Information processing method, model training method, device, equipment and medium

Abstract

The present disclosure provides an information processing method, a training method for a deep learning model, an apparatus, an electronic device, a storage medium, and a program product, which relate to the field of artificial intelligence technology, and in particular to the technical fields of large models, large language models, Transformers, conversational models, generative models, etc. The specific implementation scheme is: using the result generation model to process input information, obtain intermediate features and initial output results for responding to the input information, the result generation model is integrated with a target function plug-in, and the target function plug-in is used to evaluate the initial output result; using the target function plug-in to process the intermediate features, obtain an evaluation result for evaluating the initial output result; and adjusting the initial output result according to the evaluation result to obtain the target output result.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, in particular to the technical fields of large models, large language models, Transformers (codecs), conversational models, generative models, etc. Specifically, it relates to information processing methods, training methods, devices, electronic devices, storage media, and program products for deep learning models. Background Art

[0002] Human-computer interaction is a way for humans to interact with machines. With the continuous development of artificial intelligence technology, machines have been able to understand the information input by humans, understand the inherent meaning of the input information, and make corresponding feedback. In these operations, accurate understanding of semantics, rapid feedback, and giving corresponding opinions or suggestions have become factors that affect the smoothness of human-computer interaction. Summary of the invention

[0003] The present disclosure provides an information processing method, a training method for a deep learning model, an apparatus, an electronic device, a storage medium, and a program product.

[0004] According to one aspect of the present disclosure, there is provided an information processing method, comprising: using a result generation model to process input information to obtain intermediate features and an initial output result for responding to the input information, wherein a target function plug-in is integrated in the result generation model, and the target function plug-in is used to evaluate the initial output result; using the target function plug-in to process the intermediate features to obtain an evaluation result for evaluating the initial output result; and adjusting the initial output result according to the evaluation result to obtain a target output result.

[0005] According to another aspect of the present disclosure, a method for training a deep learning model is provided, comprising: using a result generation model to process sample input information to obtain sample intermediate features and sample initial output results for responding to the sample input information, wherein a target function plug-in is integrated in the result generation model, and the target function plug-in is used to evaluate the sample initial output results; using the target function plug-in to process the sample intermediate features to obtain a sample evaluation result for evaluating the sample initial output result; adjusting the sample initial output result according to the sample evaluation result to obtain a sample target output result; and training the deep learning model based on the sample target output result and an output label matching the sample input information to obtain a trained deep learning model, wherein the output label is used to characterize a labeling result matching the sample input information.

[0006] According to another aspect of the present disclosure, there is provided an information processing device, comprising: a result processing module, used to process input information using a result generation model to obtain intermediate features and an initial output result for responding to the input information, wherein a target function plug-in is integrated in the result generation model, and the target function plug-in is used to evaluate the initial output result; an evaluation module, used to process the intermediate features using the target function plug-in to obtain an evaluation result for evaluating the initial output result; and an adjustment module, used to adjust the initial output result according to the evaluation result to obtain a target output result.

[0007] According to another aspect of the present disclosure, a training device for a deep learning model is provided, comprising: a sample processing module, used to process sample input information using a result generation model to obtain sample intermediate features and sample initial output results for responding to the sample input information, wherein the result generation model is integrated with a target function plug-in, and the target function plug-in is used to evaluate the sample initial output results; a sample evaluation module, used to process the sample intermediate features using the target function plug-in to obtain a sample evaluation result for evaluating the sample initial output results; a sample adjustment module, used to adjust the sample initial output result according to the sample evaluation result to obtain a sample target output result; and a training module, used to train the deep learning model based on the sample target output result and an output label matching the sample input information to obtain a trained deep learning model, wherein the output label is used to characterize a labeling result matching the sample input information.

[0008] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a method as disclosed in the present disclosure.

[0009] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the method of the present disclosure.

[0010] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the method of the present disclosure is implemented.

[0011] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0013] Figure 1 An exemplary system architecture to which the information processing method and device according to an embodiment of the present disclosure can be applied is schematically shown;

[0014] Figure 2 The flowchart of the information processing method according to the embodiment of the present disclosure is schematically shown;

[0015] Figure 3A A schematic diagram schematically shows an information processing method according to an embodiment of the present disclosure;

[0016] Figure 3B A schematic diagram schematically illustrates an information processing method according to a related example;

[0017] Figure 3C A schematic diagram schematically shows an information processing method according to another related example;

[0018] Figure 4 A schematic diagram schematically shows a method for determining an initial output result according to an embodiment of the present disclosure;

[0019] Figure 5 A schematic diagram schematically shows an information processing method according to another embodiment of the present disclosure;

[0020] Figure 6 A flowchart of a deep learning model training method according to an embodiment of the present disclosure is schematically shown;

[0021] Figure 7 A schematic diagram of a mask template according to an embodiment of the present disclosure is schematically shown;

[0022] Figure 8 A block diagram schematically shows an information processing device according to an embodiment of the present disclosure;

[0023] Fig. 9 A block diagram schematically shows a training device for a deep learning model according to an embodiment of the present disclosure; and

[0024] Fig.10 A block diagram of an electronic device suitable for implementing an information processing method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0025] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0026] The present disclosure provides an information processing method, a training method for a deep learning model, an apparatus, an electronic device, a storage medium, and a program product.

[0027] According to one aspect of the present disclosure, there is provided an information processing method, comprising: using a result generation model to process input information to obtain intermediate features and an initial output result for responding to the input information, wherein a target function plug-in is integrated in the result generation model, and the target function plug-in is used to evaluate the initial output result; using the target function plug-in to process the intermediate features to obtain an evaluation result for evaluating the initial output result; and adjusting the initial output result according to the evaluation result to obtain a target output result.

[0028] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision, disclosure and application of user personal information involved comply with the provisions of relevant laws and regulations, take necessary confidentiality measures, and do not violate public order and good morals.

[0029] In the technical solution of the present disclosure, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0030] Figure 1 An exemplary system architecture to which the information processing method and apparatus according to an embodiment of the present disclosure can be applied is schematically shown.

[0031] It should be noted that Figure 1 The examples shown are only examples of system architectures to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but do not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios. For example, in another embodiment, the exemplary system architecture to which the information processing method and apparatus can be applied may include a terminal device, but the terminal device may implement the information processing method and apparatus provided by the embodiments of the present disclosure without interacting with the server.

[0032] like Figure 1As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0033] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as knowledge reading applications, web browser applications, search applications, instant messaging tools, email clients and / or social platform software, etc. (only as examples).

[0034] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0035] The server 105 may be a server that provides various services, such as a background management server (only an example) that provides support for the content browsed by the user using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0036] It should be noted that the information processing method provided in the embodiment of the present disclosure can generally be executed by the terminal device 101, 102, or 103. Accordingly, the information processing apparatus provided in the embodiment of the present disclosure can also be set in the terminal device 101, 102, or 103.

