answer similarity corresponding to the candidate answer information i and the candidate answer information j, n represents the total number of candidate answer information, k represents the candidate answer information k, e represents the natural logarithm base, and represents the answer similarity represented based on the natural logarithm base, represents the sum of n answer similarities represented based on the natural logarithm base.

[0226] Optionally, the plurality of first classification features are weighted processed based on the normalized plurality of answer similarities by using the following formula eleven:

[0227]

[0228] Wherein, represents the third classification feature corresponding to the candidate answer information i, α i,j represents the normalized answer similarity corresponding to the candidate answer information i and the candidate answer information j, represents the first classification feature corresponding to the candidate answer information j.

[0229] 405、The computer device classifies any candidate answer information based on the first classification feature and the third classification feature corresponding to the candidate answer information to obtain the classification result corresponding to the candidate answer information.

[0230] In a possible implementation, the computer device concatenates the first classification feature and the third classification feature to obtain a concatenated classification feature, and classifies the concatenated classification feature to obtain the classification result.

[0231] In another possible implementation, the computer device fuses the first classification feature and the third classification feature to obtain a fused classification feature, concatenates the first classification feature, the third classification feature, and the fused classification feature to obtain a sixth classification feature of the candidate answer information, and classifies based on the sixth classification feature to obtain the classification result. The fused classification feature can strengthen semantic information corresponding to the same part of information in the candidate answer information as other candidate answer information, and when subsequent classification is performed based on the sixth classification feature, semantic information in the candidate answer information that can be supported by other candidate answer information can be highlighted, so that the obtained classification result is more accurate.

[0232] It should be noted that, Figure 4 The embodiments shown only take obtaining a classification result corresponding to any candidate answer information in a plurality of candidate answer information as an example for description, and in another embodiment, for each candidate answer information in the plurality of candidate answer information, the corresponding classification result can be obtained by using the above-mentioned implementation.

[0233] In the method provided by the embodiments of the present application, in the process of determining the correctness of the candidate answer information, a plurality of candidate answer information is introduced, and the plurality of candidate answer information can be used as evidence to prove whether a certain candidate answer information can be the answer information of the question information, so that based on the answer similarity between each two candidate answer information, the third classification feature contains the same semantic information in other candidate answer information as the candidate answer information, and when classification is performed based on the first classification feature and the third classification feature, the influence between the plurality of candidate answer information is considered, and a more accurate classification result can be obtained, and the accuracy of the determined answer information is improved.

[0234] The above Figure 3 and Figure 4 The embodiments of the above Figure 5 , consider the influence of the evidence information and the other candidate answer information except the candidate answer information for any candidate answer information in the plurality of candidate answer information, and refer to In another embodiment, the influence of the evidence information and the other candidate answer information except the candidate answer information can be considered at the same time. After the second classification feature is obtained based on the steps 301-306 and the third classification feature is obtained based on the steps 401-404, the step 501 is performed.

[0235] 501、The computer device classifies the first classification feature, the second classification feature, and the third classification feature corresponding to the candidate answer information, to obtain a classification result corresponding to the candidate answer information.

[0236] In a possible implementation, the computer device fuses the first classification feature and the third classification feature to obtain a fused classification feature, concatenates the first classification feature, the third classification feature, and the fused classification feature to obtain a fourth classification feature of the candidate answer information, and classifies based on the second classification feature and the fourth classification feature to obtain the classification result. For example, the fourth classification feature is represented as:

[0237]

[0238] wherein, denotes the fourth classification feature, denotes the first classification feature corresponding to the candidate answer information i, denotes the third classification feature corresponding to the candidate answer information i, denotes the fused classification feature of the first classification feature and the third classification feature.

[0239] In another possible implementation, the computer device fuses the first classification feature and the second classification feature to obtain a fused classification feature, concatenates the first classification feature, the second classification feature, and the fused classification feature to obtain a fifth classification feature of the candidate answer information, and classifies based on the third classification feature and the fifth classification feature to obtain the classification result. For example, the fifth classification feature is represented as:

[0240]

[0241] wherein, denotes the fifth classification feature, denotes the first classification feature corresponding to the candidate answer information i, denotes the second classification feature corresponding to the candidate answer information i, denotes the fused classification feature of the first classification feature and the second classification feature.

[0242] In another possible implementation, the computer device fuses the first classification feature and the third classification feature to obtain a fused classification feature, concatenates the first classification feature, the third classification feature, and the fused classification feature to obtain a fourth classification feature of the candidate answer information, fuses the first classification feature and the second classification feature to obtain a fused classification feature, and concatenates the first classification feature, the second classification feature, and the fused classification feature to obtain a fifth classification feature of the candidate answer information, and classifies based on the fourth classification feature and the fifth classification feature to obtain the classification result.

[0243] Optionally, the fifth classification feature comprises the second classification feature, and the fifth classification feature is weighted based on the first confidence to obtain a weighted fifth classification feature, and classification is performed based on the fourth classification feature and the weighted fifth classification feature to obtain a classification result.

[0244] Optionally, the fourth classification feature and the weighted fifth classification feature are fused, and the fused classification feature is determined as a final classification feature corresponding to the candidate answer information, and classification is performed based on the fused classification feature to obtain a classification result.

[0245] For example, the fourth classification feature and the weighted fifth classification feature are fused by using the following formula twelve:

[0246]

[0247] wherein g i represents the classification feature after the fourth classification feature and the weighted fifth classification feature are fused, represents the fourth classification feature, represents the fifth classification feature, and score confidence represents the first confidence.

[0248] In another possible implementation, the computer device weights the second classification feature based on the first confidence corresponding to the evidence information to obtain a weighted second classification feature, and performs classification based on the first classification feature, the weighted second classification feature, and the third classification feature to obtain a classification result.

[0249] In addition, in a possible implementation, in the case that the classification result is a probability representation, for a plurality of candidate answer information, the computer device respectively obtains a classification result corresponding to each candidate answer information, and in the case that a plurality of answer information are determined from the plurality of candidate answer information, the most accurate answer information can also be determined from the plurality of answer information according to the probability size.

