Interaction processing method and electronic equipment

Abstract

The invention discloses an interaction processing method and an electronic device, and the method comprises the steps: responding to the obtained target input data, retrieving first memory data and second memory data matched with the target input data from a target knowledge base, the target input data being data input to a target application, and the second memory data being data input to the target application; the target application is an application capable of providing an artificial intelligence service or calling at least one processing model to provide the artificial intelligence service; and performing generation processing on the target input data by referring to the first memory data and the second memory data so as to obtain a target response result matched with a target user intention, wherein the configuration parameters of the first memory data and the second memory data in the target knowledge base are different, and the target user intention is determined at least based on the target input data.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to an interactive processing method and electronic device. Background Art

[0002] In AI applications, or those capable of invoking AI processing, the ability to engage in personalized conversations with users is essential. This personalized conversation is based on historical user interactions and user information. Specifically, during a conversation with a user, the historical memory data most relevant to the conversation context can be retrieved in real time. Based on this historical memory data, the large language model used for interaction can generate more personalized, targeted, and consistent responses. Summary of the Invention

[0003] In view of this, this application provides the following technical solutions:

[0004] The first aspect of the present application provides an interactive processing method, comprising:

[0005] In response to obtaining target input data, retrieving first memory data and second memory data that match the target input data from a target knowledge base, wherein the target input data is data input to a target application, and the target application is an application capable of providing an artificial intelligence service or calling at least one processing model to provide an artificial intelligence service; and

[0006] generating and processing the target input data with reference to the first memory data and the second memory data to obtain a target response result that matches the target user's intention;

[0007] The first memory data and the second memory data have different configuration parameters in the target knowledge base, and the target user intention is determined based at least on the target input data.

[0008] In one possible implementation, retrieving first memory data and second memory data matching the target input data from a target knowledge base includes:

[0009] Performing vectorization processing on the target input data to obtain first vectorized data;

[0010] Retrieving different data sources from the target knowledge base using the first vectorized data to obtain the first memory data and the second memory data;

[0011] Among them, information of at least one data field is different between different data sources, and the retrieval methods for searching different data sources are the same or different.

[0012] In one possible implementation, using the first vectorized data to retrieve different data sources from the target knowledge base to obtain the first memory data and the second memory data includes:

[0013] Retrieving a first data source from the target knowledge base using the first vectorized data in a first retrieval manner to obtain first memory data, where the first data source is a data set obtained by performing a first processing on first user data of the target user acquired within a first time period;

[0014] Retrieving a second data source from the target knowledge base using the first vectorized data in a second retrieval manner to obtain second memory data, where the second data source is a data set obtained by performing a second processing on second user data of the target user obtained within a second time period, or the second data source is a data set obtained by performing a third processing on the first data source;

[0015] The first retrieval method and the second retrieval method are the same or different, the first time period is later than the second time period, and the first processing is different from the second processing.

[0016] In one possible implementation, performing first processing on the acquired user data of the target user within the first time period to obtain the first data source includes:

[0017] Performing vectorization processing on the acquired first user data of the target user within the first time period to obtain multiple groups of second vectorized data;

[0018] performing a first clustering process on the plurality of groups of second vectorized data to obtain a plurality of first vectorized data sets;

[0019] performing a first summary generation process on the plurality of first vectorized data sets to obtain a first summary data set;

[0020] The first summary data set and the first user data are extracted and processed according to a first data structure to obtain the first data source.

[0021] In one possible implementation, the method further includes at least one of the following:

[0022] After obtaining the plurality of sets of second vectorized data, extracting and processing the plurality of sets of second vectorized data according to the first data structure to obtain the first data source;

[0023] After obtaining the first summary data set, extracting and processing the first summary data set and the plurality of sets of second vectorized data according to the first data structure to obtain the first data source;

[0024] The event data obtained by the extraction process is merged and / or deduplicated to obtain the first data source.

[0025] In one possible implementation, performing third processing on the first data source to obtain the second data source includes:

[0026] Performing a second clustering process on the first data source to obtain a plurality of second vectorized data sets;

[0027] performing a second summary generation process on the plurality of second vectorized data sets to obtain a second summary data set;

[0028] The second summary data set and the first data source are subjected to extraction processing or third summary generation processing according to a second data structure to obtain the second data source.

[0029] In one possible implementation, at least one of the following is further included:

[0030] processing the first data source into a second data source at a first time interval;

[0031] Merging and / or deduplicating the second data source based on the first data source;

[0032] Call different processing models to perform vectorized processing on corresponding types of user data;

[0033] Associating the user data or vectorized data corresponding to the first data source and the second data source.

[0034] In one possible implementation, the generating process of the target input data with reference to the first memory data and the second memory data includes at least one of the following:

[0035] Using the first memory data and the second memory data as knowledge base data of a target processing model, and using the target processing model to call the knowledge base data to generate and process the target input data;

[0036] Using the first memory data and the second memory data to update the prompt word corresponding to the target input data, and using the target processing model to generate and process the updated prompt word data;

[0037] Optimizing the user intent corresponding to the target input data using the first memory data and the second memory data, and generating and processing the optimized target user intent using a target processing model;

[0038] Performing a model parameter adjustment process on a target processing model using the first memory data and the second memory data, so as to generate the target input data using the adjusted processing model;

[0039] The weight of the target data in the target input data is adjusted using the first memory data and the second memory data, and the adjusted input data is generated and processed using a target processing model.

[0040] In one possible implementation, the generating process of the target input data with reference to the first memory data and the second memory data includes at least one of the following:

[0041] generating third memory data based on weight parameters corresponding to the first memory data and the second memory data respectively;

[0042] Based on the third memory data, the target processing model is called to generate and process the target input data to obtain the target response result; or

[0043] The response result obtained by processing the target input data by the target processing model is optimized based on the third memory data to obtain the target response result.