[0037] Alternatively, the information processing method provided in the embodiment of the present disclosure may also be generally executed by the server 105. Accordingly, the information processing apparatus provided in the embodiment of the present disclosure may generally be arranged in the server 105. The information processing method provided in the embodiment of the present disclosure may also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the information processing apparatus provided in the embodiment of the present disclosure may also be arranged in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0038] For example, the terminal devices 101, 102, and 103 can obtain the input information entered by the user into the text box, and then send the obtained input information to the server 105, and the server 105 uses the result generation model to process the input information to obtain intermediate features and initial output results for responding to the input information. The intermediate features are processed using the target function plug-in to obtain an evaluation result for evaluating the initial output result. The initial output result is adjusted according to the evaluation result to obtain the target output result. Alternatively, a server or server cluster that can communicate with the terminal devices 101, 102, 103 and / or the server 105 analyzes the input information and obtains the target output result. The server 105 can send the target output result to the terminal devices 101, 102, and 103, and the terminal devices 101, 102, and 103 feed back to the user.

[0039] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0040] It should be noted that the sequence numbers of the operations in the following method are only used as representations of the operations for the purpose of description, and should not be regarded as representing the execution order of the operations. Unless explicitly stated, the method does not need to be executed completely in the order shown.

[0041] Figure 2 The flowchart of the information processing method according to the embodiment of the present disclosure is schematically shown.

[0042] like Figure 2 As shown, the method includes operations S210 to S230.

[0043] In operation S210, the input information is processed using the result generation model to obtain intermediate features and an initial output result for responding to the input information.

[0044] In operation S220, the intermediate features are processed using the target function plug-in to obtain an evaluation result for evaluating the initial output result.

[0045] In operation S230, the initial output result is adjusted according to the evaluation result to obtain a target output result.

[0046] In one example, the result generation model may include a deep learning model, such as one or more of a convolutional neural network, a recurrent neural network, or a long short-term memory network, but is not limited thereto, and may also include a large language model (LLM), such as one or more of GPT (Generative Pre-Trained Transformer), ChatGPT (Chat Generative Pre-Trained Transformer), GLM (General Language Model), etc.

[0047] In one example, a target function plug-in may be integrated into the result generation model. The target function plug-in may include a multi-layer perceptron (MLP). However, it is not limited to this. The target function plug-in may also include a combination of other deep learning models and multi-layer perceptrons in the deep learning model except for the multi-layer perceptron. As long as it can be used to evaluate the match between the initial output result and the input information.

[0048] In one example, the target function plug-in can be used as a functional component of the result generation model. The target function plug-in can support multiple model interfaces.

[0049] In one example, adjusting the initial output result according to the evaluation result to obtain the target output result can be understood as optimizing the initial output result based on the evaluation result to obtain the target output result. The adjustment may include but is not limited to screening, modifying, updating, sorting, etc.

[0050] In one example, the intermediate features are obtained in the process of processing the input information using the result generation model. Since the intermediate features are generated based on the input information and obtained in the process of generating the initial output results, the intermediate features contain information related to the output information and the initial output results, and thus the initial output results are evaluated using the intermediate features, which can accurately control the compatibility between the input information and the target output results.

[0051] According to the embodiments of the present disclosure, since the target function plug-in is integrated into the result generation model, the initial output result output by the result generation model can be evaluated using the target function plug-in, and then the evaluation result output by the target function plug-in for evaluating the initial output result can be used to adjust the initial output result to improve the adaptability between the target output result and the input information, thereby improving the user experience in the human-computer interaction scenario.

[0052] Figure 3AThe figure schematically shows a method for processing information according to an embodiment of the present disclosure.

[0053] like Figure 3A As shown, input information 310, such as S1, S2, S3, ..., SN, can be input into a result generation model M310, such as LLM, to obtain an initial output result 320, such as T1, T2, T3, ..., TM. The intermediate feature 330, such as F, obtained in the process of processing the input information using the result generation model is input into the target function plug-in M320 to output an evaluation result 340. The initial output result 320 is adjusted according to the evaluation result 340 to obtain a target output result 350.

[0054] Figure 3B A schematic diagram schematically shows an information processing method according to a related example.

[0055] like Figure 3B As shown, the input information 310 can be input into the result generation model M310 to obtain the initial output result 320. The initial output result 320 is subjected to word segmentation and feature extraction in sequence to obtain output result processing features. The output result processing features are input into the adjustment model M330 to output the target output result 360.

[0056] Figure 3C A schematic diagram schematically shows an information processing method according to another related example.

[0057] like Figure 3C As shown, the input information 310 can be input into the result generation model M310 to obtain the initial output result 320. The initial output result 320 is subjected to word segmentation and feature extraction in sequence to obtain output result processing features. The output result processing features are input into the evaluation model M340 to output the evaluation result 370. According to the evaluation result 370, the initial output result 320 is adjusted to obtain the target output result 380.

[0058] Compared with the information processing method using the adjustment model, the information processing method using the target function plug-in can simplify the intermediate processing process because the input data input into the target function plug-in is the intermediate features obtained in the process of processing the input information using the result generation model, avoid word segmentation, feature extraction and other processing on the initial output results, and move computing power and storage resources to the result generation model, thereby improving processing efficiency.

[0059] Compared with the information processing method using the evaluation model, the information processing method using the target function plug-in can simplify the network structure of the target function plug-in because the input data of the target function plug-in is the evaluation result used to evaluate the initial output result rather than the target output result, and avoid the target function plug-in learning a large amount of input information and relevant knowledge of the initial output result, thereby improving processing efficiency while reducing the performance requirements for hardware equipment.

[0060] According to an embodiment of the present disclosure, the result generation model includes a decoder and an output layer, and the initial output result may include output sub-results.

[0061] According to an embodiment of the present disclosure, when executing Figure 2 The operation S210 shown, using the result generation model to process the input information to obtain the intermediate features may include the operations of: using a decoder to process the encoded features and the output sub-results to obtain the current output sub-result features. Using the output layer to process the current output sub-result features to obtain the current output sub-result. When it is determined that the current output sub-result meets the predetermined termination condition, based on the current output sub-result features, the intermediate features are obtained.

[0062] In one example, the encoding feature can be obtained by processing the input information using the result generation model. The input information can be preprocessed, such as word segmentation, to obtain a word sequence consisting of multiple word segments. The word segments in the word sequence are read as an embedding vector sequence according to the vocabulary. However, it is not limited to this. The word sequence can also be input into the embedding layer to obtain an embedding vector sequence. The embedding vector sequence is input into the encoder to obtain the encoding feature.

[0063] In one example, the decoder Decoder may refer to a decoding layer in a codec (Transformer), but is not limited thereto, and may also refer to a decoding layer in a large model built based on Transformer.

[0064] In one example, the output sub-result may be the previous output sub-result of the current output sub-result. For example, if the current output sub-result is Token M, then the output sub-result is Token M-1. However, this is not limited to this. The output sub-result may also be the previous output sub-result of the current output sub-result. For example, if the current output sub-result is Token M, then the output sub-results are Token 1 to Token M-1. As long as the result that has been output is used as the output sub-result.

[0065] The encoding feature and the output sub-result may be input into the decoder, the decoding feature may be output, and the decoding feature may be used as the feature of the current output sub-result. However, this is not limited thereto. The processed feature obtained by processing the decoding feature using the network layer structure between the decoder and the output layer in the large model may also be used as the feature of the current output sub-result. As long as the feature of the current output sub-result can be input into the output layer, the feature of the current output sub-result may be obtained.

[0066] In one example, based on the current output sub-result feature, obtaining the intermediate feature may refer to: using the current output sub-result feature as the intermediate feature, but is not limited thereto, and may also refer to: fusing the current output sub-result feature, the encoding feature, and the output sub-result feature to obtain the intermediate feature. Any intermediate feature that can cover the encoding feature and the feature that matches the initial output result will suffice.

