Task processing method, data generation method

Abstract

Embodiments of the present specification provide a task processing method and a data generation method. The task processing method comprises: receiving a target task and obtaining context information corresponding to the target task; obtaining task feature information corresponding to the target task, and obtaining historical feature information corresponding to the context information, wherein the historical feature information is determined based on task compression data of at least one interaction round sub-data in the context information; inputting the task feature information and the historical feature information into a task processing model to obtain a target result output by the task processing model, wherein the task processing model determines target historical feature information in the historical feature information according to the task feature information, and generates the target result according to the target historical feature information. The task processing method provided in the present specification is targeted at the target task, and improves the relevance between the target task and the target result.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of artificial intelligence technology, and in particular to a task processing method and a data generation method. Background Technology

[0002] With the continuous development of artificial intelligence technology, the application scenarios of intelligent agents are becoming more and more widespread.

[0003] In related technologies, contextual information, which consists of historical behavioral trajectory data generated during the interaction process guided by an agent, is typically managed through simple context extraction, compression, or passive retrieval. However, these methods struggle to accurately manage contextual information, potentially leading to inaccurate results when using agents to perform tasks. Summary of the Invention

[0004] In view of the above, embodiments of this specification provide a task processing method and a data generation method. One or more embodiments of this specification also relate to a task processing apparatus, a computing device, a computer-readable storage medium, and a computer program product, to address the technical deficiencies existing in the prior art.

[0005] According to a first aspect of the embodiments of this specification, a task processing method is provided, comprising:

[0006] Receive the target task and obtain the context information corresponding to the target task;

[0007] Obtain task feature information corresponding to the target task, and obtain historical feature information corresponding to the context information, wherein the historical feature information is determined based on task compressed data of at least one interaction round sub-data in the context information;

[0008] The task feature information and the historical feature information are input into the task processing model to obtain the target result output by the task processing model. The task processing model determines the target historical feature information from the historical feature information based on the task feature information and generates the target result based on the target historical feature information.

[0009] According to a second aspect of the embodiments of this specification, a data generation method is provided, comprising:

[0010] Obtain initial data information and determine the interaction round corresponding to the initial data information;

[0011] Based on the interaction rounds, the initial interaction information is divided to obtain at least one initial interaction sub-data corresponding to the interaction rounds.

[0012] Encode at least one initial interaction sub-data to generate task compressed data corresponding to at least one initial interaction round sub-data;

[0013] Based on the interaction round sub-data in the interaction history information, at least one task compression data is sequentially arranged to generate feature information corresponding to the task to be processed.

[0014] According to a third aspect of the embodiments of this specification, a task processing method is provided, applied to a cloud-side device, comprising:

[0015] The receiving end device sends the target task and obtains the context information corresponding to the target task;

[0016] Obtain task feature information corresponding to the target task, and obtain historical feature information corresponding to the context information, wherein the historical feature information is determined based on task compressed data of at least one interaction round sub-data in the context information;

[0017] The task feature information and the historical feature information are input into the task processing model to obtain the target result output by the task processing model. The task processing model determines the target historical feature information from the historical feature information based on the task feature information and generates the target result based on the target historical feature information.

[0018] The target result is sent to the end-side device.

[0019] According to a fourth aspect of the embodiments of this specification, a method for training a text adapter is provided, comprising:

[0020] Obtain the sample task, the sample context information corresponding to the sample task, and the sample result corresponding to the sample task;

[0021] The sample context information corresponding to the sample task is input into the text adapter to obtain the sample history feature information output by the text adapter. The text adapter divides the sample context information based on the interaction rounds of the sample context information and obtains the interaction round sub-data corresponding to each interaction round. The interaction round sub-data is encoded to obtain the sample history feature information corresponding to the sample task.

[0022] The sample task and the sample historical feature information are input into the task processing model to obtain the prediction result output by the task processing model;

[0023] The loss value is calculated based on the prediction result and the sample result. The adapter parameters of the text adapter are adjusted according to the loss value, and the text adapter is trained until the model training stops.

[0024] According to a fifth aspect of the embodiments of this specification, a task platform is provided, including a request interface and a response unit;

[0025] The request interface is used to receive the target task sent by the end device and obtain the context information corresponding to the target task;

[0026] The response unit is configured to acquire task feature information corresponding to the target task, acquire historical feature information corresponding to the context information, wherein the historical feature information is determined based on task compressed data of at least one interaction round sub-data in the context information; input the task feature information and the historical feature information into the task processing model to obtain the target result output by the task processing model, wherein the task processing model determines the target historical feature information in the historical feature information according to the task feature information, and generates the target result according to the target historical feature information.

[0027] According to a sixth aspect of the embodiments of this specification, a task processing apparatus is provided, comprising:

[0028] The receiving unit is configured to receive the target task and obtain the context information corresponding to the target task;

[0029] The acquisition unit is configured to acquire task feature information corresponding to the target task and acquire historical feature information corresponding to the context information, wherein the historical feature information is determined based on task compressed data of at least one interaction round sub-data in the context information.

[0030] The processing unit is configured to input the task feature information and the historical feature information into the task processing model to obtain the target result output by the task processing model, wherein the task processing model determines the target historical feature information from the historical feature information based on the task feature information, and generates the target result based on the target historical feature information.

[0031] According to a seventh aspect of the embodiments of this specification, a data generation apparatus is provided, comprising:

[0032] The acquisition unit is configured to acquire initial data information and determine the interaction round corresponding to the initial data information;

[0033] The segmentation unit is configured to segment the initial interaction information based on the interaction rounds and obtain initial interaction sub-data corresponding to at least one interaction round.

[0034] The encoding unit is configured to encode at least one initial interaction sub-data to generate task compressed data corresponding to at least one initial interaction round sub-data;

[0035] The sorting unit is configured to sequentially arrange at least one task compressed data according to the round number of at least one interaction round sub-data in the interaction history information, and generate feature information corresponding to the task to be processed.

[0036] According to an eighth aspect of the embodiments of this specification, a training apparatus for a text adapter is provided, comprising:

[0037] The acquisition unit is configured to acquire a sample task, sample context information corresponding to the sample task, and sample result corresponding to the sample task;

[0038] The processing unit is configured to input the sample context information corresponding to the sample task into a text adapter, and obtain the sample historical feature information output by the text adapter. The text adapter divides the sample context information based on the interaction rounds of the sample context information, and obtains at least one interaction round sub-data corresponding to the interaction rounds. The at least one interaction round sub-data is encoded to obtain the sample historical feature information corresponding to the sample task.

[0039] The prediction unit is configured to input the sample task and the sample historical feature information into the task processing model to obtain the prediction result output by the task processing model;

[0040] The training unit is configured to calculate a loss value based on the prediction result and the sample result, adjust the adapter parameters of the text adapter based on the loss value, and continue training the text adapter until the model training stopping condition is met.

[0041] According to a ninth aspect of the embodiments of this specification, a computing device is provided, comprising:

[0042] Memory and processor;

[0043] The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the above method.

