Response method and device based on large model, electronic equipment and storage medium

Abstract

The invention provides a response method and device based on a large model, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence, in particular to the fields of large models, large language models, dialogue models and generative models. According to the specific implementation scheme, at least one target operation record is recalled from a database according to correlation between an input text of an object and multiple operation records of the database and respective weights of the multiple operation records; inputting the input text, the at least one target operation record and the target prompt information into the large model to obtain a response text for the input text; wherein the operation record is generated based on the operation behavior of the object, and the weight is determined based on the operation behavior associated with the operation record and the historical recall times of the operation record.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to the fields of large language models, conversational models, and generative models. More specifically, the present disclosure provides a large-model-based answering method, device, electronic device, storage medium, and computer program product. Background Art

[0002] In the modern information-intensive working environment, users need to process massive amounts of data. Since information recording tools such as bookmarks, favorites, and history records lack the ability to intelligently process and capture comprehensive data, users are less efficient in finding and using information, which limits the effective use of information. Summary of the invention

[0003] The present disclosure provides a large model-based answering method, device, electronic device, storage medium, and computer program product.

[0004] According to one aspect of the present disclosure, a large model-based answering method, apparatus, electronic device, storage medium, and computer program product are provided.

[0005] According to another aspect of the present disclosure, a response device based on a large model is provided, comprising: a recall module and an input module. The recall module is used to recall at least one target operation record from a database according to the correlation between an input text of an object and a plurality of operation records in the database, and the weights of the plurality of operation records. The input module is used to input the input text, at least one target operation record and target prompt information into the large model to obtain a response text for the input text. The operation record is generated based on the operation behavior of the object, and the weight is determined based on the operation behavior associated with the operation record and the number of times the operation record has been recalled in history.

[0006] 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 the method provided by the present disclosure.

[0007] 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 a computer to execute the method provided by the present disclosure.

[0008] 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 provided by the present disclosure is implemented.

[0009] 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

[0010] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0011] Figure 1 is a schematic diagram of an application scenario of a large model-based response method and device according to an embodiment of the present disclosure;

[0012] Figure 2 is a schematic flow chart of a large model-based answering method according to an embodiment of the present disclosure;

[0013] Figure 3 is a schematic diagram of a large model-based response method according to an embodiment of the present disclosure;

[0014] Figure 4 is a schematic diagram of a large model-based response method according to an embodiment of the present disclosure;

[0015] Figure 5 is a schematic structural block diagram of a large model-based response device according to an embodiment of the present disclosure; and

[0016] Figure 6 It is a structural block diagram of an electronic device used to implement the large model-based response method of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0017] 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.

[0018] In the technical solutions disclosed in this disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the relevant laws and regulations and do not violate public order and good morals. In the technical solutions disclosed in this disclosure, the user's authorization or consent is obtained before obtaining or collecting user personal information.

[0019] For example, before collecting the user's personal information (such as the user's operation behavior and operation records), a query page can be popped up to the user in advance, asking the user whether to agree to collect the user's behavior information in order to provide personalized services, and the data can be collected after obtaining the user's authorization. If the user's authorization is not obtained, the data will not be collected.

[0020] In some embodiments, users can use information recording tools such as bookmarks and favorites to assist in their work. These information recording tools mainly rely on manual operation or automatic recording of browsing history and other non-personalized means. These information recording tools only provide static and fragmented information collection, lacking in-depth understanding and intelligent processing of user behavior.

[0021] The above-mentioned information recording tools have the following main shortcomings: First, insufficient understanding of user information. For example, the tool cannot fully capture the user's specific operations, resulting in one-sided information recording and failure to fully meet user needs. Second, insufficient personalized services. For example, due to the lack of in-depth analysis of user behavior, the tool cannot provide personalized data processing and suggestions, and cannot effectively support the user's specific work scenarios. Third, weak information integration capabilities. For example, most tools are limited to a single platform or format and cannot integrate information across platforms or multiple formats, limiting their scope of application and effectiveness.

[0022] The disclosed embodiment aims to propose a large model-based answering method. In practical applications, even if the user's needs are not clearly expressed, the target operation records with high correlation with the user's historical operation behaviors and operation records can be recalled based on the user's historical operation behaviors and operation records, and the generation of the answer text can be assisted based on the target operation records. This can shorten the information search and processing time, allowing users to quickly obtain the required information from a large amount of information, thereby improving work efficiency, the accuracy of information utilization, and providing personalized services.