[0067] In one example, the output layer may refer to a Liner layer (linear layer), but is not limited thereto, as long as it is a functional layer that can process the features of the current output sub-result and obtain the current output sub-result.

[0068] According to an embodiment of the present disclosure, when it is determined that the result type of the current output sub-result satisfies a predetermined result type, it may be determined that the current output sub-result satisfies a predetermined termination condition.

[0069] In one example, the predetermined result type may refer to a result type for indicating termination. The current output sub-result may refer to a current token or a current character. The feature corresponding to the current output sub-result of the result type for indicating termination may be used as an intermediate feature, but is not limited thereto, and the feature corresponding to any current output sub-result may also be used as an intermediate feature.

[0070] By using the principle of processing data of the decoder, the intermediate feature integrates the coding feature corresponding to the input information and the feature corresponding to the output sub-result. The intermediate feature integrating the coding feature corresponding to the input information and the feature corresponding to the output sub-result is input into the target function plug-in, so that the target function plug-in can learn the relevant knowledge of the input information and the relevant knowledge of the initial output result at the same time, enrich the data volume and data type of the input data of the target function plug-in, thereby improving the accuracy of the evaluation result output by the target function plug-in.

[0071] According to an embodiment of the present disclosure, the result generation model includes an encoder, a decoder, and an output layer.

[0072] According to the embodiments of the present disclosure, Figure 2The operation S210 shown, using the result generation model to process the input information to obtain an initial output result for responding to the input information, may include: using an encoder to process the input information to obtain a coding feature. Using a decoder to process the coding feature and the output sub-result to obtain a current output sub-result feature. Using an output layer to process the current output sub-result feature to obtain the current output sub-result. When it is determined that the current output sub-result satisfies a predetermined termination condition, obtaining an initial output result based on the current output sub-result and the output sub-result.

[0073] In one example, the encoder may refer to an encoding layer in a Transformer, but is not limited thereto, and may also refer to an encoding layer in a large model built based on the Transformer.

[0074] In one example, obtaining the initial output result based on the current output sub-result and the output sub-result may include: concatenating the current output sub-result and the output sub-result to obtain the initial output result. However, the method is not limited thereto. The method may also include: concatenating the current output sub-result, the output sub-result, and a predetermined character to obtain the initial output result.

[0075] According to an embodiment of the present disclosure, a beam search strategy is adopted for decoding, so that the current output sub-result of the initial output result is generated by decoding based on the encoding features and the features corresponding to the output sub-results, thereby improving the closeness of the contextual association relationship between multiple tokens in the initial output result and improving the semantic fluency and coherence.

[0076] Figure 4 The figure schematically shows a schematic diagram of determining an initial output result according to an embodiment of the present disclosure.

[0077] like Figure 4 As shown, the input information 410 is input into the encoder M411, and the encoding feature 420 is output. The encoding feature 420 and the previous output sub-result 431, such as TM-1, are input into the decoder M412, and the current output sub-result feature 440 is output. The current output sub-result feature 440 is input into the output layer M413 to obtain the current output sub-result 432. When it is determined that the current output sub-result 432 satisfies the predetermined termination condition, the current output sub-result TM and the multiple output sub-results, such as T1, T2, T3, and TM-1, are used as the initial output result.

[0078] By using the information processing method provided by the embodiment of the present disclosure, it is possible to use the beam search strategy of the decoder for decoding and increase the search range. In addition, the target function plug-in can be used to adjust the initial output result, further optimize the initial output result, reduce the influence of the generation probability in the decoder on the distribution of training data, and improve the controllability of the target output result. Thus, by using the concept of "task as plug-in", the discrimination task is integrated into the result generation model as a functional plug-in. After the model outputs the initial output result, the target function plug-in is used to simultaneously output the evaluation result of the matching degree between the input information and the initial output result, thereby streamlining the process, communication and computing overhead while improving the user experience.

[0079] According to an embodiment of the present disclosure, the information type of the input information may include a question type. The question type may refer to a type related to consulting and help-seeking in a question-and-answer scenario.

[0080] According to the embodiments of the present disclosure, Figure 2 The operation S230 shown, adjusting the initial output result according to the evaluation result to obtain the target output result, may include the following operations: when it is determined that the evaluation result is used to characterize that the initial output result and the input information meet a predetermined matching condition, updating the initial output result with a predetermined output result matching the question type to obtain the target output result.

[0081] In one example, the evaluation result may refer to a label indicating whether the initial output result matches the input information, for example, the evaluation result may refer to a degree of matching.

[0082] In one example, the predetermined matching condition may refer to that the matching degree between the initial output result and the input information does not satisfy the matching degree threshold. When the matching degree between the initial output result and the input information is less than or equal to the matching degree threshold, it is determined that the initial output result and the input information satisfy the predetermined matching condition. Otherwise, it is determined that the initial output result and the input information do not satisfy the predetermined matching condition.

[0083] In one example, when it is determined that the initial output result and the input information do not satisfy a predetermined matching condition, the target output result may be obtained based on the initial output result. For example, the initial output result is used as the target output result.

[0084] In one example, using a predetermined output result that matches the question type to update the initial output result to obtain a target output result may include: using the predetermined output result that matches the question type as the target output result.

[0085] For example, if it is determined that the initial output result and the input information meet the predetermined matching condition, then it is determined that the initial output result and the input information are not compatible, and directly outputting the initial output result as the target output result will cause a decrease in user experience. Pre-configured predetermined output results that match the question type, such as "I don't understand your question, please describe it again, thank you", can be used as the target output result to improve the smoothness and rationality of the interaction in the human-computer interaction scenario.

[0086] According to an embodiment of the present disclosure, the information type of the input information may include a search type. The initial output result may include multiple output sub-results.

[0087] According to the embodiments of the present disclosure, Figure 2 The operation S230 shown, adjusting the initial output result according to the evaluation result to obtain the target output result, may include: determining the sorting results of the multiple output sub-results based on the evaluation result, and sorting the multiple output sub-results based on the sorting results to obtain the target output result.

[0088] According to an embodiment of the present disclosure, the evaluation result may include multiple evaluation sub-results corresponding to the multiple output sub-results one by one, and each evaluation sub-result represents the matching degree between the corresponding output sub-result and the input information. The sorting results of the multiple output sub-results may be determined based on the multiple evaluation sub-results. For example, the sorting results of the multiple output sub-results may be determined according to the matching degree of the multiple evaluation sub-results. The multiple output sub-results may also be sorted or screened based on the sorting results to obtain the target output result.

[0089] For example, based on the sorting result, output sub-result 1, output sub-result 2, and output sub-result 3 are sorted to obtain the sorting results of output sub-result 2, output sub-result 3, and output sub-result 1. Based on the sorting result, output sub-result 2, output sub-result 3, and output sub-result 1 are used as target output results and fed back to the user.

[0090] For another example, based on the sorting result and the evaluation result, multiple output sub-results are screened, output sub-results below the matching degree threshold are deleted, and the screened multiple output sub-results are sorted to obtain output sub-result 2 and output sub-result 3. Output sub-result 2 and output sub-result 3 are used as target output results and fed back to the user.