[0044] According to a tenth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0045] According to an eleventh aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0046] According to the task processing method provided in this specification, compressed task data determined based on the context information corresponding to the target task is obtained. This enables the compressed task data to be guided by the target task, generating compressed task data associated with the target task and improving the correlation between the compressed task data and the context information corresponding to the target task. Furthermore, the compressed task data is determined based on at least one interaction round sub-data, avoiding information overflow while retaining key information within the compressed task data. Based on this, the task feature information of the target task and the historical feature information corresponding to the context information are output into the task processing model. This allows for the generation of accurate target results during the processing of the target task, based on the historical feature information obtained while retaining key information and guided by the target task, thereby improving the correlation between the target result and the target task. Attached Figure Description

[0047] Figure 1 A schematic diagram of a task processing method is shown;

[0048] Figure 2 A flowchart of a task processing method according to an embodiment of this specification is shown;

[0049] Figure 3 A flowchart is shown of a training method for a text adapter according to one embodiment of this specification;

[0050] Figure 4 A flowchart of a data generation method according to an embodiment of this specification is shown;

[0051] Figure 5 A schematic diagram of the processing procedure of a task processing method provided in one embodiment of this specification is shown;

[0052] Figure 6 This specification shows a schematic diagram of the structure of a task processing apparatus according to one embodiment;

[0053] Figure 7 This specification shows a schematic diagram of the structure of a data generation apparatus according to one embodiment;

[0054] Figure 8 A schematic diagram of the structure of a text adapter training device provided in one embodiment of this specification is shown;

[0055] Figure 9 This specification shows an architecture diagram of a task processing system provided in one embodiment;

[0056] Figure 10 A structural block diagram of a computing device provided according to one embodiment of this specification is shown. Detailed Implementation

[0057] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0058] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0059] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0060] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this manual are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0061] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0062] Intelligent Agent: An intelligent agent is a computational entity capable of autonomously perceiving its environment, making decisions, and executing actions to achieve its goals. The intelligent agents described in this specification are used to achieve task-oriented goals based on relevant tasks.

[0063] The task processing methods described in this specification are applied in scenarios where intelligent agents guide related tasks and generate corresponding question-and-answer results. For example, they can be applied to scenarios where historical trajectory information generated during historical interactions of intelligent agents guiding related tasks is accurately represented.

[0064] For example, when an agent guides a related task through at least one round of dialogue to generate the result corresponding to the related task, the historical feature information of the historical task is usually determined based on the interaction history information between the agent and historical tasks with semantic similarity to the related task. This allows the agent to generate the result corresponding to the related task based on the historical feature information.

[0065] However, in determining historical feature information based on interaction history, to avoid information overflow, related technologies typically compress the interaction history information iteratively according to the rounds of interaction. This ensures that the interaction history information of each round can represent the characteristics of previous rounds, and then uses the compressed data of the last interaction round as the historical feature information that can represent the interaction history rounds. However, with this approach, because the interaction information of each round iterates with the interaction information of the previous round, and to avoid information overflow, the iterated information is usually compressed, which can easily lead to the loss of key information after each round of compression. Therefore, the existing technology struggles to achieve a balance between preserving key information and avoiding information overflow.

[0066] In addition, such as Figure 1 As shown, Figure 1 A schematic diagram of a task processing method is shown. Figure 1 In this approach, the relevant technology stores the interaction history information obtained from iterative compression in a memory module, divides the iteratively compressed data into several text blocks, and encodes each text block to generate a corresponding block vector for subsequent retrieval. During information interaction, when the user inputs a task, the similarity between the task and the vector corresponding to each text block is calculated. Several text blocks with similarity greater than a similarity threshold are selected and concatenated with the user's input task as input to a task processing model (e.g., a Large Language Model (LLM)). However, this approach typically relies on predefined structures and static rules to determine the target result, which can easily lead to the generated target result being unable to flexibly adapt to the relevant task.

[0067] Therefore, this specification provides a task processing method that processes context information from at least one round to generate historical feature information, compresses the sub-data rows of the interaction rounds from at least one round to retain key information while avoiding information overflow. Furthermore, guided by the target task, the method processes the context information corresponding to the target task to obtain historical feature information, establishing a correlation between the historical feature information and the target task. This specification also relates to a task processing apparatus, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0068] See Figure 2 , Figure 2 A flowchart of a task processing method according to an embodiment of this specification is shown, including steps 202-206.

[0069] Step 202: Receive the target task and obtain the context information corresponding to the target task.

[0070] In one specific embodiment of this specification, a target task to be executed in the intelligent agent is determined, and the corresponding context information is obtained based on the target task.

[0071] The target task can be understood as the task that the agent needs to perform in the current scenario. Contextual information can be historical information generated during the dialogue between at least one agent based on the target task.

[0072] For example, the target task could be a task that the user enters in text form.

[0073] In one specific embodiment of this specification, obtaining the context information corresponding to the target task includes:

[0074] Obtain task dialogue information corresponding to the target task, and extract the context information from the task dialogue information; and / or, determine a reference task based on the target task, and obtain the interaction history information corresponding to the reference task, and use the interaction history information as the context information, wherein the semantic similarity between the first semantic information of the target task and the second semantic information of the reference task is greater than a preset similarity threshold.

[0075] In one specific embodiment of this specification, task dialogue information corresponding to the target task is obtained, and context information is extracted from the task dialogue information. For example, the task dialogue information is filtered according to its semantic similarity to the target task to extract task dialogue information related to the target task.

[0076] In another specific embodiment provided in this specification, a reference task is determined based on the target task, and the interaction history information corresponding to the reference task is obtained. The interaction history information is used as the context information, wherein the semantic similarity between the first semantic information of the target task and the second semantic information of the reference task is greater than a preset similarity threshold.

[0077] For example, a target task for determining the question-answering result is received, and based on the first semantic information of the target task, second semantic information with a semantic similarity greater than a similarity threshold with the first semantic information is determined, and a reference task corresponding to the second semantic information is determined.

[0078] In this context, the reference task can be understood as a historical task selected from historical tasks that has a strong semantic relevance to the target task. For example, the target task could be a question posed by a user to the agent, allowing the agent to reuse historical interaction information based on the question and the reference task to obtain the corresponding result.

[0079] In another specific embodiment provided in this specification, the context information extracted from the task dialogue information and the interaction history information obtained from the reference task are together used as the context information involved in this specification.

[0080] Based on the above, contextual information can be determined by the interaction history information obtained by at least one agent performing the target task or reference task. Furthermore, contextual information can characterize the information processing capabilities of at least one agent.

[0081] According to the above implementation method, a reference task related to the semantics of the target task is determined based on semantic information, so that relevant information corresponding to the reference task can be reused in the future, thereby assisting the agent in performing the target task efficiently.

[0082] Step 204: Obtain the task feature information corresponding to the target task, and obtain the historical feature information corresponding to the context information, wherein the historical feature information is determined based on the task compressed data of at least one interaction round sub-data in the context information.

[0083] Among them, historical feature information can be understood as a global feature representation of context information, which is generated by integrating and encoding the task compression data of all interaction rounds corresponding to the context information.

[0084] Task compressed data can be understood as a compact information representation generated by compressing and task-oriented encoding the sub-data of a single interaction round based on contextual information.

[0085] Interaction round sub-data can be understood as the interaction information corresponding to each interaction round within the context information. This interaction round sub-data can represent the interaction content of a single interaction round.

[0086] In one specific embodiment provided in this specification, task feature information corresponding to the target task and historical feature information corresponding to the context information are obtained. During the acquisition of historical feature information, at least one interaction round sub-data is obtained from the interaction history information corresponding to the context information to avoid the loss of interaction round sub-data during the interaction process. Based on this, at least one interaction round sub-data is compressed to avoid information redundancy or information overflow. Furthermore, during the compression of the interaction round sub-data, the context information of the target task is considered, thereby ensuring that the compressed task data is guided by the target task, making the correlation between the compressed task data and the reference task closer.