[0044] A second aspect of the present application provides an electronic device, including at least one processor and at least one processing model capable of running on the processor, wherein the processing model can be called by a target application to perform at least one of the following:

[0045] In response to obtaining target input data, retrieving first memory data and second memory data that match the target input data from a target knowledge base, wherein the target input data is data input to a target application, and the target application is an application capable of providing an artificial intelligence service or calling at least one processing model to provide an artificial intelligence service; and

[0046] generating and processing the target input data with reference to the first memory data and the second memory data to obtain a target response result that matches the target user's intention;

[0047] The first memory data and the second memory data have different configuration parameters in the target knowledge base, and the target user intention is determined based at least on the target input data. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0049] Figure 1 A flowchart of an interactive processing method disclosed in an embodiment of the present application;

[0050] Figure 2 A schematic diagram of the overall implementation architecture of the interactive processing solution disclosed in this application;

[0051] Figure 3 A schematic diagram of the data processing flow of the target knowledge base disclosed in the embodiments of this application;

[0052] Figure 4 A flowchart of retrieving matching memory data disclosed in an embodiment of the present application;

[0053] Figure 5 A schematic diagram of the process of recalling recent specific event data and long-term important data disclosed in the embodiments of this application;

[0054] Figure 6 A schematic diagram of the structure of an interactive processing device disclosed in an embodiment of the present application;

[0055] Figure 7 This is a schematic structural diagram of an electronic device disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0057] The embodiments of the present application can be applied to electronic devices. The present application does not limit the product form of the electronic device, which can include but is not limited to smartphones, tablet computers, wearable devices, personal computers (PCs), netbooks, etc., and can be selected according to application requirements.

[0058] Figure 1 This is a flow chart of an interactive processing method disclosed in an embodiment of the present application. Figure 1 As shown, the interaction processing method may include:

[0059] Step 101: In response to obtaining target input data, first memory data and second memory data matching the target input data are retrieved from a target knowledge base, where the target input data is data input to a target application, and the target application is an application capable of providing artificial intelligence services or calling at least one processing model to provide artificial intelligence services.

[0060] The target application can be an intelligent agent application, an application capable of invoking an AI model or intelligent agent application, such as Lenovo's intelligent agent applications Lenovo Xiaotian or AINOW, or an application capable of invoking an AI model, such as the WeChat application, Lenovo's image processing application CreatorZone, a browser application, a chatbot application, an image application, etc. The target input data can be interactive data input into the target application, such as conversation data, control action data, instruction data, etc., input into the target application's dialog box or data box. The data type of the target input data is not fixed and can be, but is not limited to, multimodal data such as voice data, string data, image data, and gesture interaction data.

[0061] The target knowledge base can be a local knowledge base or a cloud-based knowledge base, specifically a personal knowledge base, a family knowledge base, a teamwork knowledge base, or the knowledge base of an organization such as a company or school. The target knowledge base stores the memory data of one or more users with access to the knowledge base. This memory data may include text, voice, images, audio, video, and other content generated by users in their daily lives and work.

[0062] The target knowledge base includes at least first memory data and second memory data, wherein the first memory data and the second memory data have different configuration parameters in the target knowledge base. The configuration parameters may include, but are not limited to, tag information such as the generation time, location, and source of the memory data or the corresponding event.

[0063] For example, the configuration parameters may specifically include at least one of the following: storage location, divided time interval range, storage form, importance level or priority, etc. In one example, the memory data can be processed and stored separately according to the length of its generation time or storage time. For example, the first memory data can be specific event data with a first correlation with the target input data within a first historical period and the original memory data associated with the specific event data. The second memory data can be important event data with a second correlation with the target input data outside the first historical period and the original memory data associated with the important event data. Among them, the specific event data is obtained based on the processing of the original memory data, and the important event data is obtained based on the processing of the specific event data. The types and corresponding contents of the memory data contained in the target knowledge base will be introduced in detail in the embodiments below, and will not be explained in detail here.

[0064] The first memory data and the second memory data that match the target input data can be memory data whose similarity with the target input data reaches a set threshold. The similarity can be determined based on the cosine similarity, vector distance, etc. of the two matching objects, which will not be introduced in detail here. Since the first memory data and the second memory data have a certain similarity with the target input data, they can be used as prompt word (corresponding to the target input data) optimization materials, or as a reference basis for model parameter adjustment, or as a database for retrieval enhancement.

[0065] Step 102: Generate and process the target input data with reference to the first memory data and the second memory data to obtain a target response result that matches the target user's intention.

[0066] After determining the first and second memorized data, target response data matching the target user's intent can be generated based on the target input data, using the first and second memorized data as references. The target user's intent is determined based at least on the target input data. For example, if the target input data is "generate an image of a puppy," the resulting target response is an image of "a puppy squatting on the grass." This target response only meets the basic requirements of the target input data.

[0067] Of course, in addition to the target input data, the target user intention can also be determined based on other data, such as the first memory data, the second memory data, the scene the user is in, the user's occupation data, etc. For example, the part about animals in the first memory data records that the color of the animals is white, and the user's occupation data is an advertising designer. When the target input data is also "generate an image of a puppy", the target response result finally generated is an image of "a white puppy with a solid white background"; the color of the puppy "white" is based on the memory data as a reference, and the solid white background is based on the user's occupation data as an advertising designer as a reference. Because advertising design images often require the addition of text content, the background will not be designed to be messy to affect font recognition and appearance; the aforementioned target response result not only meets the basic requirements of the target input data, but also ensures that the color of the generated puppy is the user's favorite color, and is convenient for text overlay design, which is more in line with the user's intention.

[0068] Alternatively, the target user intent can be determined based on other types of data, such as hot topics or current events related to the target input data. For example, if the target input data is "What traditional activities are there during the Dragon Boat Festival?", the model can automatically search for recent news about Dragon Boat Festival dragon boat races and other popular content online, and use this as a reference for generating the target response.

[0069] The target response result may be a conversational reply that matches the target input data, a multimedia data file, a text file, a control action instruction (such as an instruction to control device configuration parameter changes or instructions to control other devices), or a processing task (such as generating a task to be executed). For example, if the current scenario is a home scenario, the user may call an intelligent assistant and input the voice message "Student **, what's the indoor temperature? I feel a little hot." The corresponding target response result includes a voice reply and a control instruction; the voice reply content may be "The indoor temperature is now 26°C. The air conditioner has been turned on for you." The control instruction is an instruction to control the air conditioner to start and cool.