[0023] The technical solution provided by the present disclosure will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] Figure 1 It is a schematic diagram of an application scenario of a large model-based response method and device according to an embodiment of the present disclosure.

[0025] It should be noted that Figure 1 What is shown is merely an example of a system architecture 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 it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0026] 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.

[0027] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptops, desktop computers, etc.

[0028] The server 105 may be a server that provides various services, such as a background management server (only an example) that provides support for websites browsed by users 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 a response text generated according to the user's input text, etc.) to the terminal device.

[0029] It should be noted that the large model-based response method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the large model-based response device provided in the embodiment of the present disclosure can generally be set in the server 105. The large model-based response method provided in the embodiment of the present disclosure can 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 large model-based response device provided in the embodiment of the present disclosure can also be set 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.

[0030] 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 the implementation requirements.

[0031] Figure 2 It is a schematic flow chart of a large model-based answering method according to an embodiment of the present disclosure.

[0032] like Figure 2 As shown, the large model-based response method 200 may include operations S210 to S220.

[0033] In operation S210 , at least one target operation record is recalled from the database according to the correlation between the input text of the object and the plurality of operation records in the database, and the weights of the plurality of operation records.

[0034] For example, the object can be a user. For example, the input text can be information entered by the user through the front-end operation page, such as "what work I did today", "write me a children's story", "my thoughts on digital human practice", etc. The user enters these contents and expects the big model to generate responses to these contents.

[0035] For example, a database can be pre-built and updated as needed, and the database can store multiple operation records. It should be noted that in the process of using an electronic device over a period of time in the past, the object can perform multiple operation behaviors through an interactive page, and these operation behaviors of the object will generate associated operation records. For example, pages include browser pages, mailbox pages, favorites pages, etc. For example, operation behaviors include browsing, clicking, swiping, scrolling, copying, editing, etc. Browsing can be browsing the content text in the page, swiping can be swiping part of the text in the page, copying can be copying part of the text in the page, and editing can be editing text in the page. The above-mentioned browsed, swiped, copied, and edited texts are called initial records. The operation record can include the initial record, and key information such as summary and keywords can also be extracted based on the initial record. The operation record can include these key information.

[0036] For example, an operation record has a corresponding weight, and the weight is determined based on the operation behavior associated with the operation record and the number of times the operation record has been recalled in history. For example, the coefficient value of each operation behavior can be pre-configured, such as the coefficient value of "edit" is 2, and the coefficient value of browse is 1. The coefficient values ​​of the operation behaviors associated with the same operation record and the number of times they have been recalled in history can be summed, and the weight of the operation record can be determined based on the summed result.

[0037] For example, similarities such as text similarity and semantic similarity between the input text and the plurality of operation records may be determined, and the relevance may be determined using the similarities.

[0038] For example, operation records with high relevance and high weight may be used as target operation records.

[0039] In operation S220, the input text, at least one target operation record and target prompt information are input into the macro model to obtain a response text for the input text.

[0040] For example, the target prompt information may be pre-configured and directly read when needed. For example, the target prompt information may be text in natural language form. For example, the large model may be a large language model (LLM).

[0041] For example, the input text, target operation record, and target prompt information can be combined into input information and then input into the big model, so that the response text is obtained based on the big model.

[0042] According to the technical solution provided by the embodiment of the present disclosure, the user's operation records can be stored in the database during the user's browsing of web pages and processing of emails. After obtaining the user's input text, even if the input text does not express the user's needs in detail, some target operation records with high relevance and high weight can be recalled from the database based on the user's historical operation behavior and operation records, and then a response text is generated based on these recalled target operation records and returned to the user, so that the generated content can better meet the user's personalized needs. In addition, since the response text is generated based on the target operation record, the response text can reflect the important information in the target operation record to a certain extent, so that the user does not need to manually summarize the information from a large number of operation records, but directly obtains the summarized information from the response text, thereby shortening the information search and processing time and improving work efficiency.

[0043] Next, the weight of the operation record is described. It should be noted that, in this embodiment, the weight can be recorded in a database, and the weight can be updated regularly or in response to a predetermined operation.