[0087] In one specific embodiment provided in this specification, during the compression of interaction round sub-data, for example, the compression rate of different interaction round sub-data is set according to the semantics of the target task, thereby ensuring the correlation between the task compressed data and the target task.

[0088] For ease of understanding, this specification uses the following method to explain the content of the historical feature information corresponding to the obtained context information.

[0089] In one or more embodiments of this specification, historical feature information is generated in the manner shown in S2042-S2048, including:

[0090] S2042: Determine the interaction round corresponding to the context information.

[0091] In one specific embodiment provided in this specification, context information consisting of information from at least one interaction round is obtained.

[0092] For example, the agent guides the reference task by generating historical information during historical interactions.

[0093] In one specific embodiment provided in this specification, context information is obtained to provide support for subsequent processing of the target task.

[0094] For example, collecting interaction information for each individual round during the execution of a target or reference task, and then assembling interaction history information based on this individual round's interaction information. Here, the individual round's interaction information can be understood as information generated separately for the interaction content of each round.

[0095] S2044: Based on the interaction rounds, divide the context information and obtain at least one interaction round sub-data corresponding to the interaction rounds.

[0096] In one specific embodiment provided in this specification, in order to avoid the loss of key information, the interaction history information is divided according to the interaction rounds to obtain at least one interaction round sub-data corresponding to the interaction round, so that the interaction round sub-data of each interaction round can be processed in the future, thereby avoiding the loss of key information.

[0097] For example, to avoid losing key information, all interaction round sub-data from all interaction rounds that constitute the context information can be obtained.

[0098] In one or more embodiments of this specification, obtaining interaction round sub-data corresponding to at least one interaction round includes:

[0099] Obtain initial interaction round sub-data and corresponding attribute information, wherein the initial interaction round sub-data is any one of the initial interaction round sub-data corresponding to at least one interaction round, and the attribute information includes at least one of the thinking content, tools used, and analysis results; based on the attribute information, segment the initial interaction round sub-data to obtain interaction round sub-data.

[0100] In one specific embodiment provided in this specification, to improve the efficiency of subsequent processing of interaction round sub-data, after dividing the interaction history information, initial interaction round sub-data is generated according to the interaction order, and the obtained initial interaction round sub-data is processed by segmentation or data filtering. Attribute information corresponding to the initial interaction round is obtained, and the initial interaction round sub-data is segmented based on the attribute information to remove redundant information. The segmented initial interaction round sub-data is then identified as the interaction round sub-data, thereby improving the efficiency of subsequent processing of the interaction round sub-data.

[0101] For example, when filtering the initial interaction round sub-data based on attribute information, if there is initial interaction round sub-data that does not contain attribute information, then the initial interaction round sub-data is filtered out. The interaction round sub-data obtained after filtering out the interaction round sub-data that does not contain attribute information and filtering out all information in the initial interaction round sub-data except for attribute information is used as the interaction round sub-data.

[0102] In one specific embodiment provided in this specification, if the attribute information in the initial interaction round sub-data includes thinking content, tools used, and analysis results, then the text sequence in the interaction round sub-data is segmented according to the thinking content, tools used, and analysis results to remove redundant information other than the thinking content, tools used, and analysis results.

[0103] For example, in an interaction round, a complete interaction between the agent and the external environment is as follows: When the objective task is to find content A in article 1, the agent considers the content in article 1 that contains content A, and uses a relevant search engine to obtain the page number, chapter, etc., of article 1 containing content A. Based on this, the attribute information corresponding to this interaction round includes: thought content: "content in article 1 that contains content A"; tool used: "search engine"; thought result: "page number, chapter, etc., of article 1 that contains content A". Based on the thought content, tool used, and thought result mentioned above, the interaction round sub-data is segmented.

[0104] According to one or more embodiments of this specification, after segmenting the context information by interaction round, the interaction round sub-data is further segmented based on attribute information. This allows the segmented interaction round sub-data to represent a complete interaction between the agent and the external environment, thereby improving the semantic coherence and contextual consistency of the interaction information within each round. Furthermore, obtaining the interaction round sub-data through attribute information segmentation allows for the expansion of the processing length of subsequent interaction history information by segmenting the interaction round sub-data for each interaction round without sacrificing the richness and diversity of the context information.

[0105] S2046: Encode at least one interactive round sub-data to generate at least one task compressed data corresponding to at least one interactive round sub-data.

[0106] In one specific embodiment provided in this specification, each interaction round sub-data is compressed into a compact feature vector, which reduces the data dimensionality while retaining key semantic information. Furthermore, the features related to the reference task target are strengthened through encoding, so that the generated historical feature information has a clear task orientation, thereby supporting subsequent semantic matching with the target task based on the historical feature information.

[0107] In one or more embodiments of this specification, at least one interaction round sub-data is encoded to generate task compressed data corresponding to each interaction round sub-data, including:

[0108] Obtain the interaction round sub-data to be processed, wherein the interaction round sub-data to be processed is any one of at least one interaction round sub-data; compress the interaction round sub-data to be processed based on the compression ratio parameter to obtain at least one compressed data; encode the at least one compressed data based on the target task to generate at least one task compressed data corresponding to the interaction round sub-data to be processed.

[0109] The compression ratio parameter can be understood as a value adjusted during actual execution or preset by humans to balance critical information and prevent information overflow. This setting improves information processing efficiency. For example, setting the compression ratio parameter to 5 means that every 5 data points can be compressed into 1 data point.

[0110] In one specific embodiment provided in this specification, when the interaction round sub-data is obtained, to avoid information overflow, the interaction round sub-data to be processed is compressed according to the compression ratio parameter to obtain at least one compressed data corresponding to the interaction round sub-data to be processed. All compressed data in each interaction round sub-data to be processed are integrated and encoded to obtain at least one compressed data that can characterize the interaction round sub-data to be processed. Then, based on the at least one compressed data corresponding to the interaction round sub-data, at least one task compressed data corresponding to the interaction round sub-data to be processed is generated.

[0111] According to one or more embodiments described above in this specification, the interaction round sub-data is compressed according to a preset compression ratio parameter, thereby achieving information compression processing while ensuring that key information is not lost.

[0112] In one or more embodiments of this specification, based on a preset compression ratio parameter, the sub-data of the interaction round to be processed is compressed to obtain at least one compressed data corresponding to the sub-data of the interaction round to be processed, including:

[0113] Based on the compression ratio parameter, the interaction round sub-data to be processed is segmented to obtain at least one sub-data block to be processed corresponding to the interaction round sub-data to be processed, wherein the amount of data information in the sub-data block to be processed is less than or equal to the set value of the preset compression ratio parameter; the data information in at least one sub-data block to be processed is compressed to obtain at least one compressed data corresponding to the interaction round sub-data to be processed.

[0114] In this context, data information can be understood as the smallest unit that makes up the sub-data of an interaction round. For example, when the sub-data of an interaction round is a text sequence, the data information can be a single text that makes up the text sequence.

[0115] In one specific embodiment provided in this specification, the sub-data of the interaction round to be processed is divided into at least one sub-data block according to the compression ratio parameter. By splitting long text into small-granularity sub-blocks, the computational complexity of compression is reduced. Each sub-data block is processed separately, and the data information in each initial sub-data block is compressed, so that the compressed data corresponding to each initial sub-data block can represent all the data information of the initial sub-data block. The compressed data corresponding to each initial sub-data block is used as at least one compressed data corresponding to the sub-data of the interaction round to be processed. This ensures that all sub-data blocks to be processed in each interaction round are compressed, avoiding information loss.

