Information processing method, device and electronic device based on large language model
Through a large language model, the fuzzy input in geographic information is processed, missing attribute items are identified and supplemented, and clear task execution information is generated, which solves the efficiency and accuracy problems caused by ambiguity in geographic information processing, and improves the efficiency of information processing and users. Experience.
Patent Information
- Application Number
- CN202411269829.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-11
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-09-11
AI Technical Summary
During the process of geographic information processing, the fuzzyness of the input information leads to a decrease in information processing efficiency and accuracy, waste of computing resources, and poor user experience.
Using an information processing method based on a large language model, by identifying the input information and supplementing the missing attribute items, clear task execution information is generated, and the target task model is determined and feedback information is generated.
It improves the accuracy and efficiency of information processing, reduces the number of conversation interactions, improves the user experience, and ensures the adequacy and accuracy of feedback information.
Smart Images

Figure CN118779542B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, specifically to the fields of natural language processing, deep learning, and more particularly to an information processing method, device, electronic device, storage medium and computer program product based on a large language model. Background Art
[0002] With the rapid growth of data in the field of Geographic Information System (GIS), geographic data obtained from various data sources are usually complex and diverse, which makes the processing process or steps of geographic information too cumbersome, thus affecting the efficiency and accuracy of information processing. Summary of the invention
[0003] The present invention provides an information processing method, device, electronic device, storage medium and computer program product based on a large language model.
[0004] According to one aspect of the present invention, there is provided an information processing method based on a large language model, comprising: in response to obtaining input information, processing the input information using a large language model to determine whether the demand intention represented by the input information is ambiguous; in response to determining that the demand intention represented by the input information is ambiguous, generating task execution information based on the input information, and the demand intention represented by the task execution information is clear; based on the task execution information, determining a target task model associated with the task execution information; based on the target task model and the task execution information, determining a first execution result related to the task execution information; generating first feedback information based on the first execution result, and displaying the first feedback information.
[0005] According to another aspect of the present invention, there is provided an information processing device based on a large language model, comprising: a processing module for, in response to obtaining input information, processing the input information using a large language model to determine whether the demand intention represented by the input information is ambiguous; a generation module for, in response to determining that the demand intention represented by the input information is ambiguous, generating task execution information according to the input information, wherein the demand intention represented by the task execution information is clear; a first determination module for determining, according to the task execution information, a target task model associated with the task execution information; a second determination module for determining, according to the target task model and the task execution information, a first execution result related to the task execution information; a feedback module for generating first feedback information based on the first execution result, and displaying the first feedback information.
[0006] According to another aspect of the present invention, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the method as described above.
[0007] According to another aspect of the present invention, a computer-readable storage medium is provided, on which executable instructions are stored. When the instructions are executed by a processor, the processor executes the method as described above.
[0008] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, which implements the method described above when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0010] Figure 1 is a flowchart of an information processing method based on a large language model according to an embodiment of the present invention;
[0011] Figure 2 is a schematic diagram of a process of constructing conversation sample data according to an embodiment of the present invention;
[0012] Figure 3A and Figure 3B is an application scenario diagram of the information processing method based on a large language model according to an embodiment of the present invention;
[0013] Figure 4 is a block diagram of an information processing device based on a large language model according to an embodiment of the present invention;
[0014] Figure 5 is a block diagram of an electronic device suitable for implementing an information processing method based on a large language model according to an embodiment of the present invention. DETAILED DESCRIPTION
[0015] The technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention and the drawings in the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.
[0016] It should be noted that the sequence numbers of the operations in the following method are only used as representations of the operations for the purpose of description, and should not be regarded as representing the execution order of the operations. Unless explicitly stated, the method does not need to be executed completely in the order shown.
[0017] In addition, in the description of the present invention, the terms “first”, “second”, etc. (if any) are used only to distinguish the description and cannot be understood as indicating or implying relative importance.