[0044] In one example, the weight is positively correlated with the number of times the operation record is recalled within a historical predetermined period. The historical predetermined period may be within a predetermined period in the past, such as the past month. For example, when the subject uses a large model to generate a response text, if a certain operation record is recalled many times, it means that the user pays more attention to the operation record, so the weight of the operation record can be increased, thereby increasing the probability of the operation record being recalled, and further improving the accuracy of the response text.

[0045] In another example, the weight is positively correlated with the number of associated operation behaviors. For example, an operation record can be associated with multiple operation behaviors. For example, a user selected a piece of text in a page last week, and the selection operation is the first operation behavior. The subject selected the same piece of text in the page again today, and the selection operation is the second operation behavior. It can be seen that the first operation behavior and the second operation behavior are both associated with the operation record determined based on the text. The more operation behaviors associated with the same operation record, the more attention the user pays to the operation record. Therefore, the weight of the operation record can be increased, thereby increasing the probability of the operation record being recalled, and then improving the accuracy of the response text.

[0046] In another example, the weight is negatively correlated with the interval duration of the associated operation behavior, and the interval duration is the duration between the moment when the operation behavior occurs and the target moment. For example, the first operation behavior increases the weight value of the operation record by 1, and the second operation behavior increases the weight value of the operation record by 3. It can be understood that the user's needs have a certain timeliness. This example makes the weight of the operation records associated with recent historical operation behaviors higher, and makes the weight of the operation records associated with historical operation behaviors that occurred a long time ago lower, so that the response text can be more in line with the user's latest needs, thereby improving the accuracy of the response text and improving the user experience.

[0047] In another example, the weight is positively correlated with the number of target category operation behaviors contained in the associated operation behavior. For example, the target category may include copying, editing, etc. The user performs operations such as copying and editing on the text in the page, indicating that these operation records can provide more useful information to the user. Therefore, the weight of the operation record can be increased, thereby increasing the probability of the operation record being recalled, so as to generate a response text based on the more important information, so that the response text is more in line with user needs, improves the accuracy of the response text, and enhances the user experience.

[0048] For example, if the subject selects a text on the page, the text or key information of the text can be recorded as an operation and stored in the database. The operation behavior corresponding to the operation record is "selection", and the weight of the operation record can be marked as 1. After a period of time, the user selects the text on the page again. At this time, the operation record already exists in the database, and the weight of the operation record can be increased by 2, that is, the weight is 3. After a predetermined period of time, the weight of the operation record does not continue to increase, and the weight can be reduced by 1, and the weight is 2 at this time. When the weight is reduced to 0, the operation record can be deleted from the database.

[0049] Next, the process of building a database for storing operation records is described.

[0050] In this embodiment, it can be determined whether the initial record meets the predetermined condition according to at least one of the scenario of the operation behavior associated with the operation record, the number of characters in the initial record, and the weight of the initial record. If the initial record meets the predetermined condition, key information can be extracted from the initial record, and the extracted key information is added to the operation record. If the initial record does not meet the predetermined condition, the initial record is added to the operation record. The key information includes, for example, at least one of a summary and a keyword.

[0051] For example, the predetermined condition includes at least one of a plurality of sub-conditions. Each sub-condition will be described below.

[0052] For example, one sub-condition is to take the scenario of the operation behavior associated with the operation record as the target scenario, and the target scenario is, for example, the scenario of a page using a browser. For example, the text edited by the object in the process of using the mailbox, this scenario does not meet the sub-condition. At this time, key information may not be extracted from the text, but the text may be directly used as an operation record. For another example, when the object uses a browser to browse the web, this scenario meets the sub-condition. At this time, the text in the web page is the initial record, and key information can be extracted from the initial record, and then the extracted key information is used as the operation record. This embodiment determines whether to extract key information from the initial record based on the scenario, so that the stored operation record can be matched with the scenario, and the input information of the large model is adjusted through the scenario so that the response text meets the scenario requirements.