[0116] According to one or more embodiments described above in this specification, data is segmented based on a compression ratio parameter to ensure that the amount of information in each sub-data block to be processed does not exceed a set value. This avoids compression deviations caused by uneven original data lengths, making the compression process more controllable. Block processing can distribute computational pressure, reducing memory usage and processing latency.

[0117] In one or more embodiments of this specification, data information in at least one sub-data block to be processed is compressed to obtain at least one compressed data, including:

[0118] Obtain initial data information and at least one adjacent data information corresponding to the initial data information in the initial sub-data block, wherein the initial sub-data block is at least one sub-data block to be processed, and the initial data information is any one of at least one data information of the initial sub-data block; encode based on at least one adjacent data information and the initial data information to obtain initial encoding information corresponding to the initial data information; encode at least one initial encoding information to obtain compressed data corresponding to the initial sub-data block.

[0119] In one specific embodiment provided in this specification, the arrangement order of each data information is determined in the initial sub-data block. Based on the arrangement order, the left adjacent data information arranged before the initial data information and / or the right adjacent data information arranged after the initial data information are obtained.

[0120] In one specific embodiment provided in this specification, when the initial sub-data block includes at least three initial data pieces, each initial data piece, excluding the first and last initial data pieces, is encoded based on its left-adjacent data pieces, right-adjacent data pieces, and its own data pieces, so that the initial encoded information of the corresponding initial data piece can characterize the information features adjacent to it. For the first initial data piece, it is encoded using its right-adjacent data pieces; for the last initial data piece, it is encoded using its left-adjacent data pieces, resulting in the initial encoded information corresponding to the complete initial data piece. Based on this, each initial encoded information is encoded to obtain the compressed data corresponding to the initial sub-data block.

[0121] According to one or more embodiments described above in this specification, initial data information and adjacent data information are determined to avoid isolated encoding and thus ensure the semantic accuracy of the encoding. Encoding based on adjacent data and initial data can also avoid misjudgments of tool functions caused by isolated encoding. Compared to directly encoding the entire initial sub-data block, encoding individual data information and its associated data in each initial sub-data block first can preserve local details more finely and provide more accurate information for subsequent compression.

[0122] S2048: Based on the round number of at least one interaction round sub-data in the context information, sequentially arrange at least one task compression data, and obtain the historical feature information corresponding to the context information.

[0123] In one specific embodiment provided in this specification, at least one interaction round sub-data is compressed into a compact feature vector, reducing data dimensionality while preserving key semantic information. Based on the feature vector obtained from compressing each interaction round sub-data, task compressed data corresponding to the interaction round sub-data is generated.

[0124] In one specific embodiment provided in this specification, the compressed task data corresponding to each interaction round's sub-data is arranged sequentially according to its original round order in the interaction history (e.g., round 1, round 2... round n), forming an ordered sequence. This ensures that the sequence of compressed data is consistent with the actual processing flow, avoiding logical breaks caused by disordered order. The round-arranged compressed task data sequence is globally encoded to capture the dependencies between rounds and generate historical feature information corresponding to the reference task.

[0125] In a specific embodiment provided in this specification, the method for generating historical feature information in S2042-S2046 above can be implemented by a pre-trained text adapter. Specifically, the acquired context information is input into the text adapter for processing. The text adapter divides the context information based on the interaction rounds, obtains interaction round sub-data corresponding to at least one interaction round, and encodes at least one interaction round sub-data to obtain the historical feature information corresponding to the reference task.

[0126] Text adapters are used to segment, encode, and compress contextual information to generate corresponding historical feature information. In practical applications, they can be generated through training with sample data.

[0127] like Figure 3 As shown, Figure 3 A flowchart of a text adapter training method according to an embodiment of this specification is shown. Specifically, the text adapter training method includes steps 302-308:

[0128] Step 302: Obtain the sample task, the sample context information corresponding to the sample task, and the sample result corresponding to the sample task.

[0129] Step 304: Input the sample context information corresponding to the sample task into the text adapter to obtain the sample history feature information output by the text adapter. The text adapter divides the sample context information based on the interaction rounds of the sample context information and obtains at least one interaction round sub-data. The at least one interaction round sub-data is encoded to obtain the sample history feature information corresponding to the sample task.

[0130] Step 306: Input the sample task and the sample historical feature information into the task processing model to obtain the prediction result output by the task processing model.

[0131] Step 308: Calculate the loss value based on the prediction result and the sample result, adjust the adapter parameters of the text adapter based on the loss value, and continue training the text adapter until the model training stops.

[0132] In one specific embodiment provided in this specification, the text adapter can be understood as a text processing model that can be pre-trained. The sample task, sample context information, and sample results can be successful cases processed by the agent, which are used as sample data to train the text adapter. Through this method, it can acquire the ability to generate historical feature information based on historical interaction information.

[0133] The trained text adapter can perform the operations described in S2042-S2048.

[0134] Step 206: Input the task feature information and the historical feature information into the task processing model to obtain the target result output by the task processing model. The task processing model determines the target historical feature information from the historical feature information based on the task feature information and generates the target result based on the target historical feature information.

[0135] In one specific implementation provided in this specification, after determining the target historical feature information, the model reuses historical experience corresponding to historical feature information that is semantically similar to the target historical feature information, and generates a target result that is adapted to the current task in combination with the specific requirements of the target task.

[0136] In one or more embodiments of this specification, the task feature information and the historical feature information are input into a task processing model to obtain the target result output by the task processing model, including:

[0137] The task feature information and the historical feature information are input into the task processing model to obtain the initial result processed by the task processing model; it is determined whether the initial result meets the set conditions; if the initial result meets the set conditions, the initial result is taken as the target result; if the initial result does not meet the set conditions, the historical feature information is adjusted until the initial result obtained based on the adjusted historical feature data meets the set conditions, and the initial result that meets the set conditions is taken as the target result.

[0138] The set conditions can be understood as primarily used to evaluate the semantic correlation between historical feature information and task feature information. For example, they can also be used to determine the accuracy of the target result and whether any key information has been omitted.

[0139] In one specific embodiment provided in this specification, the task feature information of the target task and the historical feature information of the reference task are input into the task processing model. The task processing model reuses the historical feature information to generate an initial result. The initial result is then judged to be qualified based on set conditions. If the initial result fully meets the set conditions, it is directly determined as the target result. If the initial result does not meet the set conditions, the historical feature information is adjusted in reverse, and the result is re-inputted into the model until the set conditions are met.

[0140] According to one or more embodiments described above in this specification, by dynamically adjusting historical feature information, the intelligent agent can reuse historical experience and, in conjunction with the current task requirements, determine a target result that is more relevant to the current task.

[0141] In one or more embodiments of this specification, adjusting the historical feature information includes:

[0142] Adjust the encoding parameters for encoding the sub-data of the interaction rounds; based on the adjusted encoding parameters, obtain the adjusted historical feature information corresponding to the reference task.

[0143] In one specific embodiment provided in this specification, if the initial result does not meet the set conditions, the encoding parameters for encoding the sub-data of each interaction round in the interaction history information are adjusted, for example, the weights in the encoding process are adjusted, so that the adjusted historical feature information corresponding to the reference task can be obtained according to the adjusted encoding parameters.