[0250] It should be noted that the computer device is only taken as an example in the above embodiments, and in another embodiment, the terminal and the server can interactively perform. The terminal obtains user input question information, sends the question information to the server, and the server performs the above embodiments to obtain a classification result, determines answer information of the question information, and then returns the answer information to the terminal. The terminal displays the answer information. For example, the answer information can be displayed in a text form or played in a voice form.

[0251] It should be noted that the steps 301-306 and the steps 401-404 can be executed simultaneously, or the steps 301-306 can be executed first, then the steps 401-404 are executed, or the steps 401-404 can be executed first, then the steps 301-306 are executed, and the execution sequence of the embodiments of the present application is not limited.

[0252] The overall process of the embodiments of the present application is described with reference to Figure 6 First, based on the problem information, the information obtained by retrieval is extracted to obtain a plurality of candidate answer information, and the evidence information is obtained by crawling the network, the evidence feature is obtained by semantic coding of the evidence information, the plurality of first classification features are obtained by coding based on the plurality of candidate answer information, for each candidate answer information, the evidence information and the corresponding first classification feature are compared (answer-evidence comparison), and the plurality of first classification features are compared (multi-answer comparison), and the classification result corresponding to each candidate answer information is obtained according to the results of the two comparisons.

[0253] In the method provided by the embodiments of the present application, in the process of determining the correctness of the candidate answer information, the evidence information corresponding to the problem information is introduced, and the evidence information is used as evidence to prove whether the candidate answer information can be the answer information of the problem information, so that the second classification feature obtained by adjusting the evidence feature based on the correlation between the first classification feature corresponding to the candidate answer information and the evidence feature of the evidence information can not only reflect the semantic information of the evidence information itself, but also reflect the support degree of the evidence information to the candidate answer information; and the plurality of candidate answer information is introduced, and the plurality of candidate answer information can be used as evidence to prove whether a certain candidate answer information can be the answer information of the problem information, so that the third classification feature obtained based on the answer similarity between each two candidate answer information contains the same semantic information in other candidate answer information as the candidate answer information, so that when the classification is performed based on the first classification feature, the second classification feature and the third classification feature, the influence of the evidence information and the influence between the plurality of candidate answer information are considered, and a more accurate classification result can be obtained, and the accuracy of the determined answer information is improved.

[0254] In addition, in some embodiments, the computer device can call the information processing model to process the problem information, the candidate answer information and the evidence information to obtain the classification result corresponding to the candidate answer information. First, the structure of the information processing model is described.

[0255] Reference is made to Figure 7A schematic diagram of an information processing model is shown. The information processing model includes a first processing sub-model 701, a second processing sub-model 702, and a classification sub-model 703. The first processing sub-model 701, the second processing sub-model 702, and the classification sub-model 703 are connected with each other.

[0256] In a possible implementation, referring to Figure 8 the first processing sub-model 701 includes a first extraction network 711 and a first attention network 721. The first extraction network 711 is connected with the first attention network 721, the first extraction network 711 is configured to obtain first classification features based on the question information and the candidate answer information, and the first attention network 721 is configured to obtain answer similarities between each two of the candidate answer information, and for each candidate answer information, the first attention network 721 is further configured to obtain third classification features based on the corresponding multiple answer similarities of the candidate answer information and the first classification features of other candidate answer information in the multiple candidate answer information except the candidate answer information.

[0257] In a possible implementation, referring to Figure 9 the second processing sub-model 702 includes a second extraction network 712, a second attention network 722, and a third attention network 732. The second extraction network 712, the second attention network 722, and the third attention network 732 are connected in sequence, the second extraction network 712 is configured to obtain evidence features of the evidence information corresponding to the question information, the second attention network 722 is configured to adjust the evidence features based on the association degrees between the first classification features and the evidence features, to obtain second classification features of the candidate answer information, and the third attention network 732 is configured to match the question features and the evidence features to obtain second confidences of the evidence information, and adjust the evidence features based on the second confidences.

[0258] Optionally, the second extraction network 712 includes a feature extraction layer and an attention layer. The feature extraction layer is connected with the attention layer, the feature extraction layer is configured to obtain evidence segment features corresponding to multiple evidence segments respectively, and the attention layer is configured to adjust the multiple evidence segment features based on evidence weights corresponding to the multiple evidence segments respectively, splice the adjusted multiple evidence segment features to obtain the evidence features.

[0259] Optionally, the second processing sub-model 702 further includes a self-attention network 742, the self-attention network 742 is connected with the second extraction network 712, the self-attention network 742 is configured to obtain segment similarities between each two of the multiple evidence segments, for each evidence segment, determine a segment confidence corresponding to the evidence segment based on multiple segment similarities corresponding to the evidence segment, and splice the segment confidences corresponding to the multiple evidence segments to obtain the first confidence.

[0260] In one possible implementation, see [link to relevant documentation]. Figure 10 The classification sub-model 703 includes a fusion network 713 and a classification network 723. The fusion network 713 and the classification network 723 are connected. The fusion network 713 is used at least to weight the second classification features based on the first confidence level corresponding to the evidence information to obtain the weighted second classification features. The classification network 723 is used to classify based on the first classification features and the weighted second classification features to obtain the classification result.

[0261] Based on the above Figures 7-10 The model structure of the information processing model shown is described in [reference]. Figure 11 The information processing model is invoked to process the question information, candidate answer information, and evidence information to obtain the classification results corresponding to the candidate answer information. Figure 11 This is a flowchart illustrating an information processing method provided in an embodiment of this application. The execution subject of this embodiment is a computer device. See also... Figure 11 The method includes the following steps:

[0262] 1101. The computer device calls the first processing sub-model to obtain the first classification feature based on the question information and candidate answer information.

[0263] In one possible implementation, the computer device invokes the first extraction network to obtain first classification features based on question information and candidate answer information.

[0264] 1102. The computer device calls the second processing sub-model to obtain the evidence features of the evidence information corresponding to the problem information.

[0265] In one possible implementation, the computer device invokes a second extraction network to obtain the evidentiary features of the evidentiary information corresponding to the problem information.