[0070] The interactive processing method described in this embodiment can, after obtaining the target input data, match the first memory data and the second memory data having a high similarity to the target data from the target knowledge base including memory data with different configuration parameters. Since the configuration parameters of the first memory data and the second memory data are different, based on the configuration parameters, the target input data and the memory data can be matched more accurately and quickly from different dimensions, thereby obtaining a target response result that is more in line with the target user's intention.

[0071] In order to better understand and implement the technical solution of this application, we first introduce the memory data in the target database.

[0072] Figure 2 This is a schematic diagram of the overall implementation architecture of the interactive processing solution disclosed in this application. Figure 2 As shown in the figure, the relevant content corresponding to some conversations, activities or device operations in the user's daily life or work process is used as the memory data source. These relevant contents can be text, voice pictures, audio, video and other contents, as well as the time, place, people and other data of obtaining these content files. These relevant contents are vectorized through models (including word embedding models, language models, voice models, visual / auditory models, etc.), and text descriptions or event summaries are extracted as raw memory data and stored in the target knowledge base (corresponding to Figure 2 vector database in .

[0073] Furthermore, content-related raw memory data can be clustered based on the vector distances between different raw memory data. A summary of the current cluster can be extracted from each cluster data. Specific event information can then be extracted from each cluster content summary and related raw memory data. This information is then stored as specific event data in the target knowledge base and associated with the raw memory data. During implementation, the extracted specific event data can also be merged and deduplicated to reduce data storage and improve subsequent retrieval efficiency.

[0074] Furthermore, content-related specific event data can be clustered based on the vector distance between each specific event data. A content summary of the current cluster can be extracted from each cluster data. Important event summaries can be extracted from each cluster content summary and the related specific event data. These important event summaries are stored as important event data in the target knowledge base and associated with the specific event data. Similarly, during implementation, the extracted important event data can be merged and deduplicated.

[0075] In the application scenario, the memory data can be divided into recent (short-term) data and long-term (long-term) data according to a certain time period. The recent data can be processed to obtain the corresponding specific event data, while the long-term data can be processed to obtain the corresponding important event data.

[0076] For example, taking half a year as the boundary, the original memory data within half a year is processed to obtain the corresponding specific event data, while the specific event data half a year ago is processed to obtain the important event data.

[0077] Combine Figure 2 The word embedding model is used to vectorize text; the speech recognition model is used to recognize user voice conversation content as text content and character labels; the visual model is used to vectorize images and videos and understand the content to generate text content descriptions; the auditory model is used to vectorize audio and understand the content to generate text content descriptions; the large language model is used to extract information from text data; and the event extraction prompt word is used to prompt the large language model to extract event format information, including events, locations, people, events, and other content.

[0078] Figure 3 This is a schematic diagram of the data processing flow of the target knowledge base disclosed in the embodiment of this application, which can be combined with Figure 3 The realization of the target knowledge base with a hierarchical storage structure formed by original memory data, specific event data and important event data can be understood from the content of the aforementioned embodiment. Since the memory data in the target knowledge base is a hierarchical storage structure, it is possible to highlight specific event data for recent memory data and important event data for long-term memory data, thereby highlighting the memory focus and making it easier to focus on data that is valuable to the target data information; in addition, the memory data in the target knowledge base in this application also includes clear key information such as time, place, person, and event, so that subsequent data recall is more targeted, which helps in accurate and efficient retrieval and matching.

[0079] Figure 4 This is a flowchart of retrieving matching memory data disclosed in the embodiment of this application. Figure 4 As shown, in the above embodiment, retrieving the first memory data and the second memory data that match the target input data from the target knowledge base may include:

[0080] Step 401: Perform vectorization processing on the target input data to obtain first vectorized data.

[0081] That is, the target input data is uniformly converted from its original data type to vector data. For target input data of textual data type, the first vectorized data can be directly converted based on the textual content. For target data of non-textual data type, such as speech data, image data, video data, etc., the target data can first be converted into textual content through speech recognition models, audio-visual models, word embedding models, etc., and then the converted textual content can be further converted into the first vectorized data.

[0082] Here, the target data is vectorized to obtain the first vectorized data in order to facilitate subsequent similarity comparison between the target data and the memory data in a quantitative manner.

[0083] Step 402: Utilize the first vectorized data to retrieve different data sources from the target knowledge base to obtain the first memory data and the second memory data.

[0084] The information in at least one data field differs between different data sources, and the retrieval methods for searching different data sources may be the same or different. The first memory data may be recent memory data or short-term memory data, and the second memory data may be remote memory data or long-term memory data. The information in at least one data field differs between different data sources, and the difference may be in the descriptive labels of recent memory data and remote memory data. Different data sources may include specific event data and important event data.

[0085] In a specific implementation, the first memory data may include specific event data and associated original memory data, and the second memory data may include important event data and associated original memory data.

[0086] The specific implementation of step 402 may include: using the first vectorized data to retrieve a first data source from the target knowledge base in a first retrieval method to obtain first memory data, where the first data source is a data set obtained by performing a first processing on the first user data of the target user obtained within a first time period; using the first vectorized data to retrieve a second data source from the target knowledge base in a second retrieval method to obtain second memory data, where the second data source is a data set obtained by performing a second processing on the second user data of the target user obtained within a second time period, or the second data source is a data set obtained by performing a third processing on the first data source.

[0087] The first search method and the second search method are the same or different, and the first search method includes but is not limited to semantic vector search, keyword vector search, tag vector search, etc. The first time period is later than the second time period, and the first processing is different from the second processing.

[0088] The first data source may be the aforementioned specific event data, corresponding to recent memory data; the second data source may be the aforementioned important event data, corresponding to long-term memory data. In implementation, the dividing line between recent and long-term (the first time period) may be configured by default, or may be set according to the actual application scenario. For example, the first time period may be within half a year or within a year. The first user data within the first time period may include data of the user at home, work, travel and leisure, including device operation data, social data, user-participated or related activity data or events, or social relationship data (family relationships, work relationships, etc.), corresponding Figure 2 Correspondingly, the second time period may be half a year ago or a year ago, and the second user data may be the same as or different from the first user data. Figure 2 The memory data source in .