[0144] Specifically, in the above steps, it is indicated that the operation of generating historical feature information can be performed through a text adapter. In this embodiment, the encoding parameters in the text adapter can be adjusted to re-encode the sub-data of each interaction round, thereby generating new historical feature information.

[0145] According to one or more embodiments described above in this specification, task compression data determined based on a reference task is obtained, enabling the task compression data to characterize the reference task and improving the correlation between the task compression data and the reference task. Furthermore, task compression data is determined by acquiring sub-data of each interaction round from historical information, preserving key information in the historical information while avoiding information overflow caused by directly acquiring historical information. Based on this, a target task semantically related to the reference task is determined, allowing the target task and historical feature information to be processed through a task processing model. The task processing model, by locating the target historical feature information, directly reuses the processing experience of historical feature information, avoiding redundant calculations of exploring decision paths from scratch for the target task. This reduces parameter debugging time and significantly improves processing efficiency.

[0146] The following is in conjunction with the appendix Figure 4 Taking the application of the task processing method provided in this manual in data generation as an example, the data generation method will be further explained.

[0147] See Figure 4 , Figure 4 A flowchart of a data generation method according to an embodiment of this specification is shown, including steps 402-408.

[0148] Step 402: Obtain initial data information and determine the interaction round corresponding to the initial data information.

[0149] In one specific embodiment provided in this specification, initial data information generated during historical interactions by at least one intelligent agent guiding the task to be processed is obtained.

[0150] In one specific embodiment provided in this specification, the initial data information can be understood as historical information generated during the interaction of multiple intelligent agents.

[0151] Step 404: Based on the interaction rounds, divide the initial interaction information and obtain initial interaction sub-data corresponding to at least one interaction round.

[0152] Among them, the initial interaction sub-data corresponding to each interaction round can constitute the complete initial interaction information.

[0153] Step 406: Encode at least one initial interaction sub-data to generate task compressed data corresponding to at least one initial interaction round sub-data.

[0154] Step 408: Based on the number of at least one interaction round sub-data in the interaction history information, sequentially arrange at least one task compression data to generate feature information corresponding to the task to be processed.

[0155] In one specific embodiment provided in this specification, contextual information generated during historical interactions by at least one intelligent agent guiding a task to be processed is obtained and used as initial data information. Based on this, to facilitate the retention of key information in the initial data information, the initial interaction information is divided into initial interaction round sub-data corresponding to at least one round. The at least one initial interaction round sub-data is encoded to obtain feature information corresponding to the task to be processed.

[0156] In one or more embodiments provided in this specification, the data generation method further includes: adjusting the encoding parameters in the data generation method, and regenerating the feature information corresponding to the task to be processed based on the adjusted encoding parameters.

[0157] In one specific embodiment provided in this specification, when the target result obtained from the task processing model does not meet the set conditions, the encoding parameters are adjusted based on the information fed back by the task processing model, and the feature information corresponding to the task to be processed is regenerated. Subsequently, the regenerated feature information corresponding to the task to be processed is used as historical feature information input to the task processing model.

[0158] The setting conditions can be understood as the process of determining the result of the target task based on the feature information of the initial data information, and the generation of results that are strongly correlated with the target task based on the feature information.

[0159] According to one or more embodiments described above in this specification, the effectiveness of feature information is ensured through feedback adjustment, providing reliable input for subsequent task feature matching with historical features and avoiding a decrease in processing efficiency due to low feature quality.

[0160] The following is in conjunction with the appendix Figure 5 Taking the task processing method provided in this specification as an example of its application in question-and-answering of a target task based on an intelligent agent, the task processing method will be further explained. Figure 5 This specification illustrates a schematic diagram of the processing procedure of a task processing method provided in one embodiment.

[0161] like Figure 5 As shown, the target task is obtained, and contextual information related to the target task is determined. For example, interaction history information corresponding to reference tasks semantically similar to the target task is used as contextual information. Using the above... Figure 1 The method shown determines the historical feature information corresponding to the reference task. For example, the generation process of the historical feature information is set to be carried out by a text adapter. Based on this, the interaction history information of the reference task is obtained and processed in the text adapter.

[0162] In practical applications, a text adapter can be trained using sample data and a task processing model. Specifically, the sample data includes a sample reference task, sample interaction history information corresponding to the sample context information, and sample results corresponding to the sample reference task. The interaction history information corresponding to the sample reference task is input into the text adapter. In the text adapter, based on the interaction rounds of the interaction history information, the interaction history information is divided, and interaction round sub-data corresponding to each interaction round is obtained. Each interaction round sub-data is encoded to obtain the historical feature information corresponding to the reference task.

[0163] The reference task and its corresponding historical feature information are then input into the task processing model to obtain the prediction results output by the model. The loss value is calculated based on the prediction results and sample results, and the adapter parameters of the text adapter are adjusted accordingly. Training of the text adapter continues until the training stopping condition is met. It is important to note that during the training of the text adapter, only the parameters of the text adapter are adjusted, not the model parameters of the task processing model.

[0164] In one specific embodiment provided in this specification, the interaction history information is segmented according to the interaction rounds to obtain interaction round sub-data containing corresponding attribute information. Each interaction round sub-data is a text sequence containing at least one text. Based on the attribute information, a complete interaction between the agent and the external environment in each interaction round is characterized, ensuring strong correlation and contextual consistency of information within each round. This expands the context length for subsequent processing without sacrificing the richness and diversity of contextual information, thereby providing the agent with more information support and avoiding the problem of insufficient context length in related technologies. The specific implementation method is described in S2044 above in this specification, and will not be repeated here.

[0165] Furthermore, based on the set compression ratio parameter, the sub-data for each interaction round is further segmented. For example, when the compression ratio parameter is set to k, every k texts form a sub-data block, which is then compressed to obtain compressed data. This helps to effectively control the number of texts and optimize information transmission efficiency. In addition, to improve processing efficiency and maintain contextual coherence, bidirectional encoding (e.g., a bidirectional Attention Encoder) can be used to generate a more compact contextual representation, resulting in task compressed data. This allows the core information of the interaction round sub-data to be expressed based on the encoded compressed data. The specific implementation method is described in S2046 above in this specification, and will not be repeated here.

[0166] Building upon the above, the compressed data corresponding to the sub-data blocks of each interaction round is further encoded according to the reference task. Based on task-oriented encoding, the compression of contextual information is further improved, generating denser compressed data. This dense compressed data is then sorted to obtain historical feature information. This makes the information representation more compact and allows for better flow and correlation of information between different blocks. The specific implementation method is described in S2046 of this specification, and will not be repeated here.

[0167] Based on the above, the obtained historical feature information and the task feature information corresponding to the target task are input into the task processing model (e.g., LLM) to obtain the target result output by the task processing model.

[0168] Furthermore, if the target result does not meet the set conditions, the task processing model and the text adapter are jointly trained to update the parameters of the text adapter, thereby obtaining the historical feature information generated by the text adapter with updated parameters.

[0169] In one specific embodiment provided in this specification, interaction history information is determined based on a reference task, and a reference result corresponding to the reference task is determined, wherein the reference result can be understood as the standard output result corresponding to the reference task.

[0170] After the task processing model receives the historical feature information generated by the text adapter, it generates an initial prediction based on this information. At this point, the difference between the initial result and the reference result is calculated, and this difference is compared to a difference threshold.