[0091] According to the embodiments of the present disclosure, by using the evaluation results to sort multiple output sub-results, in a retrieval scenario or a recommendation scenario, the output sub-results that match the user's input information can be sorted first, and the output sub-results that match the user's input information can be viewed by the user first, thereby improving the user experience, improving the user's viewing efficiency, and saving the user's time.

[0092] According to another embodiment of the present disclosure, Figure 2 The operation S230 shown, adjusting the initial output result according to the evaluation result to obtain the target output result, may include: determining the sorting results of the multiple output sub-results based on the evaluation result. Determining multiple target feedback data matching the multiple output sub-results from a predetermined feedback database. Sorting the multiple target feedback data based on the sorting result to obtain the target output result.

[0093] In one example, the output sub-results may be used to characterize keywords or tags that match the input information. The feedback data in the predetermined feedback database may be pre-collected data. The feedback data may include a web page containing a link to a URL (Uniform Resource Locator), but is not limited thereto, and may also include a dynamic image file, a voice file, or other feedback data. An association relationship between the output sub-results and the feedback data in the predetermined feedback database may be established in advance. In the case of determining multiple output sub-results, multiple target feedback data that match the multiple output sub-results are determined from the predetermined feedback database based on the association relationship.

[0094] In an example, the ranking results of the plurality of output sub-results may be the same as or similar to the ranking results of the plurality of target feedback data.

[0095] For example, output sub-result 1 matches feedback data A and B, output sub-result 2 matches feedback data C, and output sub-result 3 matches feedback data D and E. Based on output sub-results 1, 2, 3, target feedback data A, B, C, D, E can be determined from a predetermined feedback database.

[0096] The sorting results of output sub-results 1, 2, and 3 include: output sub-result 2, output sub-result 3, and output sub-result 1. Then the sorting results of multiple target feedback data may include: target feedback data C, target feedback data D, target feedback data E, target feedback data A, and target feedback data B.

[0097] According to an embodiment of the present disclosure, the initial output result is used as an intermediate result, and the target feedback data is determined from a predetermined feedback database using the intermediate result. The target feedback data can be used to provide feedback to the user with results that match the input information, which can improve the richness and usefulness of the target output result, thereby improving user stickiness.

[0098] According to an embodiment of the present disclosure, multiple function plug-ins may be configured. The multiple function plug-ins are matched with multiple information types. The information type may include at least one of the following: question type, retrieval type, recommendation type, translation type. A target function plug-in may be determined from the multiple function plug-ins according to the information type of the input information.

[0099] For example, sample input information of a question type and a label matching the sample input information are used as training samples to train a function plug-in matching the question type, so that the function plug-in has the ability to evaluate an initial output result for responding to the input information of the question type.

[0100] Figure 5 The figure schematically shows an information processing method according to another embodiment of the present disclosure.

[0101] like Figure 5 As shown, the input information 510 can be input into the result generation model M510 to obtain the intermediate features 520 and the initial output result 530. The information type of the input information is determined. Based on the information type of the input information, a target function plug-in M520 is determined from a plurality of function plug-ins. The intermediate features 520 are input into the target function plug-in M520 to obtain an evaluation result 540. The initial output result 530 is adjusted according to the evaluation result 540 to obtain a target output result 550.

[0102] According to an embodiment of the present disclosure, a target functional plug-in may be determined from multiple functional plug-ins by the information type of the input information. However, this is not limited thereto. A target functional plug-in may also be determined from multiple functional plug-ins by the intention recognition result. For example, when executing Figure 2 Before the operation S220 shown, the result generation method may further include: performing intent recognition on the intermediate features to obtain an intent recognition result, and determining a target function plug-in that matches the intent recognition result from the multiple function plug-ins.

[0103] In one example, the intent recognition result can characterize the user's intention or need. For example, if the user wants to know the calories of a certain food, the input information includes "What are the calories of **food?" The intermediate features can be subjected to intent recognition to obtain an intent recognition result, and the information type of the intent recognition result used to characterize the user's input information is a question-and-answer type, and a target function plug-in that matches the intent recognition result can be determined from multiple function plug-ins. However, it is not limited to this. It is also possible to pre-establish an association relationship between a function plug-in and an intent recognition result, and based on the association relationship, determine a target function plug-in that matches the intent recognition result from multiple function plug-ins.

[0104] According to the embodiments of the present disclosure, multiple functional plug-ins are used to adapt to different application scenarios and to adapt to the user intent recognition results, which can improve the intelligence and pertinence of human-computer interaction and avoid the problem of poor pertinence caused by the use of general functional plug-ins.

[0105] According to an embodiment of the present disclosure, based on the operation instruction, when executing Figure 2Before the operation S220 shown, the result generating method may further include: based on the operation instruction, determining a target function plug-in matching the operation instruction from a plurality of function plug-ins.

[0106] In one example, an operation instruction may refer to an instruction generated in a human-computer interaction interface based on a user's operation on a target interface control.

[0107] For example, the operation instruction may refer to an operation instruction generated after a user inputs input information into an input box, such as a text box, of a human-computer interaction interface and then clicks on a target interface control, such as a "query" control. However, this is not limited to this. Any operation instruction that can be associated with the input information and can represent the user's intention will suffice.

[0108] According to the embodiments of the present disclosure, the user intention can be quickly learned by utilizing the operation instructions, thereby improving the processing efficiency and the accuracy of determining the target function plug-in.

[0109] According to an embodiment of the present disclosure, when executing Figure 2 Before the operation S210 shown, the result generating method may further include: obtaining input information from the input box in response to an operation instruction for the target interface control.

[0110] In response to the operation instruction for the target interface space, the input information can be obtained from the input box to achieve a fully automated effect. The input information in the input box can be text information input by the user, but is not limited to this, and can also be voice input information input by the user in the form of voice.

[0111] According to an embodiment of the present disclosure, operation instructions and input information may also be obtained in the following manner.

[0112] For example, in response to receiving voice input information, the voice input information is converted into text to obtain text information, the text information is displayed in the input box, and the text information is used as input information.

[0113] For another example, in response to receiving the voice confirmation information, an operation instruction for the target interface control is generated.

[0114] In one example, when interacting with a user through voice, the user's voice input information can be converted and semantically recognized to obtain text information. While freeing the user's hands, the text information is displayed in the input box so that the user can confirm whether the recognized information is correct, thereby improving the fluency of human-computer interaction.

[0115] In one example, in response to receiving voice confirmation information such as "confirm" or "yes", the voice confirmation information can be converted into an operation instruction for the target interface control, so that the voice human-computer interaction is compatible with the existing manually determined operation instructions, thereby improving versatility.

[0116] According to an embodiment of the present disclosure, the result generation model and the target function plug-in can be used as follows Figure 6 The deep learning model shown in Figure 6 The deep learning model training method shown in FIG. 1 is used to train the deep learning model so that the trained deep learning model can be applied to the following examples: Figure 2 In the information processing method shown, the accuracy of the target output result of the information processing method is improved.

[0117] Figure 6 A flowchart of a deep learning model training method according to an embodiment of the present disclosure is schematically shown.

[0118] like Figure 6 As shown, the method includes operations S610 to S640.

[0119] In operation S610, the sample input information is processed using the result generation model to obtain sample intermediate features and sample initial output results for responding to the sample input information.

[0120] In operation S620, the sample intermediate features are processed using the target function plug-in to obtain a sample evaluation result for evaluating the sample initial output result.