[0053] For another example, another sub-condition is that the number of characters in the initial record is greater than or equal to a predetermined number. Too many characters are not conducive to the understanding of the large model, so the sub-condition is met when the number of characters exceeds the predetermined number. At this time, key information can be extracted from the longer initial record, and then the extracted key information can be used as an operation record. When the number of characters does not exceed the predetermined number, the sub-condition is not met, and the initial record can be used as an operation record. In this embodiment, key information is extracted from the initial record with a large number of characters, and the key information is used as an operation record. This can reduce the length of the operation record and avoid the interference of too many characters in the understanding of the large model, thereby improving the accuracy of the large model reasoning and the reasoning efficiency. At the same time, for the initial record with a small number of characters, the initial record can be used as an operation record. This does not affect the understanding of the large model, and also makes the operation record more complete, thereby improving the reasoning effect of the large model. And because the number of characters in the initial record is small, it will not cause the problem of low reasoning efficiency of the large model.

[0054] For another example, another sub-condition is that the weight of the initial record is less than or equal to the predetermined weight. For example, the weight of the initial record is large, so it is necessary to maintain the integrity of the initial record as much as possible to ensure that the semantics of the operation record is more complete. Therefore, the initial record does not meet the sub-condition. At this time, the key information of these texts can be not extracted, but the texts can be directly used as operation records. For another example, the weight of the initial record is small, so the integrity of the initial record can be appropriately reduced. The initial record does not meet the sub-condition. At this time, the key information of these texts can be extracted, and then the extracted key information can be used as an operation record. In this embodiment, the key information is extracted for the initial record with a lower weight, and the key information is used as an operation record, which can reduce the length of the operation record with low importance. At the same time, for the initial record with a higher weight, the initial record can be used as an operation record, so that the information contained in the operation record with higher importance is more complete. In the process of determining the response text by reasoning of the large model, reasoning based on the target operation record with a low weight can improve the reasoning efficiency, and reasoning based on the target operation record with a high weight can improve the accuracy of the response text.

[0055] According to another embodiment of the present disclosure, a preliminary screening may be performed by correlation first, and then a secondary screening may be performed by weight, so as to recall the target operation records. Next, the process of recalling the target operation records is described.

[0056] At least one attribute of the input text may be determined first. For example, the attribute may include the features of the input text, and the features of the input text may be obtained by extracting features from the input text through an encoder. For another example, the attribute may include entity information in the input text, and named entity recognition may be performed on the input text to obtain entities, and entities may include time, place, person, etc. For another example, the attribute may include the intent of the input text, and intent recognition may be performed on the input text to obtain the intent of the input text. In practical applications, target operation records may be recalled through multiple attributes. Compared with the method of recalling using only a single attribute, multiple attributes may increase the dimension of recall, thereby making the recalled target operation records more comprehensive and accurate.

[0057] Next, for each of the at least one attribute, a first operation record set may be determined from the database according to the similarity between the attribute and the plurality of operation records in the database. For example, if the similarity between the operation record and a predetermined number of attributes is greater than a corresponding similarity threshold, the operation record may be added to the first operation record set, and the predetermined number may be 1, 2, 3, etc., which is not limited in this embodiment.

[0058] Next, at least one target operation record may be determined from the first operation record set according to the weight of each operation record in the first operation record set. For example, the operation records in the first operation record set may be sorted according to the weight, and then based on the sorting, several operation records with the largest weights may be used as target operation records.

[0059] This embodiment first performs a preliminary screening by correlation, and then performs a secondary screening by weight to recall target operation records with high correlation and high weight. In this way, operation records with high correlation with the input text can be accurately screened, which is conducive to accurately locating the content of interest to the user, so that the recalled target operation records meet the actual needs of the user.

[0060] According to another embodiment of the present disclosure, a preliminary screening may be performed by weight first, and then a secondary screening may be performed by correlation, so as to recall the target operation records. Next, the process of recalling the target operation records is described.

[0061] At least one attribute of the input text may be determined first, and the attribute may include features of the input text, entity information in the input text, and the intention of the input text.

[0062] Next, the second operation record set can be determined from the database according to the weights of the multiple operation records in the database. For example, the operation records in the database can be sorted according to the weights, and then based on the sorting, several operation records with the largest weights are added to the second operation record set.

[0063] Next, for each of the at least one attribute, at least one target operation record can be determined from the second operation record set according to the similarity between the attribute and multiple operation records in the second operation record set. For example, when the similarity between the operation record and a predetermined number of attributes is greater than the corresponding similarity threshold, the operation record can be determined as a target operation record. The predetermined number can be 1, 2, 3, etc., and this embodiment is not limited thereto.