[0089] Figure 5 This is a flow chart of recalling recent specific event data and long-term important data disclosed in the embodiment of this application, wherein the recall of recent specific events and the recall of long-term important data can be performed simultaneously to ensure data processing efficiency. Figure 5 Understand the implementation of the interactive processing solution of this application solution.

[0090] The first processing can include extracting text descriptions based on fields such as time, location, person, and event, and associating them with the original data units. The specific implementation of this process is as described above, obtaining a description of the content of the specific event data, including clustering the original memory data, extracting summaries from each cluster, and extracting specific event information from the summaries and the original memory data. The second processing includes clustering the original memory data, extracting summaries from each cluster, clustering again based on the summaries, and extracting the summaries again.

[0091] Specifically, the first processing of the user data of the target user obtained within the first time period to obtain the first data source may include: performing vectorization processing on the first user data of the target user obtained within the first time period to obtain multiple groups of second vectorized data; performing first clustering processing on the multiple groups of second vectorized data to obtain multiple first vectorized data sets; performing first summary generation processing on the multiple first vectorized data sets to obtain a first summary data set; and extracting and processing the first summary data set and the first user data according to a first data structure to obtain the first data source.

[0092] The first clustering process is performed on the multiple sets of second vectorized data to obtain multiple first vectorized data sets. Clustering can be performed based on the same vector distance, or based on different vector distances. That is, clustering can be performed by adaptively adjusting the vector distance based on event type or scene type. For example, clustering can be performed based on the first vector distance for learning event types, and based on the second vector distance for tourism event types.

[0093] The process of generating the first summary for the plurality of first vectorized data sets can be implemented using the same model as the clustering process, or using a different model. That is, the main body that executes the content of each part of the first process can be a large model, or each part of the first process can be processed separately by multiple different small models.

[0094] The first summary data set and the first user data are extracted and processed according to the first data structure. The specific implementation process can be to extract specific event information from each cluster content summary and related original memory data, store it as specific event data in the target knowledge base, and associate it with the original memory data.

[0095] In other implementations, the acquisition of the first data source may be implemented differently.

[0096] Another implementation of obtaining the first data source is: after obtaining multiple sets of second vectorized data, extracting and processing the multiple sets of second vectorized data according to the first data structure to obtain the first data source. That is, after obtaining the multiple sets of second vectorized data, clustering processing is not performed, and event information is directly extracted from the vectorized data to obtain the first data source.

[0097] Another implementation of obtaining the first data source is: after obtaining the first summary data set, the first summary data set and the multiple sets of second vectorized data are extracted and processed according to the first data structure to obtain the first data source. That is, after obtaining the first summary data set, event information is extracted not by combining it with the first user data, but by combining it with the vectorized data. The first data structure can be determined based on the prompt word input by the user (corresponding to the target input data) to determine the extraction event format information.

[0098] Another implementation of obtaining the first data source is to merge and / or deduplicate the event data obtained through the extraction process to obtain the first data source. Specifically, since the specific event data extracted from different clusters and the original memory data may contain duplicate content, the extracted event data can be merged and / or deduplicated to reduce the amount of data stored.

[0099] Specifically, performing a third processing on the first data source to obtain a second data source may include: performing a second clustering processing on the first data source to obtain multiple second vectorized data sets; performing a second summary generation processing on the multiple second vectorized data sets to obtain a second summary data set; performing extraction processing on the second summary data set and the first data source according to a second data structure or performing a third summary generation processing on the second summary data set and the first data source to obtain the second data source.

[0100] The above content also represents the specific implementation process of determining the data set obtained by performing the third processing on the first data source as the second data source. In this implementation, for the first time, content-related specific event data is clustered based on vector distance (second clustering is performed on the first data source), content summaries are extracted for each cluster, and key event summaries are extracted from each cluster content summary and the related specific event data. These are stored as key event data in the target knowledge base and associated with the specific event data.

[0101] Among them, the second summary data set and the first data source are extracted and processed according to the second data structure or the third summary generation process is performed, which can be an important event summary extraction process, or the summary content of the important event is generated by setting important event tags or specific tags in the data structure.

[0102] In other implementations, the second data source may be obtained in different ways.

[0103] Another implementation of obtaining a second data source is to process the first data source into a second data source at a first time interval. The first time interval can be set based on actual application circumstances and is not fixed. For example, if the first time interval is three months, then during the application of the solution, as time passes, any first data source (incremental, long-term, specific event data) with an incremental storage time exceeding three months will be processed accordingly to become a second data source.

[0104] Another implementation of obtaining the second data source is to merge and / or deduplicate the second data source based on the first data source. This implementation primarily includes two steps: first, extracting incremental long-term specific event data to obtain incremental long-term important event data; and second, comparing the incremental long-term important event data with existing long-term important event data, performing merging and deduplication processing, and storing the merging and deduplication results as the second data source.

[0105] In other implementations, the processing of user data may also include: calling different processing models to perform vectorized processing on corresponding types of user data. Figure 2As shown, for text-based user data, the word embedding model can be called for vectorization processing; for voice-based user data, the speech recognition model can be called first to convert the voice content into text content, and then the word embedding model can be called for vectorization processing of the text content; for video-based user data, the visual model can be called for corresponding vectorization processing; for audio-based user data, the auditory model can be called for corresponding vectorization processing.

[0106] The above content introduces various implementations of different processing of user data and obtaining different data sources, but does not constitute a fixed restriction on the method of obtaining the data source.

[0107] In the aforementioned embodiment, the generation and processing of the target input data by referring to the first memory data and the second memory data may include: using the first memory data and the second memory data as knowledge base data of the target processing model, and using the target processing model to call the knowledge base data to generate and process the target input data.

[0108] Among them, the target processing model can be understood as Figure 2 The large language model in the example can extract the first memory data and the second memory data from the target knowledge base based on the prompt word after obtaining the prompt word (target input data), and generate a target response result with reference to the first memory data and the second memory data.

[0109] In another implementation, the target input data is generated and processed with reference to the first memory data and the second memory data, including: using the first memory data and the second memory data to update the prompt word corresponding to the target input data, and using the target processing model to generate and process the updated prompt word data.