[0121] In operation S630, the sample initial output result is adjusted according to the sample evaluation result to obtain the sample target output result.

[0122] In operation S640, a deep learning model is trained based on the sample target output result and the output label matching the sample input information to obtain a trained deep learning model.

[0123] In one example, a target function plug-in is integrated into the result generation model, and the target function plug-in is used to evaluate the initial output result of the sample.

[0124] In one example, the output label is used to characterize the annotation result that matches the sample input information.

[0125] In one example, if Figure 6 The sample input information, sample intermediate features, sample initial output results, sample evaluation results, and sample target output results shown in FIG. Figure 2The interpretations and processing methods of the input information, intermediate features, initial output results, evaluation results, and target output results shown are the same or similar, only to reflect the different embodiments. Similarly, if the only difference between the nouns involved in the training method of the deep learning model and the nouns involved in the information processing method is "sample", the interpretations and processing methods are the same or similar. No further details will be given here.

[0126] According to the embodiments of the present disclosure, the sample intermediate features are obtained in the process of processing the sample input information using the result generation model. Since the sample intermediate features are generated based on the sample input information and obtained in the process of generating the sample initial output results, the sample intermediate features contain information related to the sample output information and the sample initial output results, and thus the sample initial output results are evaluated using the sample intermediate features, which can accurately control the adaptability between the sample input information and the sample target output results. Using the sample target output results and the output labels that match the sample input information to train the deep learning model can improve the training efficiency while improving the accuracy of the trained deep learning model.

[0127] According to an embodiment of the present disclosure, the result generation model includes a decoder and an output layer.

[0128] According to the embodiments of the present disclosure, Figure 6 The operation S610 shown, using the result generation model to process the sample input information to obtain the sample intermediate feature, may include: using a decoder to process the sample coding feature and the sample output sub-result to obtain the sample current output sub-result feature. The sample coding feature is obtained by processing the sample input information using the result generation model. The sample current output sub-result feature is processed using the output layer to obtain the sample current output sub-result. When it is determined that the sample current output sub-result meets the predetermined termination condition, the sample intermediate feature is obtained based on the sample current output sub-result feature.

[0129] By using the principle of processing data of the decoder, the sample intermediate feature integrates the sample coding feature corresponding to the sample input information and the feature corresponding to the sample output sub-result. The sample intermediate feature integrating the sample coding feature corresponding to the sample input information and the feature corresponding to the sample output sub-result is input into the target function plug-in, so that the target function plug-in can simultaneously learn the relevant knowledge of the sample input information and the relevant knowledge of the sample initial output result, enrich the data volume and data type of the sample input data of the target function plug-in, thereby improving the accuracy of the sample evaluation result output by the target function plug-in.

[0130] According to an embodiment of the present disclosure, the result generation model includes an encoder, a decoder, and an output layer.

[0131] According to the embodiments of the present disclosure, Figure 6 The operation S610 shown in FIG. 6 is to process the sample input information using the result generation model to obtain a sample initial output result for responding to the sample input information, including: processing the sample input information using an encoder to obtain a sample coding feature. Processing the sample coding feature and the sample output sub-result using a decoder to obtain a sample current output sub-result feature. Processing the sample current output sub-result feature using an output layer to obtain a sample current output sub-result. When it is determined that the sample current output sub-result satisfies a predetermined termination condition, obtaining a sample initial output result based on the sample current output sub-result and the sample output sub-result.

[0132] According to an embodiment of the present disclosure, a beam search strategy is adopted for decoding, so that the sample current output sub-result of the sample initial output result is generated by decoding based on the sample encoding features and the features corresponding to the sample output sub-results, thereby improving the closeness of the contextual association relationship between multiple tokens in the sample initial output result, improving the semantic coherence and consistency of the sample initial output result, and thereby improving the accuracy of the sample target initial output result obtained after adjusting the sample initial output result.

[0133] According to an embodiment of the present disclosure, the training method of the deep learning model may further include: determining that the current output sub-result of the sample satisfies a predetermined termination condition when determining that the result type of the current output sub-result of the sample satisfies a predetermined result type.

[0134] According to an embodiment of the present disclosure, adjusting the sample initial output result according to the sample evaluation result to obtain the sample target output result may include: when it is determined that the sample evaluation result indicates that the sample initial output result and the sample input information meet a predetermined matching condition, updating the sample initial output result using the sample predetermined output result to obtain the sample target output result. When it is determined that the sample evaluation result indicates that the sample initial output result and the sample input information do not meet the predetermined matching condition, obtaining the sample target output result based on the sample initial output result.

[0135] According to the embodiments of the present disclosure, Figure 6 The operation S640 shown, based on the sample output result and the output label matching the sample input information, training the deep learning model to obtain the trained deep learning model, may include: obtaining a first loss value based on the sample target output result and the output label matching the sample input information. Obtaining a second loss value based on the sample evaluation result and the evaluation label matching the sample input information. The evaluation label is used to characterize the labeled evaluation result matching the sample input information. Based on the first loss value and the second loss value, training the deep learning model to obtain the trained deep learning model.

[0136] In one example, a first loss function and a second loss function may be constructed. The first loss function is used to process the sample target output result and the output label matching the sample input information to obtain a first loss value. The second loss function is used to process the sample evaluation result and the evaluation label matching the sample input information to obtain a second loss value. The first loss value and the second loss value are weighted and summed to obtain a target loss value. Based on the target loss value, the model parameters of the deep learning model are adjusted to obtain a trained deep learning model.

[0137] In one example, the function types of the first loss function and the second loss function may be the same or different. For example, they may be a cross entropy loss function, but are not limited thereto, as long as they are loss functions used for model training.

[0138] The deep learning model is trained using the first loss value and the second loss value to obtain a trained deep learning model, which can train a result generation model integrated with a target function plug-in, thereby improving training accuracy and training efficiency.

[0139] According to an embodiment of the present disclosure, using a decoder to process sample coding features and sample output sub-results to obtain sample current output sub-result features may include: inputting the sample coding features, sample output sub-results and a mask matrix into the decoder to obtain sample current output sub-result features.

[0140] In one example, the mask matrix may be a matrix used for masking information, and may also be referred to as an attention mask matrix (Attention Mask).

[0141] In one example, the attention mechanism layer in the decoder can process the sample encoding features, the sample output sub-result and the mask matrix through the following formula (1) to obtain the sample current output sub-result features.

[0142]

[0143] Q represents the query vector, K represents the key vector, V represents the value vector, and d k Represents the coefficient.

[0144] In one example, Q, K, and V in formula (1) can be obtained based on the sample encoding features and the sample output sub-results.

[0145] In one example, based on Figure 7 The mask template shown is used to generate the mask matrix.

[0146] Figure 7 A schematic diagram of a mask template according to an embodiment of the present disclosure is schematically shown.

[0147] like Figure 7As shown, the horizontal column K of the mask template represents the key vector, and the vertical column Q of the mask template represents the query vector. Tokens S1, S2, and S3 represent sample input information. Tokens T1, T2, T3, and T4 represent output labels that match the sample input information. Tokens gMask, START, and gEND represent identifiers, respectively.

[0148] like Figure 7 As shown in the figure, the blank box indicates that the content is known and can be visualized. The crossed box indicates that the content is masked and cannot be seen.