[0064] This embodiment first performs a preliminary screening by correlation, and then performs a secondary screening by weight to recall target operation records with high correlation and high weight. In this way, the candidate set of operation records can be quickly narrowed down by weight sorting, without calculating the correlation between the full amount of operation records and the input text, thereby improving the recall efficiency.

[0065] In practical applications, the execution order of correlation calculation and weight sorting can be determined according to the scene of the input text. For example, when the scene of the input text is within the predetermined scene range, it can be preliminarily screened by correlation first, and then secondary screened by weight. When the scene of the input text is outside the predetermined scene range, it can be preliminarily screened by weight first, and then secondary screened by correlation.

[0066] According to another embodiment of the present disclosure, after obtaining the input text, a reference operation record is determined from a plurality of operation records in a database according to the input text, wherein the reference operation record and the input text are used for tasks of the same category. Then, a target style is determined from a plurality of predetermined styles according to the reference operation record. Then, target prompt information is determined according to the target style and a predetermined prompt information template.

[0067] For example, the subject wants to create a story, and the input text is "write a story". Correspondingly, the task is story creation. The stories that the subject has edited before can be queried from the database, and the information can be used as a reference operation record. Multiple predetermined styles can also be pre-configured, such as "meticulous", "simple", "short", and "medium", and then the target style can be determined from the reference operation record. For example, the user has edited three stories before, and all three stories are relatively simple and have fewer words. The target style can be "simple" and "short", and then "simple" and "short" can be combined with the predetermined prompt information template to obtain the target prompt information. After obtaining the target prompt information, the large model is prompted based on the target prompt information to generate a response text for the input text.

[0068] In this embodiment, the target style is determined according to the user's historical operation records, and then a response text is generated based on the target style, which can make the response text more in line with the user's personal style, improve the user experience, and improve the accuracy of the response.

[0069] Figure 3 It is a schematic diagram of a large model-based response method according to an embodiment of the present disclosure.

[0070] In this embodiment, after obtaining the user's input text 301, the input text 301 can be parsed to obtain the attribute 302 of the input text 301, and the attribute 302 can include the characteristics of the input text 301, the entity information in the input text 301, and the intention of the input text 301. Then, according to the correlation between the attribute 302 and the multiple operation records 303 in the database, and the weights of the multiple operation records 303, the target operation record 304 is recalled from the database.

[0071] In addition, after obtaining the user's input text 301, the task category 305 can be determined based on the input text 301, and then a reference operation record 306 for the same category of task is determined from the multiple operation records 303 in the database. Then, a target style 307 is determined based on the reference operation record 306. Then, the target style 307 and the predetermined prompt information template 308 are combined into a target prompt information template 309.

[0072] Next, the input text 301 , the target operation record 304 and the target prompt information template 309 may be combined into input information and input into the large model 310 , and the response text 311 for the input text 301 may be obtained using the large model 310 .

[0073] Figure 4 It is a schematic diagram of a large model-based response method according to an embodiment of the present disclosure.

[0074] In this embodiment, the system includes an operation record parsing module 410 , an input text parsing module 420 , a recall module 430 , an orchestration engine 440 and a basic capability module 450 .

[0075] The operation record parsing module 410 may include a scene classification submodule 411, a key information extraction submodule 412, and a structured content representation submodule 413. The scene classification submodule 411 is used to divide the scenes of the operation record according to the information source, type, etc. For example, email and web browsing are different scenes. The key information extraction submodule 412 can determine whether to extract key information from the operation record according to the scene of the operation record. The structured content representation submodule 413 is used to convert the data into a pre-configured format and interact with the large model through a standardized format so that the large model can better process the data.

[0076] The input text parsing module 420 may include a feature extraction submodule 421, a named entity recognition submodule 422, and an intention analysis submodule 413. The feature extraction submodule 421 may perform feature extraction operations on the input text. The named entity recognition submodule 422 may determine entity information contained in the input text. The intention analysis submodule 413 may determine the intention of the input text.