[0110] For example, if the target input data is the voice data "Please explain the necessity of health care", the prompt words recognized by the large language model may include "sword", "necessity", etc., but there are many "health care"-related contents that are homophonic with "sword" in the first memory data and the first memory data, and these contents mostly involve health maintenance, enhancing immunity, etc. Therefore, it can be determined that the prompt word corresponding to the target input data is not "sword", but "health care", and the prompt word can be updated from "sword" to "health care".

[0111] In another implementation, the target input data is generated and processed with reference to the first memory data and the second memory data, including: optimizing the user intention corresponding to the target input data using the first memory data and the second memory data, and generating and processing the optimized target user intention using the target processing model.

[0112] For example, the target input data is "Please introduce the multimodal model". Based on this target input data, it is determined that the user's intention is "to understand the concept of the multimodal model". However, after obtaining the first memory data and the second memory data, it is found that they include many introductions to the applications, functions, and scenarios of the multimodal model. Therefore, it can be determined that the user has a certain degree of knowledge reserve about the multimodal model. The user's true intention may not be to simply understand the basic concepts of the multimodal model, but to have an in-depth understanding of its implementation principles, thereby updating the user's intention to "introduce the multimodal model in depth."

[0113] In another implementation, the target input data is generated and processed with reference to the first memory data and the second memory data, including: using the first memory data and the second memory data to adjust the model parameters of the target processing model, so as to generate and process the target input data using the adjusted processing model.

[0114] For example, during the application of the target processing model, the first memory data and the second memory data are used to determine that the user's historical memory data has personalized preferences for event records, such as conciseness, simplicity, etc., and the model parameters of the target processing model can be adjusted accordingly so that it can generate target response results according to user preferences in the subsequent processing process.

[0115] In another implementation, the target input data is generated and processed with reference to the first memory data and the second memory data, including: using the first memory data and the second memory data to adjust the weight of the target data in the target input data, and using the target processing model to generate and process the adjusted input data.

[0116] For example, the target input data is "generate an image with a cat and a dog", where the target data includes "cat" and "dog", and both have the same weight of 0.5; while in the first memory data and the second memory data, the number of images including cats accounts for a large proportion, and in images with both cats and dogs, cats are often foreground or near-field objects, so the weights of the target data "cat" and "dog" can be adjusted, such as the weight of "cat" is 0.7 and the weight of "dog" is 0.3.

[0117] In another implementation, the target input data is generated and processed with reference to the first memory data and the second memory data, including: generating third memory data based on weight parameters corresponding to the first memory data and the second memory data respectively.

[0118] In this implementation, the first memory data corresponds to recent data and has a larger weight, such as 0.7, while the second memory data corresponds to long-term data and has a smaller weight, such as 0.3. The two reference data (the first memory data and the second memory data) can be merged and the merged third memory data can be used to assist the target processing model in generating the target data. The merging process can be a weighted process of the first memory data and the second memory data.

[0119] On the basis of the above-mentioned implementation, the target input data is generated and processed with reference to the first memory data and the second memory data, further including: calling the target processing model based on the third memory data to generate and process the target input data to obtain the target response result; or, optimizing the response result obtained by processing the target input data by the target processing model based on the third memory data to obtain the target response result.

[0120] After obtaining the third memory data, the third memory data can be used only as reference data for generating the target response result, or the response result can still be generated with reference to the first memory data and the second memory data, and the response result can be further optimized and updated based on the third memory data to obtain the target response result.

[0121] The above content details various different implementations of generating and processing the target input data with reference to the first memory data and the second memory data, which facilitates technical personnel in the field to better understand and implement the technical solution of the present application.

[0122] For the sake of simplicity, the aforementioned method embodiments are described as a series of action combinations. However, those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application.

[0123] The above embodiments disclosed in the present application describe the method in detail. The method of the present application can be implemented using various devices. Therefore, the present application also discloses a device, and a specific embodiment is given below for detailed description.

[0124] Figure 6 This is a schematic diagram of the structure of an interactive processing device disclosed in an embodiment of the present application. Figure 6 As shown, the interaction processing device 60 may include:

[0125] A data retrieval module 601 is configured to retrieve, in response to obtaining target input data, first memory data and second memory data that match the target input data from a target knowledge base, wherein the target input data is data input to a target application, and the target application is an application capable of providing an artificial intelligence service or invoking at least one processing model to provide an artificial intelligence service;

[0126] The response processing module 602 is used to generate and process the target input data with reference to the first memory data and the second memory data to obtain a target response result that matches the target user's intention.

[0127] The first memory data and the second memory data have different configuration parameters in the target knowledge base, and the target user intention is determined based at least on the target input data.

[0128] The interactive processing device described in this embodiment can, after obtaining the target input data, match the first memory data and the second memory data that have a high similarity to the target data from the target knowledge base including memory data with different configuration parameters. Since the configuration parameters of the first memory data and the second memory data are different, based on the configuration parameters, the target input data and the memory data can be matched more accurately and quickly from different dimensions, thereby obtaining a target response result that is more in line with the target user's intention.

[0129] In one implementation, the data retrieval module includes: a vectorization processing module, used to perform vectorization processing on the target input data to obtain first vectorized data; a retrieval processing module, used to use the first vectorized data to retrieve different data sources from the target knowledge base to obtain the first memory data and the second memory data; wherein, the information of at least one data field between different data sources is different, and the retrieval methods for searching different data sources are the same or different.

[0130] In one implementation, the retrieval processing module can be specifically used to: use the first vectorized data to retrieve a first data source from the target knowledge base in a first retrieval method to obtain first memory data, where the first data source is a data set obtained by performing a first processing on the first user data of the target user obtained within a first time period; use the first vectorized data to retrieve a second data source from the target knowledge base in a second retrieval method to obtain second memory data, where the second data source is a data set obtained by performing a second processing on the second user data of the target user obtained within a second time period, or the second data source is a data set obtained by performing a third processing on the first data source; wherein the first retrieval method and the second retrieval method are the same or different, the first time period is later than the second time period, and the first processing is different from the second processing.