[0149] A mask matrix can be generated based on the mask template. The sample input information, the mask matrix and the output label matching the sample input information are used to perform processing as shown in formula (1) to train the deep learning model and obtain a trained deep learning model.

[0150] During the parallel training process, the mask matrix can be used to ensure that the decoder obtains the sample output sub-results and sample encoding features before the current output sub-result of the sample, and cannot obtain the output sub-results of the sample current output sub-result and after the current output sub-result of the sample, thereby improving training accuracy while improving training efficiency.

[0151] Figure 8 The block diagram schematically shows an information processing device according to an embodiment of the present disclosure.

[0152] like Figure 8 As shown, the information processing device 800 includes: a result processing module 810, an evaluation module 820 and an adjustment module 830.

[0153] The result processing module 810 is used to process the input information using the result generation model to obtain intermediate features and initial output results for responding to the input information. The result generation model is integrated with a target function plug-in, which is used to evaluate the initial output results.

[0154] The evaluation module 820 is used to process the intermediate features using the target function plug-in to obtain an evaluation result for evaluating the initial output result.

[0155] The adjustment module 830 is used to adjust the initial output result according to the evaluation result to obtain the target output result.

[0156] According to an embodiment of the present disclosure, the result generation model includes a decoder and an output layer, and the initial output result includes output sub-results.

[0157] According to an embodiment of the present disclosure, the result processing module includes: a first decoding submodule, a first output submodule, and a first determination submodule.

[0158] The first decoding submodule is used to process the encoding feature and the output sub-result using a decoder to obtain the current output sub-result feature. The encoding feature is obtained by processing the input information using the result generation model.

[0159] The first output submodule is used to use the output layer to process the characteristics of the current output subresult to obtain the current output subresult.

[0160] The first determination submodule is used to obtain an intermediate feature based on a feature of the current output subresult when it is determined that the current output subresult satisfies a predetermined termination condition.

[0161] According to an embodiment of the present disclosure, the result generation model includes an encoder, a decoder and an output layer, and the initial output result includes output sub-results.

[0162] According to an embodiment of the present disclosure, the result processing module includes: an encoding submodule, a second decoding submodule, a second output submodule, and a second determination submodule.

[0163] The encoding submodule is used to process input information using the encoder to obtain encoding features.

[0164] The second decoding submodule is used to use a decoder to process the encoding features and the output sub-results to obtain current output sub-result features.

[0165] The second output submodule is used to use the output layer to process the characteristics of the current output subresult to obtain the current output subresult.

[0166] The second determination submodule is used to obtain an initial output result based on the current output subresult and the output subresult when it is determined that the current output subresult satisfies a predetermined termination condition.

[0167] According to an embodiment of the present disclosure, the information processing device further includes: a determination module.

[0168] The determination module is used to determine that the current output sub-result meets the predetermined termination condition when it is determined that the result type of the current output sub-result meets the predetermined result type.

[0169] According to an embodiment of the present disclosure, the information type of the input information includes a question type.

[0170] According to an embodiment of the present disclosure, the adjustment module includes: an update submodule.

[0171] The updating submodule is used to update the initial output result with a predetermined output result matching the question type to obtain a target output result when it is determined that the evaluation result represents that the initial output result and the input information meet a predetermined matching condition.

[0172] According to an embodiment of the present disclosure, the information type of the input information includes a search type, and the initial output result includes a plurality of output sub-results.

[0173] According to an embodiment of the present disclosure, the adjustment module includes: a first sorting submodule and a second sorting submodule.

[0174] The first sorting submodule is used to determine the sorting results of the multiple output sub-results based on the evaluation results.

[0175] The second sorting submodule is used to sort the multiple output sub-results based on the sorting result to obtain the target output result.

[0176] According to an embodiment of the present disclosure, the information type of the input information includes a search type, and the initial output result includes a plurality of output sub-results;

[0177] According to an embodiment of the present disclosure, the adjustment module includes: a third sorting submodule, a feedback determination submodule, and a fourth sorting submodule.

[0178] The third sorting submodule is used to determine the sorting results of the multiple output sub-results based on the evaluation results.

[0179] The feedback determination submodule is used to determine a plurality of target feedback data matching the plurality of output sub-results from a predetermined feedback database.

[0180] The fourth sorting submodule is used to sort the multiple target feedback data based on the sorting result to obtain the target output result.

[0181] According to an embodiment of the present disclosure, the information processing device further includes: an intention recognition module and a first plug-in determination module.

[0182] The intent recognition module is used to perform intent recognition on the intermediate features and obtain intent recognition results.

[0183] The first plug-in determination module is used to determine a target function plug-in that matches the intention recognition result from multiple function plug-ins.

[0184] According to an embodiment of the present disclosure, the information processing device further includes: an instruction response module.

[0185] The instruction response module is used to obtain input information from the input box in response to the operation instruction for the target interface control.

[0186] According to an embodiment of the present disclosure, the information processing device further includes: a second plug-in determination module

[0187] The second plug-in determination module is used to determine a target functional plug-in matching the operation instruction from a plurality of functional plug-ins based on the operation instruction.

[0188] According to an embodiment of the present disclosure, the information processing device further includes: a voice response module, a display module, and a determination response module.

[0189] The voice response module is used to convert the voice input information into text in response to receiving the voice input information to obtain text information.

[0190] The display module is used to display text information in the input box.

[0191] The determination response module is used to generate an operation instruction for a target interface control in response to receiving the voice confirmation information.

[0192] Fig. 9 A block diagram of a deep learning model training device according to an embodiment of the present disclosure is schematically shown.

[0193] like Fig. 9 As shown, the training device 900 of the deep learning model includes: a sample processing module 910, a sample evaluation module 920, a sample adjustment module 930 and a training module 940.

[0194] The sample processing module 910 is used to process the sample input information using the result generation model to obtain the sample intermediate features and the sample initial output result for responding to the sample input information. The result generation model is integrated with a target function plug-in, which is used to evaluate the sample initial output result.

[0195] The sample evaluation module 920 is used to process the intermediate features of the sample using the target function plug-in to obtain a sample evaluation result for evaluating the initial output result of the sample.

[0196] The sample adjustment module 930 is used to adjust the sample initial output result according to the sample evaluation result to obtain the sample target output result.

[0197] The training module 940 is used to train the deep learning model based on the sample target output result and the output label matching the sample input information to obtain a trained deep learning model. The output label is used to represent the annotation result matching the sample input information.

[0198] According to an embodiment of the present disclosure, the result generation model includes a decoder and an output layer, and the sample initial output result includes a sample output sub-result.

[0199] According to an embodiment of the present disclosure, the sample processing module includes: a first sample decoding submodule, a first sample output submodule, and a first sample determination submodule.

[0200] The first sample decoding submodule is used to process the sample coding feature and the sample output sub-result by using a decoder to obtain the sample current output sub-result feature. The sample coding feature is obtained by processing the sample input information by using the result generation model.

[0201] The first sample output submodule is used to use the output layer to process the characteristics of the sample current output subresult to obtain the sample current output subresult.

[0202] The first sample determination submodule is used to obtain the sample intermediate feature based on the feature of the sample current output subresult when it is determined that the sample current output subresult meets the predetermined termination condition.

[0203] According to an embodiment of the present disclosure, the result generation model includes an encoder, a decoder and an output layer, and the sample initial output result includes a sample output sub-result.