[0266] Optionally, the computer device calls the feature extraction layer to obtain the evidence fragment features corresponding to multiple evidence fragments; calls the attention layer to adjust the multiple evidence fragment features based on the evidence weights corresponding to the multiple evidence fragments; and splices the adjusted multiple evidence fragment features to obtain the evidence features.

[0267] 1103. The computer device calls the second processing sub-model, and adjusts the evidence features based on the correlation between the first classification features and the evidence features to obtain the second classification features.

[0268] In one possible implementation, the computer device invokes a second attention network to adjust the evidence features based on the correlation between the first classification features and the evidence features, thereby obtaining the second classification features of the candidate answer information.

[0269] In one possible implementation, the computer device invokes a third attention network to match the problem features and evidence features to obtain a second confidence level corresponding to the evidence information. Based on the second confidence level, the evidence features are adjusted to obtain the adjusted evidence features.

[0270] 1104. The computer device calls the classification sub-model, classifies based on the first classification feature and the second classification feature, and obtains the classification result corresponding to the candidate answer information. The classification result indicates whether the candidate answer information is the answer information of the question information.

[0271] In one possible implementation, the computer device invokes a fusion network to weight the second classification features based on the first confidence level corresponding to the evidence information, thereby obtaining the weighted second classification features; and invokes a classification network to classify based on the first classification features and the weighted second classification features, thereby obtaining the classification result.

[0272] Optionally, the computer device invokes an attention network to obtain the segment similarity between every two evidence segments in multiple evidence segments; for each evidence segment, the segment confidence corresponding to the evidence segment is determined based on the segment similarity corresponding to the evidence segment; the segment confidence corresponding to multiple evidence segments is concatenated to obtain the first confidence.

[0273] The implementation methods of steps 1101-1104 are the same as those of steps 302-307 in the above embodiments, and will not be repeated here.

[0274] The method provided in this application introduces evidence information corresponding to the question information during the process of determining the correctness of candidate answer information. This evidence information serves as proof of whether the candidate answer information can be the answer to the question information. Based on the correlation between the first classification feature corresponding to the candidate answer information and the evidence feature of the evidence information, the second classification feature obtained by adjusting the evidence feature not only reflects the semantic information of the evidence information itself, but also reflects the degree of support of the evidence information for the candidate answer information. Thus, when calling the information processing model to classify based on the first and second classification features, the influence of the evidence information is considered, resulting in a more accurate classification result and improving the accuracy of the determined answer information.

[0275] With the above Figure 4 Similar to the embodiments shown, the embodiments of this application can invoke an information processing model to classify candidate answer information by considering the influence of other candidate answer information on each candidate answer information when the question information has multiple candidate answer information. Figure 12 This is a flowchart illustrating an information processing method provided in an embodiment of this application. The execution subject of this embodiment is a computer device. See also... Figure 12The method comprises the following steps:

[0276] 1201. The computer device invokes a first processing sub-model, and obtains a plurality of first classification features based on the question information and the plurality of candidate answer information.

[0277] In a possible implementation, the computer device invokes a first extraction network, and obtains the first classification features based on the question information and the candidate answer information.

[0278] 1202. The computer device invokes the first processing sub-model, and obtains an answer similarity between each two pieces of candidate answer information in the plurality of candidate answer information.

[0279] In a possible implementation, the computer device invokes a first attention network, and obtains the answer similarity between each two pieces of candidate answer information in the plurality of candidate answer information.

[0280] 1203. The computer device invokes the first processing sub-model, and for any piece of candidate answer information, obtains a third classification feature based on a plurality of answer similarities corresponding to the piece of candidate answer information and the first classification features corresponding to other pieces of candidate answer information in the plurality of candidate answer information except the piece of candidate answer information.

[0281] In a possible implementation, for each piece of candidate answer information, the computer device invokes the first attention network, and obtains the third classification feature based on the plurality of answer similarities corresponding to the piece of candidate answer information and the first classification features corresponding to other pieces of candidate answer information in the plurality of candidate answer information except the piece of candidate answer information.

[0282] 1204. The computer device invokes a classification sub-model, and for any piece of candidate answer information, performs classification based on the first classification feature and the third classification feature corresponding to the piece of candidate answer information, to obtain a classification result corresponding to the piece of candidate answer information.

[0283] The embodiments of steps 1201-1204 are the same as the embodiments of steps 402-405 in the above-described embodiments, and will not be described here again.

[0284] The method provided in the embodiments of the present application introduces a plurality of candidate answer information in the process of determining the correctness of the candidate answer information. The plurality of candidate answer information can serve as evidence to prove whether a certain piece of candidate answer information can be the answer information of the question information, so that the third classification feature obtained based on the answer similarity between each two pieces of candidate answer information contains the same semantic information as other pieces of candidate answer information. Therefore, when the information processing model is invoked and classification is performed based on the first classification feature and the third classification feature, the influence between the plurality of candidate answer information is considered, and a more accurate classification result can be obtained, thereby improving the accuracy of the determined answer information.

[0285] Similar to the embodiments shown above Figure 5 , in the embodiments of the present application, based on the second classification features obtained in steps 1101-1103 and the third classification features obtained in steps 1201-1203, the computer device can further call a classification sub-model to classify the first classification features, the second classification features and the third classification features corresponding to the candidate answer information, and obtain the classification result corresponding to the candidate answer information.

[0286] Taking an example of the question information having three candidate answer information, referring to Figure 13 , each candidate answer information is spliced with the question information, and the three combined information after splicing is input into the first extraction network in the first processing sub-model, and the first extraction network is called to encode the three combined information respectively, and three first classification features and are obtained. and are then input into the first attention network, and the first attention network processes and to obtain the third classification features corresponding to each candidate answer information and Meanwhile, the question information and the evidence information are input into the second extraction network in the second processing sub-model respectively, and the second extraction network is called to obtain the question feature of the question information and the evidence feature of the evidence information respectively. The evidence feature is then input into the third attention network, the third attention network is called to match the question feature and the evidence feature, and the second confidence corresponding to the evidence information is obtained, based on which the evidence feature is adjusted to obtain the adjusted evidence feature (in Figure 13 which is represented by three evidence segment features), and then the adjusted evidence feature and the three first classification features and are input into the second attention network to obtain the second classification features corresponding to each candidate answer information and After that, the three first classification features, the three second classification features and the three third classification features are input into the classification sub-model to obtain the classification result corresponding to each candidate answer information respectively.