[0131] In one implementation, the device may further include a data source processing module, and the data source processing module may obtain the first data source by: performing vectorization processing on the first user data of the target user obtained within the first time period to obtain multiple groups of second vectorized data; performing first clustering processing on the multiple groups of second vectorized data to obtain multiple first vectorized data sets; performing first summary generation processing on the multiple first vectorized data sets to obtain a first summary data set; and extracting and processing the first summary data set and the first user data according to a first data structure to obtain the first data source.

[0132] In one implementation, the data source processing module can also be used for at least one of the following: after obtaining multiple groups of second vectorized data, extracting and processing the multiple groups of second vectorized data according to the first data structure to obtain the first data source; after obtaining the first summary data set, extracting and processing the first summary data set and the multiple groups of second vectorized data according to the first data structure to obtain the first data source; merging and / or deduplicating the event data obtained by the extraction and processing to obtain the first data source.

[0133] In one implementation, the device may further include a data source processing module, and the data source processing module may obtain the second data source by: performing a second clustering process on the first data source to obtain multiple second vectorized data sets; performing a second summary generation process on the multiple second vectorized data sets to obtain a second summary data set; performing extraction processing on the second summary data set and the first data source according to a second data structure or performing a third summary generation process to obtain the second data source.

[0134] In one implementation, the data source processing module can also be used for at least one of the following: processing the first data source into a second data source at a first time interval; merging and / or deduplicating the second data source based on the first data source; calling different processing models to vectorize corresponding types of user data; and associating user data or vectorized data corresponding to the first data source and the second data source.

[0135] In one implementation, the response processing module can be specifically used for at least one of the following: using the first memory data and the second memory data as the knowledge base data of the target processing model, and using the target processing model to call the knowledge base data to generate and process the target input data; using the first memory data and the second memory data to update the prompt word corresponding to the target input data, and using the target processing model to generate and process the updated prompt word data; using the first memory data and the second memory data to optimize the user intention corresponding to the target input data, and using the target processing model to generate and process the optimized target user intention; using the first memory data and the second memory data to adjust the model parameters of the target processing model, so as to use the adjusted processing model to generate and process the target input data; using the first memory data and the second memory data to adjust the weight of the target data in the target input data, and using the target processing model to generate and process the adjusted input data.

[0136] In one implementation, the response processing module can be specifically used for at least one of the following: generating third memory data based on the weight parameters corresponding to the first memory data and the second memory data respectively; calling the target processing model based on the third memory data to generate and process the target input data to obtain the target response result; or, optimizing the response result obtained by processing the target input data by the target processing model based on the third memory data to obtain the target response result.

[0137] The specific implementation of the interactive processing device and the modules it contains can be found in the corresponding content introduction in the method embodiment, which will not be repeated here.

[0138] Any of the interactive processing devices described in the above embodiments includes a processor and a memory. The data retrieval module, response processing module, vectorization processing module, retrieval processing module, data source processing module, etc. in the above embodiments are all stored in the memory as program modules, and the processor executes the above program modules stored in the memory to realize corresponding functions.

[0139] The processor contains a kernel, which retrieves the corresponding program module from the memory. There can be one or more kernels, and the kernel parameters can be adjusted to process the returned data.

[0140] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0141] In an exemplary embodiment, a computer-readable storage medium is also provided, which can be directly loaded into the internal memory of a computer and contains software code. After being loaded and executed by a computer, the computer program can implement the steps shown in any embodiment of the above-mentioned interactive processing method.

[0142] In an exemplary embodiment, a computer program product is also provided, which can be directly loaded into the internal memory of a computer and contains software code. After being loaded and executed by a computer, the computer program can implement the steps shown in any embodiment of the interactive processing method described above.

[0143] Furthermore, an embodiment of the present application provides an electronic device. Figure 7 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present application. Figure 7 As shown, the electronic device 70 includes at least one processor 701, at least one memory 702 connected to the processor, at least one processing model 703 capable of running on the processor, and a bus 704. The processor, memory, and processing model communicate with each other via the bus. The processing model can be called by a target application to perform at least one of the following:

[0144] In response to obtaining target input data, retrieving first memory data and second memory data that match the target input data from a target knowledge base, wherein the target input data is data input to a target application, and the target application is an application capable of providing an artificial intelligence service or calling at least one processing model to provide an artificial intelligence service; and

[0145] generating and processing the target input data with reference to the first memory data and the second memory data to obtain a target response result that matches the target user's intention;

[0146] The first memory data and the second memory data have different configuration parameters in the target knowledge base, and the target user intention is determined based at least on the target input data.

[0147] In another embodiment, the processing model in the electronic device can also be called to execute: in response to obtaining target input data, executing a corresponding retrieval operation in a target knowledge base using a target retrieval strategy.

[0148] The target search strategy is obtained by configuring a plurality of different search methods based on the target score data, and may include one or more search strategies. The configuration process may include adjusting the initial weight of each search method, configuring whether each search method performs search operations in parallel or serially, and selecting a search method or a combination of search methods.

[0149] In implementations where the target search strategy includes multiple search strategies, the multiple search strategies may be executed in parallel or sequentially, or one of the search strategies may be executed based on the weight configuration of each search strategy, such as executing the search strategy with the highest weight configuration. The search operations include, but are not limited to, at least one of tag search, keyword search, vector search, and knowledge graph search. The various search operations may be performed in parallel or serially, and this application does not impose any fixed restrictions on this.

[0150] In another embodiment, the processing model in the electronic device can also be called to execute: performing corresponding processing on the retrieval results obtained by the retrieval operation based on the target scoring data of the electronic device to obtain the target retrieval results.

[0151] Among them, the target rating data can be the hardware performance rating of the electronic device or the current resource usage rating, or the rating of the corresponding retrieval model (it can be understood that different retrieval methods correspond to different retrieval models, and the ratings obtained by different devices when running different models may be different).

[0152] The corresponding processing of the retrieval results obtained by the retrieval operation based on the target scoring data of the electronic device may include fusion processing, that is, deduplication and / or merging of the results retrieved by each retrieval method; or, deduplication and weighted sorting of candidate results of different strategies, and outputting the best or top-ranked result as the target output result; or, sorting and outputting all or the top two results; or, targeted partial output or full output.