[0171] When the difference between the initial result and the reference result is less than the difference threshold, it indicates that the initial result is highly matched with the standard output result of the reference task.

[0172] When the difference between the initial result and the reference result exceeds a difference threshold, it indicates that the task processing model, based on current historical feature information, cannot predict a standard output result that highly matches the reference task. In this case, to obtain an accurate initial result based on the reference task, the first gradient value between the corresponding parameters of each network layer in the task processing model is calculated. This first gradient value is backpropagated to the input of the text adapter to calculate the second gradient value of the parameters in the text adapter. Based on the second gradient value, the parameters in the text adapter are updated so that they are adjusted in the direction of decreasing difference. Therefore, when historical feature information is obtained based on the continuously adjusted parameters in the text adapter and used as input to the task processing model, the initial result obtained by the task processing model highly matches the reference task.

[0173] It is important to understand that the parameters of the task processing model mentioned above remain unchanged.

[0174] Corresponding to the above method embodiments, this specification also provides embodiments of a task processing device. Figure 6 A schematic diagram of a task processing apparatus according to one embodiment of this specification is shown. Figure 6 As shown, the device includes:

[0175] The receiving unit 602 is configured to receive the target task and obtain the context information corresponding to the target task;

[0176] The acquisition unit 604 is configured to acquire task feature information corresponding to the target task and acquire historical feature information corresponding to the context information, wherein the historical feature information is determined based on task compressed data of at least one interaction round sub-data in the context information.

[0177] The processing unit 606 is configured to input the task feature information and the historical feature information into the task processing model to obtain the target result output by the task processing model, wherein the task processing model determines the target historical feature information from the historical feature information based on the task feature information, and generates the target result based on the target historical feature information.

[0178] Optionally, the acquisition unit 604 is further configured as follows:

[0179] Obtain task dialogue information corresponding to the target task, and extract the context information from the task dialogue information; and / or, determine a reference task based on the target task, and obtain the interaction history information corresponding to the reference task, and use the interaction history information as the context information, wherein the semantic similarity between the first semantic information of the target task and the second semantic information of the reference task is greater than a preset similarity threshold.

[0180] Optionally, the acquisition unit 604 is further configured as follows:

[0181] Determine the interaction round corresponding to the context information; based on the interaction round, divide the context information and obtain at least one interaction round sub-data corresponding to the interaction round; encode the at least one interaction round sub-data to generate at least one task compressed data corresponding to the at least one interaction round sub-data; arrange at least one task compressed data in order according to the round of the at least one interaction round sub-data in the context information, and obtain the historical feature information corresponding to the context information.

[0182] Optionally, the acquisition unit 604 is further configured as follows:

[0183] Obtain the interaction round sub-data to be processed, wherein the interaction round sub-data to be processed is any one of at least one interaction round sub-data; compress the interaction round sub-data to be processed based on the compression ratio parameter to obtain at least one compressed data; encode the at least one compressed data based on the target task to generate at least one task compressed data corresponding to the interaction round sub-data to be processed.

[0184] Optionally, the acquisition unit 604 is further configured as follows:

[0185] Based on the compression ratio parameter, the interaction round sub-data to be processed is segmented to obtain at least one sub-data block to be processed corresponding to the interaction round sub-data, wherein the amount of data information in the sub-data block to be processed is less than or equal to the set value of the compression ratio parameter; the data information in at least one sub-data block to be processed is compressed to obtain at least one compressed data corresponding to the interaction round sub-data.

[0186] Optionally, the acquisition unit 604 is further configured as follows:

[0187] Obtain initial data information and at least one adjacent data information corresponding to the initial data information in the initial sub-data block, wherein the initial sub-data block is at least one sub-data block to be processed, and the initial data information is any one of at least one data information of the initial sub-data block; encode based on at least one adjacent data information and the initial data information to obtain initial encoding information corresponding to the initial data information; encode at least one initial encoding information to obtain compressed data corresponding to the initial sub-data block.

[0188] Optionally, the acquisition unit 604 is further configured as follows:

[0189] Obtain initial interaction round sub-data and corresponding attribute information, wherein the initial interaction round sub-data is any one of the initial interaction round sub-data corresponding to at least one interaction round, and the attribute information includes at least one of the thinking content, tools used, and analysis results; based on the attribute information, segment the initial interaction round sub-data to obtain interaction round sub-data.

[0190] Optionally, the acquisition unit 604 is further configured as follows:

[0191] The task feature information and the historical feature information are input into the task processing model to obtain the initial result processed by the task processing model; it is determined whether the initial result meets the set conditions; if the initial result meets the set conditions, the initial result is taken as the target result; if the initial result does not meet the set conditions, the historical feature information is adjusted until the initial result obtained based on the adjusted historical feature data meets the set conditions, and the initial result that meets the set conditions is taken as the target result.

[0192] Optionally, the acquisition unit 604 is further configured as follows:

[0193] Adjust the encoding parameters for encoding the sub-data of the interaction rounds; based on the adjusted encoding parameters, obtain the adjusted historical feature information corresponding to the reference task.

[0194] Optionally, the acquisition unit 604 is further configured as follows:

[0195] Determine the interaction round corresponding to the context information; input the context information into a text adapter to obtain the historical feature information output by the text adapter, wherein the text adapter divides the context information based on the interaction round, obtains interaction round sub-data corresponding to at least one interaction round, encodes at least one interaction round sub-data, and obtains the historical feature information corresponding to the context information.

[0196] The above is an illustrative scheme of a task processing device according to this embodiment. It should be noted that the technical solution of this task processing device and the technical solution of the task processing method described above belong to the same concept. For details not described in detail in the technical solution of the task processing device, please refer to the description of the technical solution of the task processing method described above.

[0197] Corresponding to the above method embodiments, this specification also provides embodiments of a data generation apparatus. Figure 7 A schematic diagram of a data generation apparatus according to one embodiment of this specification is shown. Figure 7 As shown, the device includes:

[0198] The acquisition unit 702 is configured to acquire initial data information and determine the interaction round corresponding to the initial data information;

[0199] The segmentation unit 704 is configured to segment the initial interaction information based on the interaction rounds and obtain initial interaction sub-data corresponding to at least one interaction round.

[0200] The encoding unit 706 is configured to encode at least one initial interaction sub-data to generate task compressed data corresponding to at least one initial interaction round sub-data;

[0201] The sorting unit 708 is configured to sequentially arrange at least one task compressed data according to the round number of at least one interaction round sub-data in the interaction history information, and generate feature information corresponding to the task to be processed.

[0202] Optionally, the coding unit 706 is further configured as follows:

[0203] Adjust the encoding parameters, and based on the adjusted encoding parameters, regenerate the feature information corresponding to the task to be processed.

[0204] Corresponding to the above method embodiments, this specification also provides embodiments of a text adapter training device. Figure 8 A schematic diagram of a text adapter training device according to one embodiment of this specification is shown. Figure 8 As shown, the device includes:

[0205] The acquisition unit 802 is configured to acquire a sample task, sample context information corresponding to the sample task, and sample result corresponding to the sample task;

[0206] The processing unit 804 is configured to input the sample context information corresponding to the sample task into a text adapter, and obtain the sample history feature information output by the text adapter. The text adapter divides the sample context information based on the interaction rounds of the sample context information, and obtains at least one interaction round sub-data corresponding to the interaction rounds. The at least one interaction round sub-data is encoded to obtain the sample history feature information corresponding to the sample task.