[0287] In addition, the information processing model in the above embodiment needs to be trained before use, which can be trained by the computer device or other computer devices, and the device for training the information processing model is not limited in the embodiment of the application.

[0288] The computer device obtains sample question information, sample candidate answer information, sample evidence information, and a sample classification result corresponding to the sample candidate answer information, calls the information processing model, processes the sample question information, the sample candidate answer information, and the sample evidence information, and obtains a predicted classification result corresponding to the sample candidate answer information, and trains the information processing model based on the sample classification result and the predicted classification result.

[0289] Optionally, the information processing model is trained by using the following loss function:

[0290] L = -(y·logp + (1-y)·log(1-p))

[0291] Wherein, L represents a loss value, p represents a predicted classification result, and y represents a sample classification result.

[0292] In a possible implementation, the computer device combines the sample question information and the sample candidate answer information to obtain sample combined information, calls the information processing model, processes the sample question information, the sample combined information, and the sample evidence information, and obtains a predicted classification result corresponding to the sample candidate answer information.

[0293] It should be noted that the training process described above is only an example of one-time training, and in another embodiment, the information processing model can be iteratively trained multiple times.

[0294] It should be noted that the sample answer information in the one-time training process described above is one or more.

[0295] The information processing method shown in the above embodiment can be applied in various scenarios, such as medical scenarios, examination scenarios, intelligent transportation scenarios, or other scenarios. The following takes the medical scenario as an example, and the information processing method is described with reference to Figure 14 The method is performed by a computer device, and the information processing process is described.

[0296] 1401. Obtain a plurality of candidate medical answer information corresponding to medical question information and medical evidence information.

[0297] 1402. Based on the medical question information and the plurality of candidate medical answer information, respectively obtain a first classification feature corresponding to each of the plurality of candidate medical answer information.

[0298] 1403. Obtain an evidence feature of the medical evidence information and a question feature of the medical question.

[0299] 1404、match the question features and the evidence features to obtain a second confidence degree corresponding to the medical evidence information.

[0300] 1405、adjust the evidence features based on the second confidence degree to obtain adjusted evidence features.

[0301] 1406、adjust the evidence features based on the association degree between the first classification features and the evidence features to obtain second classification features.

[0302] 1407、obtain an answer similarity between each two candidate medical answer information in the plurality of candidate medical answer information.

[0303] 1408、for each candidate medical answer information, obtain third classification features based on the plurality of answer similarities corresponding to the candidate medical answer information and the first classification features corresponding to other candidate medical answer information in the plurality of candidate medical answer information except the candidate medical answer information.

[0304] 1409、classify each candidate medical answer information based on the first classification features, the second classification features and the third classification features corresponding to the candidate medical answer information to obtain a classification result corresponding to the candidate medical answer information.

[0305] In a possible implementation manner, a user can input medical question information in a medical question and answer system, and a computer device searches the input medical question information through the medical question and answer system to obtain a plurality of candidate medical answer information and medical evidence information.

[0306] The implementation manners of steps 1401-1409 are the same as those shown in the implementation manners shown in Figure 3 、 Figure 4 and Figure 5 , and will not be described here.

[0307] In addition, in order to illustrate the accuracy of the information processing method provided in the embodiments of the present application, the information processing method provided in the embodiments of the present application (denoted as the present method in the following table) is compared with the information processing method in the prior art which only considers one candidate answer information (denoted as baseline in the following table), and see Table 1:

[0308] Table 1

[0309]

[0310] As can be seen from the above Table 1, compared with the baseline, the present method can improve the accuracy of judging the correct answer as much as possible (+12%) on the basis of ensuring the recall of the correct answer information.

[0311] For another example, see Figure 15As shown in the schematic diagram, the medical question information is "what is the reason for low fever in the afternoon every day", and the baseline is used to determine the treatment scheme (candidate medical answer information) of low fever "clinically, it is generally not recommended to take medicine to reduce the temperature for low fever, and it is suggested to drink more hot water when the fever is below 38 degrees...". In order to the medical answer information of the medical question information, the method provided by the embodiment of the present application searches the related medical evidence information "there are mainly the following reasons for low fever every day: first, the infection caused by the factor needs to be excluded...", "low fever every day is very likely to be caused by viral infection, and it may also be due to bacteria in the body...", "low fever in the afternoon every day, first of all, it is necessary to consider whether there is tubercle bacillus infection...", "finally, it is necessary to consider whether there is non-infectious disease", "there are many reasons for fever, which can be divided into infectious fever and non-infectious fever", and "low fever occurs every day, which is generally related to chronic infectious diseases and metabolic diseases". In the embodiment of the present application, the medical evidence information introduced is similar between each evidence segment, so the credibility of the medical evidence information is relatively high, and the medical evidence information is also related to the medical question information, but the correlation degree between the medical evidence information and the candidate medical answer information is very low, so the information processing method provided by the embodiment of the present application is used to determine that the candidate medical answer information is the wrong answer.

[0312] Figure 16 is a structural schematic diagram of an information processing device provided by an embodiment of the present application. Referring to Figure 16 , the device comprises:

[0313] A first feature acquisition module 1601 is configured to acquire a first classification feature based on question information and candidate answer information, the first classification feature at least representing semantic information of the candidate answer information.

[0314] A second feature acquisition module 1602 is configured to acquire evidence features of evidence information corresponding to the question information, the evidence information being used to provide evidence that the candidate answer information is answer information of the question information.

[0315] A first feature adjustment module 1603 is configured to adjust the evidence features based on a correlation degree between the first classification features and the evidence features, to obtain second classification features.

[0316] A classification module 1604 is configured to perform classification based on the first classification features and the second classification features, to obtain a classification result corresponding to the candidate answer information, the classification result indicating whether the candidate answer information is answer information of the question information.