[0077] The recall module 430 may include a weight sorting submodule 431 and a relevance calculation submodule 432 , which may determine the execution order of weight sorting and relevance calculation according to the scenario, and recall the target operation record from multiple operation records in the determined order.

[0078] It should be noted that the weight of the operation record is determined based on the operation behavior associated with the operation record and the number of times the operation record has been recalled in history. The weight of the operation record can be increased when the operation record meets the conditions of active recall, spaced repetition, and operation reinforcement. Among them, active recall means that after the user provides input text many times, an operation record is repeatedly recalled, then the weight of the operation record can be increased. Spaced repetition means that the user performs the same operation behavior at regular intervals, and these operation behaviors are associated with the same operation record, then the weight of the operation records associated with these operation behaviors can be increased. Operation reinforcement means that the user takes target category operation behaviors such as editing and copying for a certain operation record, then the weight of the operation record can be increased.

[0079] The orchestration engine 440 may include a plug-in management submodule 441, a prompt information management submodule 442, and an execution framework 443. For example, some functions are managed in the form of plug-ins, and the plug-in management submodule 441 is used to manage these functions. For example, the data processing process related to the large model requires the use of prompt information (Prompt), such as extracting key information through the large model and generating a response text using the large model. The prompt information management submodule 442 is used to manage the prompt information used in the process of calling the large model. The execution framework 443 is used to manage the call link of the large model and the prompt information used in the process of calling the large model.

[0080] The basic capability module 450 may include the following underlying models, such as a large model 451 , a fine-tuning model 452 , and an encoding network 453 .

[0081] The system in this embodiment can record the user's operation behavior and operation records. In addition, it can provide response information to the user according to the scenario, thereby improving work efficiency and personalized service level.

[0082] This solution can be applied to various environments that require efficient information management and processing, such as internal knowledge management systems, personal data assistants, academic research support tools, etc. It is especially suitable for users who need to quickly access and process a large amount of dynamic information, such as researchers, journalists, lawyers, etc. For example, based on the user's historical operation behavior and operation records, it can provide relevant data and reference materials when the user writes a report, or provide industry dynamics when conducting market analysis.

[0083] Figure 5 It is a schematic structural block diagram of a large model-based response device according to an embodiment of the present disclosure.

[0084] like Figure 5 As shown, the large model-based response device 500 may include a recall module 510 and an input module 520 .

[0085] The recall module 510 is used to recall at least one target operation record from the database according to the correlation between the input text of the object and multiple operation records in the database, and the weights of the multiple operation records. The operation record is generated based on the operation behavior of the object, and the weight is determined based on the operation behavior associated with the operation record and the number of times the operation record has been recalled in history.

[0086] The input module 520 is used to input input text, at least one target operation record and target prompt information into the large model to obtain a response text for the input text.

[0087] According to another embodiment of the present disclosure, the weight satisfies at least one of the following conditions: the weight is positively correlated with the number of times the operation record is recalled within a historical predetermined period; the weight is positively correlated with the number of associated operation behaviors; the weight is negatively correlated with the interval duration of the associated operation behaviors, and the interval duration is the duration between the moment when the operation behavior is generated and the target moment; and the weight is positively correlated with the number of target category operation behaviors contained in the associated operation behavior.

[0088] According to another embodiment of the present disclosure, it also includes: a reference operation record determination module, a style determination module and a target prompt information determination module. The reference operation record determination module is used to determine a reference operation record from a plurality of operation records in a database according to an input text, wherein the reference operation record and the input text are used for tasks of the same category. The style determination module is used to determine a target style from a plurality of predetermined styles according to the reference operation record. The target prompt information determination module is used to determine the target prompt information according to the target style and a predetermined prompt information template.

[0089] According to another embodiment of the present disclosure, the operation record is determined by the following modules: a determination module, a first adding module, and a second adding module. The determination module is used to determine whether the initial record meets a predetermined condition based on the scenario of the operation behavior associated with the operation record, the number of characters in the initial record, and the weight of the initial record. The first adding module is used to extract key information from the initial record in response to determining that the initial record meets the predetermined condition, and add the extracted key information to the operation record. The second adding module is used to add the initial record to the operation record in response to determining that the initial record does not meet the predetermined condition.

[0090] According to another embodiment of the present disclosure, the predetermined condition includes at least one of the following: the scene is a target scene; the number of characters in the initial record is greater than or equal to a predetermined number; the weight of the initial record is less than or equal to a predetermined weight.