[0153] In one implementation, performing corresponding retrieval operations in a target knowledge base using a target retrieval strategy may include: identifying the user intention represented by the target input data; obtaining target scoring data of the electronic device; determining the target weights corresponding to the multiple different retrieval methods based on the user intention and the target scoring data; and generating a target retrieval strategy for the target input data based on the target weights.

[0154] The user intent represented by the target input data can be determined by an intent recognition model, or by a large model or retrieval model deployed in the electronic device. The target scoring data can evaluate the hardware capabilities of different scenarios or configurations based on the device's hardware performance (CPU, GPU, or memory, etc.).

[0155] Among them, various search methods can include tag search, keyword search, vector search, knowledge graph search, etc., and the final target weight can be obtained after adjusting the initial weight. Specifically, the policy manager dynamically calculates the weights of tag, keyword, vector, and knowledge graph search based on the device's hardware performance (CPU, GPU, or memory, etc.) and user feedback history; if the device performance is insufficient or certain search capabilities are unavailable, the weights of the corresponding search capabilities will be adjusted.

[0156] The target search strategy for the target input data is generated based on the target weight. The adjusted weight can be used to determine the final search strategy. For example, if the device's hardware does not support knowledge graph search, the system will readjust its score to 0. At this time, the system will use other strategies, such as keyword search strategy and tag search strategy, to make up for the shortfall by adjusting their weights. At this time, the scores of other tag search, keyword search, and vector search are 0.1, 0.45, and 0.45 respectively.

[0157] In one implementation, determining the target weights corresponding to the multiple different retrieval methods based on the user intention and the target scoring data may include: configuring the initial weights corresponding to the multiple different retrieval methods supported by the electronic device based on the type of the user intention; adjusting the initial weights based on the target scoring data to obtain the target weights corresponding to each retrieval method.

[0158] Among them, the initial weights corresponding to various different retrieval methods are configured based on the type of user intention. The initial weight can be determined according to the type of input data. For example, the user first enters the query content, whether it is a natural sentence, keyword or label, the system will perform pre-processing such as word segmentation, stop word removal and synonym expansion, and then use the intent recognition module to determine whether the query is label-type, keyword-type or semantic-type, and identify whether it is related to the existing knowledge graph, and set initial weights for different retrieval strategies (such as label retrieval, keyword retrieval, vector retrieval, etc.).

[0159] When adjusting the initial weights, it is necessary to consider whether the current hardware capabilities and / or software model capabilities of the electronic device support all retrieval methods, which retrieval methods are not supported, or how many of each retrieval method can be supported, and then make corresponding adjustments.

[0160] Based on the foregoing content, correspondingly, the target retrieval strategy for the target input data generated based on the target weight may include: determining the target retrieval method used to perform the retrieval operation based on the target weight; when the target retrieval method is unique, performing the retrieval operation on the target input data with the unique target retrieval method; when the target retrieval method is not unique, performing multiple serial or parallel retrieval operations on the target input data with non-unique retrieval methods.

[0161] In implementation, if the device performance is insufficient or certain retrieval capabilities are unavailable, the weight of the corresponding retrieval capability can be adjusted. Alternatively, if the device hardware does not support knowledge graph retrieval, the system will reset its score to 0. In this case, the system will use other strategies, such as keyword retrieval strategy and tag retrieval strategy, to make up for the shortfall by adjusting their weights.

[0162] If the target search method is unique, the search can be performed based solely on that method, or the relevant strategies of other search methods can be partially integrated into the selected search method. If the target search method is not unique, it can be further determined whether to execute multiple search operations in parallel or serially, depending on the device resources (whether serial or parallel support is currently available) or user needs (whether efficiency and speed (parallel) or accuracy (serial) are considered).

[0163] In one implementation, determining the retrieval method used to perform the retrieval operation based on the target weight may include: adjusting the initial weights of each retrieval method in different proportions based on the target scoring data to obtain the target weight, wherein the proportion is related to the retrieval method adapted to the current performance of the electronic device; determining a retrieval method having a target weight greater than a first threshold as the retrieval method used to perform the retrieval operation; and / or, when the target weight of the first retrieval method among the multiple different retrieval methods is adjusted to 0, using the initial weight of the first retrieval method to adjust the weight of at least one of the remaining retrieval methods again.

[0164] When a device lacks a certain retrieval capability (such as tag retrieval), it compensates for the lack of capability through keyword mapping or semantic conversion mechanisms. For example, if the device's hardware does not support knowledge graph retrieval, the system will reset its score to 0. At this time, the system will use other strategies, such as keyword retrieval strategy and tag retrieval strategy, to make up for the shortfall by adjusting their weights.

[0165] In one implementation, executing corresponding retrieval operations in a target knowledge base using a target retrieval strategy may include at least one of the following: executing multiple retrieval operations in parallel or serially on the target input data based on multiple different retrieval methods supported by the electronic device; configuring corresponding initial weights for the multiple different retrieval methods based on target feature data of the target input data, and executing multiple serial retrieval operations on the target input data based on the priority corresponding to each retrieval method, wherein the priority is related to the initial weight; configuring different target weights for the multiple different retrieval methods based on the target feature data and the target scoring data of the target input data, so as to determine to execute the retrieval operation in at least one of the retrieval methods based on the target weight.

[0166] Based on the various different search methods supported by the electronic device, multiple search operations are performed in parallel or serially on the target input data, that is, each search method is executed, but the target weight is used to perform screening and processing when the final search result is output.

[0167] The target feature data of the target input data is used to configure corresponding initial weights for the various search methods. For example, the initial weights of the various search methods are determined directly based on specific features of the input data, such as labels, types, key content, sentence patterns (knowledge graphs), etc. The target feature data here does not refer to the intent type after intent recognition.

[0168] In one implementation, the corresponding processing of the retrieval results obtained by the retrieval operation based on the target scoring data of the electronic device may include: determining the target weight corresponding to each retrieval method based on the target scoring data of the electronic device; fusing the retrieval results obtained by each retrieval method based on the target weight to output the target retrieval result; or, weighted sorting the retrieval results obtained by each retrieval method based on the target weight, and taking the retrieval results in the target sequence as the target retrieval result.