[0317] The device provided in the embodiments of the present application introduces the evidence information corresponding to the question information in the process of determining the correctness of the candidate answer information, and takes the evidence information as the evidence proving whether the candidate answer information can be the answer information of the question information, so that the second classification feature obtained by adjusting the evidence feature based on the association degree between the first classification feature corresponding to the candidate answer information and the evidence feature can not only reflect the semantic information of the evidence information itself, but also reflect the support degree of the evidence information to the candidate answer information, so that when classification is performed based on the first classification feature and the second classification feature, the influence of the evidence information can be considered to obtain a more accurate classification result, and the accuracy of the determined answer information is improved.

[0318] Optionally, the device further comprises:

[0319] The similarity obtaining module is configured to obtain an answer similarity between each two of the candidate answer information.

[0320] The second feature adjusting module is configured to, for each candidate answer information, obtain a third classification feature based on a plurality of answer similarities corresponding to the candidate answer information and the first classification features corresponding to other candidate answer information in the plurality of candidate answer information except the candidate answer information.

[0321] The classification module 1604 is configured to, for each candidate answer information, perform classification based on the first classification feature, the second classification feature and the third classification feature corresponding to the candidate answer information to obtain the classification result.

[0322] Optionally, the classification module 1604 is configured to:

[0323] fuse the first classification feature and the third classification feature to obtain fused classification features;

[0324] splice the first classification feature, the third classification feature and the fused classification features to obtain fourth classification features of the candidate answer information;

[0325] perform classification based on the second classification feature and the fourth classification feature to obtain the classification result.

[0326] Optionally, the second feature adjusting module comprises:

[0327] The normalization unit is configured to normalize the plurality of answer similarities to obtain normalized plurality of answer similarities.

[0328] The adjusting unit is configured to perform weighted processing on the first classification features corresponding to other candidate answer information in the plurality of candidate answer information except the candidate answer information based on the normalized plurality of answer similarities to obtain the third classification features.

[0329] Optionally, the classification module 1604 is configured to:

[0330] weight the second classification feature according to a first confidence degree corresponding to the evidence information, to obtain a weighted second classification feature, the first confidence degree indicating a degree of credibility of the evidence information;

[0331] classify based on the first classification feature and the weighted second classification feature to obtain the classification result.

[0332] Optionally, the evidence information includes a plurality of evidence segments, and the apparatus further includes a confidence degree acquisition module configured to:

[0333] acquire a segment similarity between each two of the plurality of evidence segments;

[0334] for each evidence segment, determine a segment confidence degree corresponding to the evidence segment based on a plurality of segment similarities corresponding to the evidence segment;

[0335] splice the segment confidence degrees corresponding to the plurality of evidence segments to obtain the first confidence degree.

[0336] Optionally, the evidence information includes a plurality of evidence segments, and the second feature acquisition module 1602 is configured to:

[0337] acquire an evidence segment feature corresponding to each of the plurality of evidence segments;

[0338] adjust the plurality of evidence segment features based on evidence weights corresponding to the plurality of evidence segments, the evidence weights indicating degrees of importance of the evidence segments;

[0339] splice the adjusted plurality of evidence segment features to obtain the evidence feature.

[0340] Optionally, the second feature acquisition module 1602 is configured to:

[0341] acquire a question feature of the question information;

[0342] match the question feature and the evidence feature to obtain a second confidence degree corresponding to the evidence information, the second confidence degree indicating a degree of credibility of the evidence information as evidence of the question information;

[0343] adjust the evidence feature based on the second confidence degree.

[0344] Optionally, the classification module 1604 is configured to:

[0345] fuse the first classification feature and the second classification feature to obtain a fused classification feature;

[0346] The first classification feature, the second classification feature and the fused classification feature are spliced to obtain a fifth classification feature of the candidate answer information;

[0347] Classification is performed based on the fifth classification feature to obtain the classification result.

[0348] Optionally, the first feature acquisition module 1602 is configured to:

[0349] The question information and the candidate answer information are combined to obtain combined information;

[0350] The combined information is subjected to feature extraction to obtain the first classification feature.

[0351] Optionally, the apparatus further includes an answer acquisition module configured to:

[0352] Search is performed based on the question information and other question information having the same meaning as the question information;

[0353] The candidate answer information is extracted from the searched information.

[0354] Optionally, the information processing model includes a first processing sub-model, a second processing sub-model and a classification sub-model,

[0355] The first processing sub-model is configured to acquire the first classification feature based on the question information and the candidate answer information;

[0356] The second processing sub-model is configured to acquire an evidence feature of evidence information corresponding to the question information;

[0357] The second processing sub-model is further configured to adjust the evidence feature based on a correlation degree between the first classification feature and the evidence feature to obtain the second classification feature of the candidate answer information;

[0358] The classification sub-model is configured to perform classification based on the first classification feature and the second classification feature to obtain a classification result corresponding to the candidate answer information.

[0359] Optionally, the first attention network in the first processing sub-model is configured to acquire an answer similarity between each two of a plurality of the candidate answer information;

[0360] The first attention network is further configured to, for each candidate answer information, acquire a third classification feature based on a plurality of answer similarities corresponding to the candidate answer information and first classification features of other candidate answer information in the plurality of candidate answer information except the candidate answer information;

[0361] The classification sub-model is also used for classifying, for each candidate answer information, based on the first classification feature, the second classification feature and the third classification feature corresponding to the candidate answer information, to obtain the classification result.

[0362] Optionally, the fusion network in the classification sub-model is configured to perform weighted processing on the second classification feature according to a first confidence degree corresponding to the evidence information, to obtain a weighted second classification feature, the first confidence degree indicating a degree of credibility of the evidence information.

[0363] The classification network in the classification sub-model is configured to perform classification based on the first classification feature and the weighted second classification feature, to obtain the classification result.

[0364] Optionally, the evidence information includes a plurality of evidence segments, and the self-attention network in the second processing sub-model is configured to:

[0365] obtain a segment similarity between each two evidence segments in the plurality of evidence segments;

[0366] for each evidence segment, determine a segment confidence degree corresponding to the evidence segment based on a plurality of segment similarities corresponding to the evidence segment;

[0367] splice the segment confidence degrees corresponding to the plurality of evidence segments to obtain the first confidence degree.