[0204] According to an embodiment of the present disclosure, the sample processing module includes: a sample encoding submodule, a second sample decoding submodule, a second sample output submodule, and a second sample determination submodule.

[0205] The sample encoding submodule is used to process sample input information using an encoder to obtain sample encoding features.

[0206] The second sample decoding submodule is used to process the sample encoding characteristics and the sample output sub-result by using a decoder to obtain the sample current output sub-result characteristics.

[0207] The second sample output submodule is used to use the output layer to process the characteristics of the sample current output subresult to obtain the sample current output subresult.

[0208] The second sample determination submodule is used to obtain the sample initial output result based on the sample current output subresult and the sample output subresult when it is determined that the sample current output subresult meets a predetermined termination condition.

[0209] According to an embodiment of the present disclosure, the training module includes: a first loss determination submodule, a second loss determination submodule and a training submodule.

[0210] The first loss determination submodule is used to obtain a first loss value based on the sample target output result and the output label matching the sample input information.

[0211] The second loss determination submodule is used to obtain a second loss value based on the sample evaluation result and the evaluation label that matches the sample input information. The evaluation label is used to characterize the labeled evaluation result that matches the sample input information.

[0212] The training submodule is used to train the deep learning model based on the first loss value and the second loss value to obtain a trained deep learning model.

[0213] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0214] According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a method as in the embodiment of the present disclosure.

[0215] According to an embodiment of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute a method as described in an embodiment of the present disclosure.

[0216] According to an embodiment of the present disclosure, a computer program product includes a computer program, and when the computer program is executed by a processor, the method according to the embodiment of the present disclosure is implemented.

[0217] Fig.10 A schematic block diagram of an example electronic device 1000 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0218] like Fig.10 As shown, the device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a random access memory (RAM) 1003. In the RAM 1003, various programs and data required for the operation of the device 1000 can also be stored. The computing unit 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0219] A number of components in the device 1000 are connected to an input / output (I / O) interface 1005, including: an input unit 1006, such as a keyboard, a mouse, etc.; an output unit 1007, such as various types of displays, speakers, etc.; a storage unit 1008, such as a disk, an optical disk, etc.; and a communication unit 1009, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1009 allows the device 1000 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0220] The computing unit 1001 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1001 performs the various methods and processes described above, such as information processing methods. For example, in some embodiments, the information processing method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by the computing unit 1001, one or more steps of the information processing method described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to perform the information processing method in any other appropriate manner (e.g., by means of firmware).

[0221] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0222] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0223] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0224] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0225] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0226] A computer system may include a client and a server. The client and the server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises through computer programs running on respective computers and having a client-server relationship to each other. The server may be a cloud server, a server in a distributed system, or a server combined with a blockchain.

[0227] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0228] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. An information processing method, comprising: Processing the input information using a result generation model to obtain intermediate features and an initial output result for responding to the input information, wherein the result generation model is integrated with a target function plug-in, and the target function plug-in is used to evaluate whether the initial output result matches the input information; Processing the intermediate features using the target function plug-in to obtain an evaluation result for evaluating the initial output result; and Adjusting the initial output result according to the evaluation result to obtain a target output result; The result generation model includes a decoder and an output layer, and the initial output result includes output sub-results. The method of using the result generation model to process the input information to obtain intermediate features includes: Processing the encoding feature and the output sub-result using the decoder to obtain a current output sub-result feature, wherein the encoding feature is obtained by processing the input information using the result generation model; Processing the features of the current output sub-result using the output layer to obtain the current output sub-result; and In the case where it is determined that the current output sub-result satisfies a predetermined termination condition, the intermediate feature is obtained based on the feature of the current output sub-result.

2. The method according to claim 1, wherein: The result generation model includes an encoder, a decoder and an output layer, and the initial output result includes output sub-results. The using result generation model to process the input information to obtain an initial output result for responding to the input information includes: Processing the input information using the encoder to obtain encoding features; Processing the encoding feature and the output sub-result using the decoder to obtain a current output sub-result feature; Processing the features of the current output sub-result using the output layer to obtain the current output sub-result; and In the case where it is determined that the current output sub-result satisfies a predetermined termination condition, the initial output result is obtained based on the current output sub-result and the output sub-result.

3. The method according to claim 1, further comprising: In the case where it is determined that the result type of the current output sub-result satisfies the predetermined result type, it is determined that the current output sub-result satisfies the predetermined termination condition.

4. The method according to claim 1, wherein: The information type of the input information includes question type; The adjusting the initial output result according to the evaluation result to obtain the target output result includes: In the case where it is determined that the evaluation result indicates that the initial output result and the input information meet a predetermined matching condition, the initial output result is updated using a predetermined output result matching the question type to obtain the target output result.

5. The method according to claim 1, wherein: The information type of the input information includes a search type, and the initial output result includes a plurality of output sub-results; The adjusting the initial output result according to the evaluation result to obtain the target output result includes: Based on the evaluation result, determining a ranking result of the plurality of output sub-results; as well as Based on the sorting result, the multiple output sub-results are sorted to obtain the target output result.

6. The method according to claim 1, wherein: The information type of the input information includes a search type, and the initial output result includes a plurality of output sub-results; The adjusting the initial output result according to the evaluation result to obtain the target output result includes: Based on the evaluation result, determining a ranking result of the plurality of output sub-results; Determining a plurality of target feedback data matching the plurality of output sub-results from a predetermined feedback database; and Based on the sorting result, the plurality of target feedback data are sorted to obtain a target output result.

7. The method according to any one of claims 1 to 6, further comprising: Performing intent recognition on the intermediate features to obtain an intent recognition result; as well as The target function plug-in matching the intention recognition result is determined from a plurality of function plug-ins.

8. The method according to any one of claims 1 to 6, further comprising: In response to an operation instruction for a target interface control, the input information is obtained from the input box.

9. The method according to claim 8, further comprising: Based on the operation instruction, the target function plug-in matching the operation instruction is determined from a plurality of function plug-ins.

10. The method according to claim 8, further comprising: In response to receiving voice input information, converting the voice input information into text to obtain text information; Displaying the text information in the input box; as well as In response to receiving the voice confirmation information, the operation instruction for the target interface control is generated.

11. A method for training a deep learning model, comprising: Processing sample input information using a result generation model to obtain sample intermediate features and a sample initial output result for responding to the sample input information, wherein a target function plug-in is integrated in the result generation model, and the target function plug-in is used to evaluate whether the sample initial output result matches the sample input information; Processing the sample intermediate features using the target function plug-in to obtain a sample evaluation result for evaluating the sample initial output result; Adjusting the sample initial output result according to the sample evaluation result to obtain a sample target output result; and Based on the sample target output result and the output label matching the sample input information, training the deep learning model to obtain a trained deep learning model, wherein the output label is used to represent the annotation result matching the sample input information; The result generation model includes a decoder and an output layer, and the sample initial output result includes a sample output sub-result. The method of using the result generation model to process sample input information to obtain sample intermediate features includes: Processing the sample coding feature and the sample output sub-result by using the decoder to obtain the sample current output sub-result feature, wherein the sample coding feature is obtained by processing the sample input information by using the result generation model; Processing the features of the sample current output sub-result using the output layer to obtain the sample current output sub-result; and When it is determined that the current output sub-result of the sample satisfies a predetermined termination condition, the sample intermediate feature is obtained based on the feature of the current output sub-result of the sample.