[0207] The prediction unit 806 is configured to input the sample task and the sample historical feature information into the task processing model to obtain the prediction result output by the task processing model.

[0208] Training unit 808 is configured to calculate a loss value based on the prediction result and the sample result, adjust the adapter parameters of the text adapter based on the loss value, and continue training the text adapter until the model training stopping condition is met.

[0209] See Figure 9 , Figure 9 This specification illustrates an architecture diagram of a task processing system provided in one embodiment of the specification. The task processing system may include a client 100 and a server 200.

[0210] Client 100 is used to send the target task to server 200;

[0211] Server 200 is configured to receive a target task and obtain context information corresponding to the target task; obtain task feature information corresponding to the target task and historical feature information corresponding to the context information, wherein the historical feature information is determined based on task compressed data of at least one interaction round sub-data in the context information; input the task feature information and the historical feature information into a task processing model to obtain the target result output by the task processing model, wherein the task processing model determines target historical feature information from the historical feature information based on the task feature information and generates the target result based on the target historical feature information; and send the target result to client 100.

[0212] Client 100 is also used to receive the target result of the target task sent by server 200.

[0213] The task processing system may include multiple clients 100 and a server 200. Clients 100 can be referred to as edge devices, and server 200 can be referred to as cloud devices. Multiple clients 100 can establish communication connections through server 200. In a task processing scenario, server 200 is used to provide task processing services between multiple clients 100. Each client 100 can act as a sender or receiver, communicating through server 200.

[0214] Users can interact with server 200 through client 100 to receive data sent by other clients 100, or send data to other clients 100, etc. In a task processing scenario, users can publish data streams to server 200 through client 100, server 200 can generate task processing based on the data stream, and push the task processing to other clients that have established communication.

[0215] In this system, client 100 and server 200 establish a connection via a network. The network provides the medium for communication between client 100 and server 200. The network can include various connection types, such as wired or wireless communication links or fiber optic cables. Data transmitted by client 100 may need to undergo encoding, transcoding, compression, or other processing before being published to server 200.

[0216] Client 100 can be a browser, an app (application), a web application such as an H5 (HyperText Markup Language 5) application, a lightweight application (also known as a mini-program), or a cloud application. Client 100 can be developed based on the software development kit (SDK) of the corresponding service provided by server 200, such as a real-time communication (RTC) SDK. Client 100 can be deployed on a computing device and depends on the device or certain apps on the device to run. The computing device may have a display screen and support information browsing, such as a personal mobile terminal like a mobile phone, tablet, or personal computer. Various other types of applications can also be configured on the computing device, such as human-computer interaction applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, and social media platform software.

[0217] Server 200 may include servers providing various services, such as servers providing communication services to multiple clients, servers supporting backend training of models used on clients, and servers processing data sent by clients. It should be noted that server 200 can be implemented as a distributed server cluster composed of multiple servers, or as a single server. The server can also be a server in a distributed system, or a server integrated with blockchain. The server can also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0218] It is worth noting that the task processing methods provided in the embodiments of this specification are generally executed by the server. However, in other embodiments of this specification, the client may also have similar functions to the server, thereby executing the task processing methods provided in the embodiments of this specification. In other embodiments, the task processing methods provided in the embodiments of this specification may also be executed jointly by the client and the server.

[0219] Figure 10 A structural block diagram of a computing device according to one embodiment of this specification is shown. The components of the computing device 1000 include, but are not limited to, a memory 1010 and a processor 1020. The processor 1020 is connected to the memory 1010 via a bus 1030, and a database 1050 is used to store data.

[0220] The computing device 1000 also includes an access device 1040, which enables the computing device 1000 to communicate via one or more networks 1060. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 1040 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0221] In one embodiment of this application, the aforementioned components of the computing device 1000 and Figure 10 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 10 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this application. Those skilled in the art can add or replace other components as needed.

[0222] The computing device 1000 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 1000 can also be a mobile or stationary server.

[0223] The processor 1020 is used to execute the following computer program / instructions, which, when executed by the processor, implement the steps of the above-described task processing method.

[0224] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the task processing method described above belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the task processing method described above.

[0225] In one or more embodiments of this specification, the computing device can be understood as an integrated smart terminal, including but not limited to a server, desktop computer, PC (Personal Computer), all-in-one model machine, mobile phone, tablet computer or other portable smart terminal, etc., and the computing device may have the model described in the above embodiments of this application pre-installed.

[0226] Specifically, this computing device can pre-install various types of models, including but not limited to models in natural language processing, visual processing, speech processing, code processing, and multimodal task processing, thus providing diverse model selection. In different product forms, this computing device can support one or more model usage methods, including but not limited to model training, model invocation, model fine-tuning, model deployment, model inference, and application. In some product forms, this computing device also supports model management, including but not limited to multi-type model management (supporting the management of discriminative, generative, and other model types), model version control (supporting the control of different model versions), and model evaluation (evaluating model performance and effectiveness based on model evaluation tools). In other product forms, this computing device can also create applications based on models, providing API (Application Programming Interface) calling capabilities. Users can call models into created applications through the API interface, and application management tools are also provided to manage and monitor the applications.

[0227] Furthermore, the computing device can also include data management (supporting the creation and management of model tuning datasets), a training center (providing abundant training resources to help users learn and master AI (Artificial Intelligence) technology), and basic control capabilities (providing enterprise-level basic control capabilities to ensure the security and efficient operation of the system). Through the above functions, it provides a comprehensive and integrated device for AI development, training, deployment, and application.

[0228] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the task processing method described above belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the task processing method described above.

[0229] An embodiment of this specification also provides a computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the task processing method described above.

[0230] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, the computer-readable storage medium embodiments are described simply because they are substantially similar to the task processing method embodiments; relevant parts can be referred to the descriptions of the task processing method embodiments.

[0231] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described task processing method.

[0232] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the task processing method described above belong to the same concept. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the task processing method described above.

[0233] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0234] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0235] It should be noted that the above description describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments of this specification.

[0236] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0237] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A task processing method, comprising: Receive the target task and obtain the context information corresponding to the target task; Obtain task feature information corresponding to the target task, and obtain historical feature information corresponding to the context information. The historical feature information is determined based on compressed task data of at least one interaction round sub-data in the context information. The compressed task data of at least one interaction round sub-data is generated as follows: obtain interaction round sub-data to be processed, wherein the interaction round sub-data to be processed is any one of at least one interaction round sub-data; compress the interaction round sub-data to be processed based on a compression ratio parameter to obtain at least one compressed data; encode the at least one compressed data based on the target task to generate at least one compressed task data corresponding to the interaction round sub-data to be processed. The task feature information and the historical feature information are input into the task processing model to obtain the target result output by the task processing model. The task processing model determines the target historical feature information from the historical feature information based on the task feature information and generates the target result based on the target historical feature information.

2. The method as described in claim 1, wherein obtaining the context information corresponding to the target task includes: Obtain the task dialogue information corresponding to the target task, and extract the context information from the task dialogue information; And / or, A reference task is determined based on the target task, and the interaction history information corresponding to the reference task is obtained. The interaction history information is used as the context information, wherein the semantic similarity between the first semantic information of the target task and the second semantic information of the reference task is greater than a preset similarity threshold.