[0368] Optionally, the first extraction network in the first processing sub-model is configured to obtain the first classification feature based on the question information and the candidate answer information.

[0369] The second extraction network in the second processing sub-model is configured to obtain an evidence feature of the evidence information corresponding to the question information.

[0370] The second attention network in the second processing sub-model is configured to adjust the evidence feature based on an association degree between the first classification feature and the evidence feature, to obtain the second classification feature of the candidate answer information.

[0371] Optionally, the evidence information includes a plurality of evidence segments,

[0372] The feature extraction layer in the second extraction network is configured to obtain evidence segment features corresponding to the plurality of evidence segments, respectively.

[0373] The attention layer in the second extraction network is configured to adjust a plurality of evidence segment features based on evidence weights corresponding to the plurality of evidence segments, the evidence weights indicating importance degrees of the evidence segments.

[0374] The attention layer is also configured to splice the adjusted plurality of evidence segment features to obtain the evidence feature.

[0375] Optionally, the second extraction network in the second processing sub-model is further configured to obtain a question feature of the question information.

[0376] The third attention network in the second processing sub-model is configured to match the question feature and the evidence feature to obtain a second confidence degree corresponding to the evidence information, the second confidence degree representing a credibility of the evidence information as the evidence of the question information.

[0377] The third attention network is further configured to adjust the evidence feature based on the second confidence degree.

[0378] Optionally, the training process of the information processing model comprises:

[0379] obtaining sample question information, sample candidate answer information, sample evidence information, and a sample classification result corresponding to the sample candidate answer information;

[0380] invoking the information processing model to process the sample question information, the sample candidate answer information, and the sample evidence information to obtain a predicted classification result corresponding to the sample candidate answer information;

[0381] training the information processing model based on the sample classification result and the predicted classification result.

[0382] All the optional technical solutions described above can be combined to form optional embodiments of the present application, which will not be described again.

[0383] It should be noted that the information processing apparatus provided in the above embodiments is only used as an example to illustrate the division of the above functional modules, and in actual applications, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the information processing apparatus and the information processing method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process is described in the method embodiments, which will not be described here.

[0384] The computer device provided in the embodiments of the present application includes a processor and a memory, and the memory stores at least one computer program, which is loaded and executed by the processor to implement the operations performed by the information processing method of the above embodiments.

[0385] Optionally, the computer device is provided as a terminal. Figure 17 Fig. 17 is a structural schematic diagram of a terminal 1700 provided in the embodiments of the present application. The terminal 1700 includes a processor 1701 and a memory 1702.

[0386] The processor 1701 can include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like. The processor 1701 can be implemented in at least one of a hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), a PLA (Programmable Logic Array). The processor 1701 can also include a main processor and a coprocessor, the main processor being a processor for processing data in an awake state, also referred to as a CPU (Central Processing Unit), and the coprocessor being a low-power processor for processing data in a standby state. In some embodiments, the processor 1701 can be integrated with a GPU (Graphics Processing Unit) for rendering and drawing content required to be displayed by the display screen. In some embodiments, the processor 1701 can further include an AI (Artificial Intelligence) processor for processing computing operations related to machine learning.

[0387] The memory 1702 can include one or more computer-readable storage media that can be non-transitory. The memory 1702 can also include high-speed random access memory and nonvolatile, computer-readable storage media such as one or more magnetic disk storage devices, flash memory devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 1702 is used to store at least one computer program for being executed by the processor 1701 to implement the information processing method provided by the method embodiments in the present application.

[0388] In some embodiments, the terminal 1700 can also optionally include a peripheral device interface 1703 and at least one peripheral device. The processor 1701, the memory 1702, and the peripheral device interface 1703 can be connected through a bus or a signal line. Each peripheral device can be connected to the peripheral device interface 1703 through a bus, a signal line, or a circuit board. Specifically, the peripheral device includes at least one of a radio frequency circuit 1704, a display screen 1705, a camera assembly 1706, an audio circuit 1707, a positioning assembly 1708, and a power supply 1709.

[0389] The peripheral interface 1703 can be used to connect at least one I / O (Input / Output) related peripheral device to the processor 1701 and the memory 1702. In some embodiments, the processor 1701, the memory 1702 and the peripheral interface 1703 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1701, the memory 1702 and the peripheral interface 1703 can be implemented on a separate chip or circuit board, and the present embodiments are not limited in this regard.

[0390] The radio frequency circuit 1704 is used to receive and send RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 1704 communicates with the communication network and other communication devices through electromagnetic signals. The radio frequency circuit 1704 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals into electrical signals. Optionally, the radio frequency circuit 1704 includes an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a subscriber identity module card, and the like. The radio frequency circuit 1704 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes but is not limited to: the World Wide Web, a metropolitan area network, an intranet, various generations of mobile communication networks (2G, 3G, 4G and 5G), a wireless local area network and / or a WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 1704 can also include NFC (Near Field Communication) related circuitry, and the present application is not limited in this regard.

[0391] The display screen 1705 is configured to display a UI (User Interface). The UI can include graphics, text, icons, video, and any combination thereof. When the display screen 1705 is a touch display screen, the display screen 1705 is further configured to capture touch signals on or above the surface of the display screen 1705. The touch signals can be input to the processor 1701 as control signals for processing. In this case, the display screen 1705 can also be configured to provide virtual buttons and / or virtual keyboard, also known as soft buttons and / or soft keyboard. In some embodiments, the display screen 1705 can be one, disposed on the front panel of the terminal 1700; in other embodiments, the display screen 1705 can be at least two, respectively disposed on different surfaces of the terminal 1700 or in a folding design; in other embodiments, the display screen 1705 can be a flexible display screen, disposed on a curved surface or a folding surface of the terminal 1700. Even, the display screen 1705 can also be disposed in an irregular shape other than a rectangle, i.e., a special-shaped screen. The display screen 1705 can be made of materials such as LCD (Liquid Crystal Display) or OLED (Organic Light-Emitting Diode).