12. The method according to claim 11, wherein: The result generation model includes an encoder, a decoder and an output layer, and the sample initial output result includes a sample output sub-result. The using the result generation model to process the sample input information to obtain a sample initial output result for responding to the sample input information includes: Processing the sample input information using the encoder to obtain sample encoding features; Processing the sample encoding feature and the sample output sub-result using the decoder to obtain a sample current output sub-result feature; Processing the features of the sample current output sub-result using the output layer to obtain the sample current output sub-result; and In the case where it is determined that the sample current output sub-result satisfies a predetermined termination condition, the sample initial output result is obtained based on the sample current output sub-result and the sample output sub-result.

13. The method according to claim 11, wherein: The step of training the deep learning model based on the sample target output result and the output label matching the sample input information to obtain a trained deep learning model includes: Obtaining a first loss value based on the sample target output result and an output label matching the sample input information; Obtaining a second loss value based on the sample evaluation result and an evaluation label that matches the sample input information, wherein the evaluation label is used to characterize the labeled evaluation result that matches the sample input information; and Based on the first loss value and the second loss value, the deep learning model is trained to obtain a trained deep learning model.

14. An information processing device, comprising: A result processing module, used to process input information using a result generation model to obtain intermediate features and an initial output result for responding to the input information, wherein the result generation model is integrated with a target function plug-in, and the target function plug-in is used to evaluate whether the initial output result matches the input information; an evaluation module, configured to process the intermediate features using the target function plug-in to obtain an evaluation result for evaluating the initial output result; and An adjustment module, used to adjust the initial output result according to the evaluation result to obtain a target output result; The result generation model includes a decoder and an output layer, and the initial output result includes output sub-results. The result processing module includes: A first decoding submodule, configured to process the encoding feature and the output sub-result using the decoder to obtain a current output sub-result feature, wherein the encoding feature is obtained by processing the input information using the result generation model; A first output submodule, configured to use the output layer to process the features of the current output subresult to obtain the current output subresult; and The first determination submodule is configured to obtain the intermediate feature based on the feature of the current output subresult when it is determined that the current output subresult satisfies a predetermined termination condition.

15. The device according to claim 14, wherein: The result generation model includes an encoder, a decoder and an output layer, and the initial output result includes output sub-results. The result processing module includes: An encoding submodule, used for processing the input information using the encoder to obtain encoding features; A second decoding submodule, configured to process the encoding feature and the output subresult using the decoder to obtain a current output subresult feature; A second output submodule, configured to process the features of the current output subresult using the output layer to obtain the current output subresult; and The second determining submodule is configured to obtain the initial output result based on the current output subresult and the output subresult when it is determined that the current output subresult satisfies a predetermined termination condition.

16. The apparatus according to claim 14, further comprising: The determination module is used to determine that the current output sub-result satisfies the predetermined termination condition when it is determined that the result type of the current output sub-result satisfies the predetermined result type.

17. The device according to claim 14, wherein: The information type of the input information includes question type; The adjustment module comprises: The updating submodule is used to update the initial output result with a predetermined output result matching the question type to obtain the target output result when it is determined that the evaluation result indicates that the initial output result and the input information meet a predetermined matching condition.

18. The device according to claim 14, wherein: The information type of the input information includes a search type, and the initial output result includes a plurality of output sub-results; The adjustment module comprises: A first sorting submodule, configured to determine a sorting result of the plurality of output sub-results based on the evaluation result; and The second sorting submodule is used to sort the multiple output sub-results based on the sorting result to obtain the target output result.

19. The device according to claim 14, wherein: The information type of the input information includes a search type, and the initial output result includes a plurality of output sub-results; The adjustment module comprises: A third sorting submodule, configured to determine a sorting result of the plurality of output sub-results based on the evaluation result; a feedback determination submodule, configured to determine a plurality of target feedback data matching the plurality of output sub-results from a predetermined feedback database; and The fourth sorting submodule is used to sort the multiple target feedback data based on the sorting result to obtain a target output result.

20. The device according to any one of claims 14 to 19, further comprising: An intention recognition module, used to perform intention recognition on the intermediate features to obtain an intention recognition result; as well as The first plug-in determination module is used to determine the target function plug-in that matches the intention recognition result from multiple function plug-ins.

21. The apparatus according to any one of claims 14 to 19, further comprising: The instruction response module is used to obtain the input information from the input box in response to the operation instruction for the target interface control.

22. The apparatus according to claim 21, further comprising: The second plug-in determination module is used to determine the target function plug-in matching the operation instruction from multiple function plug-ins based on the operation instruction.

23. The apparatus according to claim 21, further comprising: A voice response module, configured to convert the voice input information into text in response to receiving the voice input information to obtain text information; A display module, used to display the text information in the input box; as well as A determination response module is used to generate the operation instruction for the target interface control in response to receiving the voice confirmation information.

24. A training device for a deep learning model, comprising: A sample processing module, used to process sample input information using a result generation model to obtain sample intermediate features and a sample initial output result for responding to the sample input information, wherein the result generation model is integrated with a target function plug-in, and the target function plug-in is used to evaluate whether the sample initial output result matches the sample input information; A sample evaluation module, used to process the sample intermediate features using the target function plug-in to obtain a sample evaluation result for evaluating the sample initial output result; A sample adjustment module, configured to adjust the sample initial output result according to the sample evaluation result to obtain a sample target output result; and A training module, configured to train the deep learning model based on the sample target output result and the output label matching the sample input information, so as to obtain a trained deep learning model, wherein the output label is used to represent the annotation result matching the sample input information; The result generation model includes a decoder and an output layer, and the sample initial output result includes a sample output sub-result. The sample processing module comprises: A first sample decoding submodule, configured to use the decoder to process a sample encoding feature and the sample output subresult to obtain a sample current output subresult feature, wherein the sample encoding feature is obtained by using the result generation model to process the sample input information; A first sample output submodule, configured to use the output layer to process the features of the sample current output subresult to obtain the sample current output subresult; and The first sample determination submodule is used to obtain the sample intermediate feature based on the feature of the sample current output subresult when it is determined that the sample current output subresult meets a predetermined termination condition.

25. The device according to claim 24, wherein: The result generation model includes an encoder, a decoder and an output layer, and the sample initial output result includes a sample output sub-result. The sample processing module comprises: A sample encoding submodule, used to process the sample input information using the encoder to obtain sample encoding features; A second sample decoding submodule, configured to use the decoder to process the sample encoding feature and the sample output subresult to obtain a sample current output subresult feature; A second sample output submodule, configured to use the output layer to process the features of the sample current output subresult to obtain the sample current output subresult; and The second sample determination submodule is configured to obtain the sample initial output result based on the sample current output subresult and the sample output subresult when it is determined that the sample current output subresult satisfies a predetermined termination condition.

26. The device according to claim 24, wherein: The training module includes: A first loss determination submodule, configured to obtain a first loss value based on the sample target output result and an output label matching the sample input information; a second loss determination submodule, configured to obtain a second loss value based on the sample evaluation result and an evaluation tag matching the sample input information, wherein the evaluation tag is used to characterize the labeled evaluation result matching the sample input information; and A training submodule is used to train the deep learning model based on the first loss value and the second loss value to obtain a trained deep learning model.

27. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 13.

28. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 13.

29. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any one of claims 1 to 13.

Images