[0091] According to another embodiment of the present disclosure, the recall module includes: an attribute determination submodule, a first set determination submodule and a first determination submodule. The attribute determination submodule is used to determine at least one attribute of the input text. The first set determination submodule is used to determine a first operation record set from the database according to the similarity between the attribute and multiple operation records in the database for each attribute in the at least one attribute. The first determination submodule is used to determine at least one target operation record from the first operation record set according to the weight of each operation record in the first operation record set.

[0092] According to another embodiment of the present disclosure, the recall module includes: an attribute determination submodule, a second set determination submodule, and a second determination submodule. The attribute determination submodule is used to determine at least one attribute of the input text. The second set determination submodule is used to determine a second operation record set from the database according to the respective weights of a plurality of operation records in the database. The second determination submodule is used to determine at least one target operation record from the second operation record set according to the similarity between the attribute and a plurality of operation records in the second operation record set for each attribute in the at least one attribute.

[0093] According to another embodiment of the present disclosure, the attribute includes at least one of the following: a feature of the input text, entity information in the input text, and an intention of the input text.

[0094] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, including at least one processor; and a memory communicatively connected to the at least one processor; 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 the above-mentioned large model-based response method.

[0095] According to an embodiment of the present disclosure, the present disclosure also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the above-mentioned large model-based response method.

[0096] According to an embodiment of the present disclosure, the present disclosure also provides a computer program product, including a computer program, which implements the above-mentioned large model-based response method when executed by a processor.

[0097] Figure 6 It is a block diagram of an electronic device for implementing the large model-based answering method of an embodiment of the present disclosure. 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.

[0098] like Figure 6As shown, the device 600 includes a computing unit 601, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the device 600 can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0099] A number of components in the device 600 are connected to the I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a disk, an optical disk, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows the device 600 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0100] The computing unit 601 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 601 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 601 performs the various methods and processes described above, such as a large model-based response method. For example, in some embodiments, the large model-based response method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded into the RAM 603 and executed by the computing unit 601, one or more steps of the large model-based response method described above may be performed. Alternatively, in other embodiments, the computing unit 601 may be configured to perform a large model-based response method in any other appropriate manner (e.g., by means of firmware).

[0101] 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.

[0102] 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.

[0103] 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 device, 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.

[0104] 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).

[0105] 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.

[0106] A computer system may include clients and servers. Clients and servers are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship to each other.

[0107] 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.

[0108] 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. A response method based on a large model, comprising: Recalling at least one target operation record from the database according to the correlation between the input text of the object and a plurality of operation records in the database, and the weights of the plurality of operation records; as well as Inputting the input text, the at least one target operation record and the target prompt information into the macro model to obtain a response text for the input text; The operation record is generated based on the operation behavior of the object, and the weight is determined based on the operation behavior associated with the operation record and the number of times the operation record has been recalled in history.

2. The method according to claim 1, wherein: The weight satisfies at least one of the following conditions: The weight is positively correlated with the number of times the operation record is recalled within a historical predetermined period; The weight is positively correlated with the number of associated operation behaviors; The weight is negatively correlated with the interval duration of the associated operation behavior, and the interval duration is the duration between the moment when the operation behavior is generated and the target moment; as well as The weight is positively correlated with the number of target category operation behaviors included in the associated operation behavior.

3. The method according to claim 1, further comprising: Determine, according to the input text, a reference operation record from the plurality of operation records in the database, wherein the reference operation record and the input text are used for tasks of the same category; Determining a target style from a plurality of predetermined styles according to the reference operation record; and The target prompt information is determined according to the target style and a predetermined prompt information template.

4. The method according to claim 1, wherein: The operation record is determined in the following manner: Determining whether the initial record meets a predetermined condition according to at least one of a scenario of an operation behavior associated with the operation record, the number of characters in the initial record, and a weight of the initial record; In response to determining that the initial record satisfies the predetermined condition, extracting key information from the initial record, and adding the extracted key information to the operation record; as well as In response to determining that the initial record does not satisfy the predetermined condition, the initial record is added to the operation record.