[0392] The camera assembly 1706 is configured to capture images or videos. Optionally, the camera assembly 1706 includes a front-facing camera and a rear-facing camera. The front-facing camera is disposed on the front panel of the terminal, and the rear-facing camera is disposed on the back of the terminal. In some embodiments, the rear-facing camera is at least two, which is any one of a main camera, a depth-of-field camera, a wide-angle camera, and a telephoto camera, to realize the background blur function of the main camera and the depth-of-field camera, the panoramic shooting and VR (Virtual Reality) shooting function of the main camera and the wide-angle camera, or other fusion shooting functions. In some embodiments, the camera assembly 1706 can further include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. The dual-color temperature flash refers to the combination of a warm light flash and a cold light flash, which can be used for light compensation under different color temperatures.

[0393] The audio circuit 1707 can include a microphone and a speaker. The microphone is used to collect sound waves of a user and an environment, and convert the sound waves into an electrical signal input to the processor 1701 for processing, or input to the radio frequency circuit 1704 to realize voice communication. For the purpose of stereo sound collection or noise reduction, the microphone can be multiple, respectively arranged at different parts of the terminal 1700. The microphone can also be an array microphone or an omnidirectional collection type microphone. The speaker is used to convert an electrical signal from the processor 1701 or the radio frequency circuit 1704 into sound waves. The speaker can be a traditional diaphragm speaker, or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, not only can the electrical signal be converted into a sound wave audible to humans, but also can be converted into a sound wave inaudible to humans for ranging purposes. In some embodiments, the audio circuit 1707 can also include a headphone jack.

[0394] The positioning component 1708 is used to position the current geographic position of the terminal 1700 to realize navigation or LBS (Location Based Service). The positioning component 1708 can be a positioning component based on the GPS (Global Positioning System) of the United States, the Beidou system of China, the Glonass positioning system of Russia or the Galileo positioning system of the European Union.

[0395] The power supply 1709 is used to supply power to each component in the terminal 1700. The power supply 1709 can be alternating current, direct current, disposable battery or rechargeable battery. When the power supply 1709 includes a rechargeable battery, the rechargeable battery can be a wired charging battery or a wireless charging battery. The wired charging battery is a battery charged through a wired line, and the wireless charging battery is a battery charged through a wireless coil. The rechargeable battery can also be used to support fast charging technology.

[0396] Those skilled in the art can understand that, Figure 17 The structure shown in the figure does not constitute a limitation on the terminal 1700, and can include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0397] Optionally, the computer device is provided as a server. Figure 18is a structural schematic diagram of a server provided by an embodiment of the present application. The server 1800 can have great differences due to different configurations or performances, and can include one or more processors (Central Processing Units, CPUs) 1801 and one or more memories 1802. The memory 1802 stores at least one computer program, which is loaded and executed by the processor 1801 to implement the method provided by each method embodiment described above. Of course, the server can also have a wired or wireless network interface, a keyboard, an input and output interface, and other components for implementing device functions, and the like, so as to perform input and output. The server can also include other components for implementing device functions, which are not described here.

[0398] An embodiment of the present application further provides a computer readable storage medium, which stores at least one computer program. The at least one computer program is loaded and executed by a processor to implement the operations performed by the information processing method of the above embodiment.

[0399] An embodiment of the present application further provides a computer program product, which includes a computer program. The computer program is executed by a processor to implement the operations performed by the information processing method of the above embodiment.

[0400] In some embodiments, the computer program related to the embodiments of the present application can be deployed on a computer device for execution, or on multiple computer devices located in one place, or on multiple computer devices distributed in multiple places and interconnected through a communication network. The multiple computer devices distributed in multiple places and interconnected through a communication network can constitute a blockchain system.

[0401] Those skilled in the art can understand that all or part of the steps of the above embodiments can be completed by hardware, or by a program instructing relevant hardware, which can be stored in a computer readable storage medium. The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0402] The above is only an optional embodiment of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An information processing method characterized by comprising: The method comprises: obtaining first classification features based on question information and candidate answer information, wherein the first classification features at least represent semantic information of the candidate answer information; obtaining evidence features of evidence information corresponding to the question information, wherein the evidence information is used to provide evidence that the candidate answer information is answer information of the question information; adjusting the evidence features based on the correlation between the first classification features and the evidence features to obtain second classification features; performing classification based on the first classification features and the second classification features to obtain a classification result corresponding to the candidate answer information, wherein the classification result indicates whether the candidate answer information is answer information of the question information.

2. The method of claim 1, wherein, After the first classification features are obtained based on the question information and the candidate answer information, the method further comprises: obtaining answer similarities between each two of a plurality of candidate answer information; for each candidate answer information, obtaining third classification features based on a plurality of answer similarities corresponding to the candidate answer information and first classification features corresponding to other candidate answer information in the plurality of candidate answer information except the candidate answer information; the classification based on the first classification features and the second classification features to obtain the classification result corresponding to the candidate answer information comprises: for each candidate answer information, performing classification based on the first classification features, the second classification features and the third classification features corresponding to the candidate answer information to obtain the classification result.

3. The method of claim 2, wherein, the classification based on the first classification features, the second classification features and the third classification features corresponding to the candidate answer information to obtain the classification result comprises: fusing the first classification features and the third classification features to obtain fused classification features; splicing the first classification features, the third classification features and the fused classification features to obtain fourth classification features of the candidate answer information; performing classification based on the second classification features and the fourth classification features to obtain the classification result.

4. The method of claim 2, wherein, the third classification features obtained based on the plurality of answer similarities corresponding to the candidate answer information and the first classification features corresponding to other candidate answer information in the plurality of candidate answer information except the candidate answer information comprises: normalizing the plurality of answer similarities to obtain normalized plurality of answer similarities; performing weighted processing on the first classification features corresponding to other candidate answer information in the plurality of candidate answer information except the candidate answer information based on the normalized plurality of answer similarities to obtain the third classification features.