3. The method as described in claim 2, wherein the historical feature information corresponding to the context information is obtained using the following method: Determine the interaction round corresponding to the context information; Based on the interaction rounds, the context information is divided to obtain at least one interaction round sub-data corresponding to the interaction rounds; Encode at least one interaction round sub-data to generate at least one task compressed data corresponding to at least one interaction round sub-data; Based on the round number of at least one interaction round sub-data in the context information, at least one task compressed data is sequentially arranged to obtain the historical feature information corresponding to the context information.

4. The method as described in claim 1, wherein the interaction round sub-data to be processed is compressed based on a compression ratio parameter to obtain at least one compressed data, comprising: Based on the compression ratio parameter, the interaction round sub-data to be processed is segmented to obtain at least one sub-data block to be processed corresponding to the interaction round sub-data, wherein the amount of data information in the sub-data block to be processed is less than or equal to the set value of the compression ratio parameter. Compress the data information in at least one sub-data block to be processed to obtain at least one compressed data corresponding to the sub-data of the interaction round to be processed.

5. The method as described in claim 4, wherein data information in at least one sub-data block to be processed is compressed to obtain at least one compressed data, comprising: Obtain initial data information from the initial sub-data block and at least one adjacent data information corresponding to the initial data information, wherein the initial sub-data block is at least one sub-data block to be processed, and the initial data information is any one of at least one data information of the initial sub-data block; Encode the initial data information based on at least one adjacent data information and the initial data information to obtain the initial encoded information corresponding to the initial data information; Encode at least one initial encoding information to obtain compressed data corresponding to the initial sub-data block.

6. The method as described in claim 3, wherein obtaining at least one interaction round corresponding to interaction round sub-data includes: Obtain initial interaction round sub-data and obtain attribute information corresponding to the initial interaction round sub-data, wherein the initial interaction round sub-data is any one of the initial interaction round sub-data corresponding to at least one interaction round, and the attribute information includes at least one of the thinking content, tools used, and analysis results; Based on the attribute information, the initial interaction round sub-data is segmented to obtain the interaction round sub-data.

7. The method as described in claim 3, wherein the task feature information and the historical feature information are input into the task processing model to obtain the target result output by the task processing model, comprising: The task feature information and the historical feature information are input into the task processing model to obtain the initial result processed by the task processing model. Determine whether the initial result meets the set conditions; If the initial result meets the set conditions, the initial result will be taken as the target result; If the initial result does not meet the set conditions, the historical feature information is adjusted until the initial result obtained based on the adjusted historical feature data meets the set conditions, and the initial result that meets the set conditions is taken as the target result.

8. The method of claim 7, wherein adjusting the historical feature information includes: Adjust the encoding parameters for encoding the sub-data of the interaction rounds; Based on the adjusted encoding parameters, the adjusted historical feature information corresponding to the reference task is obtained.

9. The method as described in claim 3, wherein the historical feature information corresponding to the context information is obtained using the following method: Determine the interaction round corresponding to the context information; The context information is input to the text adapter to obtain the historical feature information output by the text adapter, wherein... The text adapter divides the context information based on the interaction rounds, obtains interaction round sub-data corresponding to at least one interaction round, encodes the at least one interaction round sub-data, and obtains historical feature information corresponding to the context information.

10. A data generation method, comprising: Obtain the context information corresponding to the task to be processed, and determine the interaction round corresponding to the context information; Based on the interaction rounds, the initial interaction information is divided, and at least one initial interaction sub-data corresponding to the interaction round is obtained. Encode at least one initial interaction sub-data to generate task compressed data corresponding to at least one initial interaction round sub-data. The task compressed data corresponding to at least one initial interaction round sub-data is generated as follows: obtain interaction round sub-data to be processed, wherein the interaction round sub-data to be processed is any one of at least one initial interaction round sub-data; compress the interaction round sub-data to be processed based on a compression ratio parameter to obtain at least one compressed data; encode the at least one compressed data based on the task to be processed to generate at least one task compressed data corresponding to the interaction round sub-data to be processed. Based on the number of at least one interaction round sub-data in the interaction history information, at least one task compressed data is sequentially arranged to generate feature information corresponding to the task to be processed.

11. The method of claim 10, further comprising: Adjust the encoding parameters, and based on the adjusted encoding parameters, regenerate the feature information corresponding to the task to be processed.

12. A task processing method, applied to cloud-side devices, comprising: The receiving end device sends the target task and obtains the context information corresponding to the target task; Obtain task feature information corresponding to the target task, and obtain historical feature information corresponding to the context information. The historical feature information is determined based on compressed task data of at least one interaction round sub-data in the context information. The compressed task data of at least one interaction round sub-data is generated as follows: obtain interaction round sub-data to be processed, wherein the interaction round sub-data to be processed is any one of at least one interaction round sub-data; compress the interaction round sub-data to be processed based on a compression ratio parameter to obtain at least one compressed data; encode the at least one compressed data based on the target task to generate at least one compressed task data corresponding to the interaction round sub-data to be processed. The task feature information and the historical feature information are input into the task processing model to obtain the target result output by the task processing model. The task processing model determines the target historical feature information from the historical feature information based on the task feature information and generates the target result based on the target historical feature information. The target result is sent to the end-side device.

13. A method for training a text adapter, comprising: Obtain the sample task, the sample context information corresponding to the sample task, and the sample result corresponding to the sample task; The sample context information corresponding to the sample task is input into a text adapter to obtain sample history feature information output by the text adapter. The text adapter divides the sample context information based on the interaction rounds of the sample context information, obtaining at least one interaction round sub-data. The at least one interaction round sub-data is then encoded to obtain sample history feature information corresponding to the sample task. Encoding the at least one interaction round sub-data to obtain sample history feature information corresponding to the sample task includes: obtaining interaction round sub-data to be processed, wherein the interaction round sub-data to be processed is any one of at least one interaction round sub-data; compressing the interaction round sub-data to be processed based on a compression ratio parameter to obtain at least one compressed data; encoding the at least one compressed data based on the sample task to generate at least one task compressed data corresponding to the interaction round sub-data to be processed; and determining the sample history feature information corresponding to the sample task based on the at least one task compressed data. The sample task and the sample historical feature information are input into the task processing model to obtain the prediction result output by the task processing model; The loss value is calculated based on the prediction result and the sample result. The adapter parameters of the text adapter are adjusted according to the loss value, and the text adapter is trained until the model training stops.

14. A task platform, comprising a request interface and a response unit; The request interface is used to receive the target task sent by the end device and obtain the context information corresponding to the target task; The response unit is used to obtain task feature information corresponding to the target task and historical feature information corresponding to the context information, wherein... Historical feature information is determined based on task compressed data of at least one interaction round sub-data in the context information. The task compressed data of at least one interaction round sub-data is generated in the following manner: obtaining interaction round sub-data to be processed, wherein the interaction round sub-data to be processed is any one of at least one interaction round sub-data; compressing the interaction round sub-data to be processed based on a compression ratio parameter to obtain at least one compressed data; encoding the at least one compressed data based on the target task to generate at least one task compressed data corresponding to the interaction round sub-data to be processed; inputting the task feature information and the historical feature information into the task processing model to obtain the target result output by the task processing model, wherein the task processing model determines the target historical feature information in the historical feature information based on the task feature information, and generates the target result based on the target historical feature information.

15. A computing device, comprising: Memory and processor; The memory is used to store computer programs or instructions, and the processor is used to execute the computer programs or instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 13.

16. A computer-readable storage medium storing a computer program or instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 13.

17. A computer program product comprising a computer program or instructions which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 13.

Images