[0169] The fusion processing can be understood as a merging process, and can further include a deduplication process.

[0170] The search results obtained through each search method are weighted and ranked based on the target weight. Specifically, if the initial semantic matching (vector search) score is 0.6, the upper limit calculated by the hardware score is 0.9. Therefore, the final assignable score of the vector search strategy can be "strengthened" to 0.9. Because the device has stronger computing power, the system can invest more resources in semantic matching (vector search), thereby "strengthening" the score of this strategy and making it dominate the result ranking. For semantic queries, the high score of this strategy is more in line with actual needs.

[0171] The target sorting can determine the number of output results and the order thereof according to needs, output area size, user habits, and the like.

[0172] In one implementation, the electronic device may further be used for at least one of the following:

[0173] In response to obtaining feedback information from a target user regarding the target search result, adjusting weight parameters corresponding to each search method based on the feedback information; adjusting weight parameters corresponding to each search method based on target historical search feedback data; and performing corresponding processing on the search results obtained by the search operation based on the target score data and target historical search feedback data to obtain the target search result. The weight parameter may be an initial weight or a target weight, and the target historical search feedback data may be user feedback data in historical search records.

[0174] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0175] It should also be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0176] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0177] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An interactive processing method, comprising: In response to obtaining target input data, retrieving first memory data and second memory data that match the target input data from a target knowledge base, wherein the target input data is data input to a target application, and the target application is an application capable of providing an artificial intelligence service or calling at least one processing model to provide an artificial intelligence service; and generating and processing the target input data with reference to the first memory data and the second memory data to obtain a target response result that matches the target user's intention; The first memory data and the second memory data have different configuration parameters in the target knowledge base, and the target user intention is determined based at least on the target input data.

2. The method according to claim 1, wherein Retrieving first memory data and second memory data matching the target input data from a target knowledge base, comprising: Performing vectorization processing on the target input data to obtain first vectorized data; Retrieving different data sources from the target knowledge base using the first vectorized data to obtain the first memory data and the second memory data; Among them, information of at least one data field is different between different data sources, and the retrieval methods for searching different data sources are the same or different.

3. The method according to claim 2, wherein: Retrieving different data sources from the target knowledge base using the first vectorized data to obtain the first memory data and the second memory data includes: Retrieving a first data source from the target knowledge base using the first vectorized data in a first retrieval manner to obtain first memory data, where the first data source is a data set obtained by performing a first processing on first user data of the target user acquired within a first time period; Retrieving a second data source from the target knowledge base using the first vectorized data in a second retrieval manner to obtain second memory data, where the second data source is a data set obtained by performing a second processing on second user data of the target user obtained within a second time period, or the second data source is a data set obtained by performing a third processing on the first data source; The first retrieval method and the second retrieval method are the same or different, the first time period is later than the second time period, and the first processing is different from the second processing.

4. The method according to claim 3, wherein: Performing a first processing on the acquired user data of the target user within the first time period to obtain a first data source includes: Performing vectorization processing on the acquired first user data of the target user within the first time period to obtain multiple groups of second vectorized data; performing a first clustering process on the plurality of groups of second vectorized data to obtain a plurality of first vectorized data sets; performing a first summary generation process on the plurality of first vectorized data sets to obtain a first summary data set; The first summary data set and the first user data are extracted and processed according to a first data structure to obtain the first data source.

5. The method according to claim 4, wherein Also includes at least one of the following: After obtaining the plurality of sets of second vectorized data, extracting and processing the plurality of sets of second vectorized data according to the first data structure to obtain the first data source; After obtaining the first summary data set, extracting and processing the first summary data set and the plurality of sets of second vectorized data according to the first data structure to obtain the first data source; The event data obtained by the extraction process is merged and / or deduplicated to obtain the first data source.

6. The method according to claim 3, wherein: Performing a third process on the first data source to obtain a second data source includes: Performing a second clustering process on the first data source to obtain a plurality of second vectorized data sets; performing a second summary generation process on the plurality of second vectorized data sets to obtain a second summary data set; The second summary data set and the first data source are subjected to extraction processing or third summary generation processing according to a second data structure to obtain the second data source.

7. The method according to claim 3, further comprising at least one of the following: processing the first data source into a second data source at a first time interval; Merging and / or deduplicating the second data source based on the first data source; Call different processing models to perform vectorized processing on corresponding types of user data; Associating the user data or vectorized data corresponding to the first data source and the second data source.

8. The method according to claim 1, wherein generating the target input data with reference to the first memory data and the second memory data comprises at least one of the following: Using the first memory data and the second memory data as knowledge base data of a target processing model, and using the target processing model to call the knowledge base data to generate and process the target input data; Using the first memory data and the second memory data to update the prompt word corresponding to the target input data, and using the target processing model to generate and process the updated prompt word data; Optimizing the user intent corresponding to the target input data using the first memory data and the second memory data, and generating and processing the optimized target user intent using a target processing model; Performing a model parameter adjustment process on a target processing model using the first memory data and the second memory data, so as to generate the target input data using the adjusted processing model; The weight of the target data in the target input data is adjusted using the first memory data and the second memory data, and the adjusted input data is generated and processed using a target processing model.

9. The method according to claim 1, wherein generating the target input data with reference to the first memory data and the second memory data comprises at least one of the following: generating third memory data based on weight parameters corresponding to the first memory data and the second memory data respectively; Based on the third memory data, a target processing model is called to generate and process the target input data to obtain the target response result; or, The response result obtained by processing the target input data by the target processing model is optimized based on the third memory data to obtain the target response result.

10. An electronic device comprising at least one processor and at least one processing model capable of running on the processor, wherein the processing model can be called by a target application to perform at least one of the following: In response to obtaining target input data, retrieving first memory data and second memory data that match the target input data from a target knowledge base, wherein the target input data is data input to a target application, and the target application is an application capable of providing an artificial intelligence service or calling at least one processing model to provide an artificial intelligence service; and generating and processing the target input data with reference to the first memory data and the second memory data to obtain a target response result that matches the target user's intention; in, The first memory data and the second memory data have different configuration parameters in the target knowledge base, and the target user intention is determined based at least on the target input data.

Images