5. The method according to claim 4, wherein: The predetermined condition includes at least one of the following: The scene is a target scene; The number of characters in the initial record is greater than or equal to a predetermined number; The weight of the initial record is less than or equal to a predetermined weight.

6. The method according to claim 1, wherein: The step of recalling at least one target operation record from the database according to the correlation between the input text of the object and the plurality of operation records in the database, and the weights of the plurality of operation records, comprises: determining at least one attribute of the input text; For each attribute of the at least one attribute, determining a first operation record set from the database according to a similarity between the attribute and the plurality of operation records in the database; and At least one target operation record is determined from the first operation record set according to the weight of each operation record in the first operation record set.

7. The method according to claim 1, wherein: The step of recalling at least one target operation record from the database according to the correlation between the input text of the object and the plurality of operation records in the database, and the weights of the plurality of operation records, comprises: determining at least one attribute of the input text; determining a second operation record set from the database according to respective weights of the plurality of operation records in the database; and For each of the at least one attribute, at least one target operation record is determined from the second operation record set according to a similarity between the attribute and a plurality of operation records in the second operation record set.

8. The method according to any one of claims 6 to 7, wherein: The attribute includes at least one of the following: a feature of the input text, entity information in the input text, and an intention of the input text.

9. A response device based on a large model, comprising: A recall module, configured to recall at least one target operation record from the database according to the correlation between the input text of the object and the multiple operation records in the database, and the weights of the multiple operation records; as well as An input module, used for inputting the input text, the at least one target operation record and the target prompt information into the macro model to obtain a response text for the input text; The operation record is generated based on the operation behavior of the object, and the weight is determined based on the operation behavior associated with the operation record and the number of times the operation record has been recalled in history.

10. The device according to claim 9, wherein: The weight satisfies at least one of the following conditions: The weight is positively correlated with the number of times the operation record is recalled within a historical predetermined period; The weight is positively correlated with the number of associated operation behaviors; The weight is negatively correlated with the interval duration of the associated operation behavior, and the interval duration is the duration between the moment when the operation behavior is generated and the target moment; as well as The weight is positively correlated with the number of target category operation behaviors included in the associated operation behavior.

11. The apparatus according to claim 9, further comprising: a reference operation record determination module, configured to determine a reference operation record from the plurality of operation records in the database according to the input text, wherein the reference operation record and the input text are used for tasks of the same category; a style determination module, configured to determine a target style from a plurality of predetermined styles according to the reference operation record; and The target prompt information determination module is used to determine the target prompt information according to the target style and a predetermined prompt information template.

12. The device according to claim 9, wherein: The operation record is determined by the following modules: a determination module, configured to determine whether the initial record satisfies a predetermined condition according to at least one of a scenario of an operation behavior associated with the operation record, the number of characters in the initial record, and a weight of the initial record; A first adding module, configured to extract key information from the initial record in response to determining that the initial record satisfies the predetermined condition, and add the extracted key information to the operation record; as well as The second adding module is configured to add the initial record to the operation record in response to determining that the initial record does not meet the predetermined condition.

13. The device according to claim 12, wherein: The predetermined condition includes at least one of the following: The scene is a target scene; The number of characters in the initial record is greater than or equal to a predetermined number; The weight of the initial record is less than or equal to a predetermined weight.

14. The device according to claim 9, wherein: The recall module includes: An attribute determination submodule, used to determine at least one attribute of the input text; A first set determination submodule, configured to determine, for each attribute of at least one attribute, a first operation record set from the database according to a similarity between the attribute and the plurality of operation records in the database; and The first determining submodule is configured to determine at least one target operation record from the first operation record set according to a weight of each operation record in the first operation record set.

15. The device according to claim 9, wherein: The recall module includes: An attribute determination submodule, used to determine at least one attribute of the input text; A second set determination submodule, configured to determine a second operation record set from the database according to respective weights of the plurality of operation records in the database; and The second determination submodule is configured to determine, for each attribute of the at least one attribute, at least one target operation record from the second operation record set according to a similarity between the attribute and a plurality of operation records in the second operation record set.

16. The device according to any one of claims 14 to 15, wherein: The attribute includes at least one of the following: a feature of the input text, entity information in the input text, and an intention of the input text.

17. 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 that can be executed 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 8.

18. 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 8.

19. 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 8.

Images