5. The method of claim 1, wherein, the classification based on the first classification features and the second classification features corresponding to the candidate answer information to obtain the classification result comprises: performing weighted processing on the second classification features based on a first confidence corresponding to the evidence information to obtain weighted second classification features, wherein the first confidence refers to a credibility of the evidence information; performing classification based on the first classification features and the weighted second classification features to obtain the classification result.

6. The method of claim 5, wherein, The evidence information includes a plurality of evidence segments, and before the second classification feature is obtained by weighting the second classification feature according to the first confidence degree corresponding to the evidence information, the method further includes: Obtain the segment similarity between each two evidence segments in the plurality of evidence segments; For each evidence segment, determine the segment confidence degree corresponding to the evidence segment based on the plurality of segment similarities corresponding to the evidence segment; Splice the segment confidence degrees corresponding to the plurality of evidence segments to obtain the first confidence degree.

7. The method of claim 1, wherein, The evidence information includes a plurality of evidence segments, and the evidence feature of the evidence information corresponding to the problem information is obtained, including: Respectively obtain the evidence segment feature corresponding to each of the plurality of evidence segments; Respectively adjust the plurality of evidence segment features based on the evidence weight corresponding to the plurality of evidence segments, the evidence weight indicating the importance of the evidence segment; Splice the plurality of adjusted evidence segment features to obtain the evidence feature.

8. The method of claim 1, wherein, Before the second classification feature is obtained by adjusting the evidence feature based on the correlation between the first classification feature and the evidence feature, the method further includes: Obtain the problem feature of the problem information; Match the problem feature and the evidence feature to obtain the second confidence degree corresponding to the evidence information, the second confidence degree indicating the credibility of the evidence information as the evidence of the problem information; Adjust the evidence feature based on the second confidence degree.

9. The method of claim 1, wherein, The classification result corresponding to the candidate answer information is obtained by classifying based on the first classification feature and the second classification feature, including: Fuse the first classification feature and the second classification feature to obtain a fused classification feature; Splice the first classification feature, the second classification feature, and the fused classification feature to obtain a fifth classification feature of the candidate answer information; Classify based on the fifth classification feature to obtain the classification result.

10. The method according to any one of claims 1 to 9, characterized in that, The first classification feature is obtained based on the problem information and the candidate answer information, including: Combine the problem information and the candidate answer information to obtain combined information; Extract the first classification feature from the combined information.

11. The method according to any one of claims 1 to 9, characterized in that, Before the first classification feature is obtained based on the problem information and the candidate answer information, the method further includes: Search based on the problem information and other problem information with the same meaning as the problem information; Extract the candidate answer information from the searched information.

12. The method of claim 1, wherein, The information processing model includes a first processing sub-model, a second processing sub-model, and a classification sub-model, The first processing sub-model is configured to obtain the first classification feature based on the problem information and the candidate answer information; The second processing sub-model is configured to obtain the evidence feature of the evidence information corresponding to the problem information; The second processing sub-model is further configured to adjust the evidence feature based on the correlation between the first classification feature and the evidence feature to obtain the second classification feature of the candidate answer information; The classification sub-model is configured to perform classification based on the first classification feature and the second classification feature to obtain a classification result corresponding to the candidate answer information.

13. The method of claim 12, wherein, The first attention network in the first processing sub-model is configured to obtain an answer similarity between each two of the candidate answer information. The first attention network is further configured to, for each candidate answer information, obtain a third classification feature based on a plurality of answer similarities corresponding to the candidate answer information and first classification features corresponding to other candidate answer information except the candidate answer information in the plurality of candidate answer information. The classification sub-model is further configured to, for each candidate answer information, perform classification based on the first classification feature, the second classification feature and the third classification feature corresponding to the candidate answer information to obtain the classification result.

14. The method of claim 12, wherein, The fusion network in the classification sub-model is configured to perform weighted processing on the second classification feature according to a first confidence degree corresponding to the evidence information, to obtain a weighted second classification feature, the first confidence degree indicating a credibility of the evidence information. The classification network in the classification sub-model is configured to perform classification based on the first classification feature and the weighted second classification feature to obtain the classification result.

15. The method of claim 14, wherein, The evidence information includes a plurality of evidence segments, and the self-attention network in the second processing sub-model is configured to: obtain a segment similarity between each two of the plurality of evidence segments; for each evidence segment, determine a segment confidence degree corresponding to the evidence segment based on a plurality of segment similarities corresponding to the evidence segment; and splice segment confidence degrees corresponding to the plurality of evidence segments to obtain the first confidence degree.

16. The method of claim 12, wherein, The first extraction network in the first processing sub-model is configured to obtain the first classification feature based on the question information and the candidate answer information. The second extraction network in the second processing sub-model is configured to obtain an evidence feature of evidence information corresponding to the question information. The second attention network in the second processing sub-model is configured to adjust the evidence feature based on an association degree between the first classification feature and the evidence feature to obtain the second classification feature of the candidate answer information.

17. An information processing apparatus comprising: The apparatus comprises: A first feature obtaining module configured to obtain a first classification feature based on question information and candidate answer information, the first classification feature at least representing semantic information of the candidate answer information; A second feature obtaining module configured to obtain an evidence feature of evidence information corresponding to the question information, the evidence information configured to provide evidence that the candidate answer information is answer information of the question information; A first feature adjusting module configured to adjust the evidence feature based on an association degree between the first classification feature and the evidence feature to obtain a second classification feature. The classification module is configured to perform classification based on the first classification feature and the second classification feature to obtain a classification result corresponding to the candidate answer information, the classification result indicating whether the candidate answer information is answer information of the question information.

18. A computer device, comprising: The computer device comprises a processor and a memory, and the memory stores at least one computer program, which is loaded and executed by the processor to implement the operations performed by the information processing method according to any one of claims 1 to 16.

19. A computer-readable storage medium, characterized in that, The computer readable storage medium stores at least one computer program, which is loaded and executed by the processor to implement the operations performed by the information processing method according to any one of claims 1 to 16.

20. A computer program product comprising a computer program, characterized in that, The computer program, when executed by the processor, implements the operations performed by the information processing method according to any one of claims 1 to 16.

Images