[0018] In the technical solution of the present invention, the collection, storage, use, processing, transmission, provision, disclosure and application of the data involved (for example, including but not limited to user personal information) shall comply with the relevant laws and regulations of the relevant countries and regions and shall not violate public order and good morals.
[0019] In the technical solution of the present invention, before obtaining or collecting relevant data, the authorization or consent of the data owner is obtained.
[0020] Figure 1 is a flowchart of an information processing method based on a large language model according to an embodiment of the present invention.
[0021] like Figure 1 As shown, the information processing method 100 based on the large language model includes operations S110 to S150.
[0022] In operation S110 , in response to obtaining input information, the input information is processed using a large language model to determine whether the demand intention represented by the input information is ambiguous.
[0023] In operation S120 , in response to determining that the demand intention represented by the input information is ambiguous, task execution information is generated according to the input information, and the demand intention represented by the task execution information is clear.
[0024] In operation S130 , a target task model associated with the task execution information is determined according to the task execution information.
[0025] In operation S140 , a first execution result related to the task execution information is determined according to the target task model and the task execution information.
[0026] In operation S150, first feedback information is generated based on the first execution result, and the first feedback information is displayed.
[0027] According to an embodiment of the present invention, the input information may be natural language information used to characterize the user's information acquisition needs. The input information may be obtained based on the user's input operation on the information input box in the interactive interface, or may be obtained by recognizing the user's input voice information based on voice recognition technology. The embodiment of the present invention does not limit the way of obtaining the input information.
[0028] In an embodiment of the present invention, the demand intention represented by the input information can be clear or vague. For example, the input information can be "the wheat distribution in the xxx area in 2023", and the input information includes necessary information, such as the time range (2023), the spatial range (xxx area) and the land feature category (wheat), which can clearly represent the user's acquisition demand intention for the wheat distribution in the xxx area in 2023. For example, the input information can be "please provide the distribution of land features", which is vague. It lacks necessary information, such as the specific category of "land features", the spatial range of "land features", etc., which makes it impossible to determine the user's acquisition demand intention.
[0029] It is understandable that the ambiguity of input information will affect the subsequent understanding and execution of the input information, causing a certain deviation between the final execution result and the user's demand intention, which not only affects the efficiency and accuracy of information processing, resulting in a waste of computing resources, but also affects the user's experience. In view of this, after obtaining the input information, the large language model can be used to process the input information to determine whether the demand intention represented by the input information is ambiguous.
[0030] According to an embodiment of the present invention, a large language model (LLM) is a deep learning model trained using a large amount of text data, such as a language model based on a Transformer architecture. A large language model can understand and generate natural language text. A large language model usually contains a large number of parameters and can capture complex patterns and structures in a language, thereby performing well in a variety of natural language processing tasks.
[0031] According to an embodiment of the present invention, preset annotation information can be used to construct dialogue sample data, and a large language model can be pre-trained based on the dialogue sample data, so that the large language model can judge the ambiguity of the demand intention represented by the input information based on the input information expressed in natural language, and if the demand intention represented by the input information is ambiguous, identify the missing attribute items in the input information, and use at least one information interaction between the large language model and the user to supplement and summarize the attribute values of the missing attribute items to obtain task execution information for representing clear demand intentions.
[0032] According to an embodiment of the present invention, the preset annotation information includes, for example, annotation information corresponding to each of a plurality of task instructions. Each task instruction is natural language information for representing the user's information acquisition needs. A plurality of task instructions may include, for example, a task instruction representing a clear demand intention and a task instruction representing a vague demand intention. The annotation information corresponding to each task instruction includes, for example, but is not limited to, the demand intention represented by the task instruction, the judgment content as to whether the demand intention represented by the task instruction is vague, the known attribute item information and the missing attribute item information in the task instruction. Among them, the known attribute item information includes known attribute items, attribute identifiers of known attribute items and attribute values of known attribute items. The missing attribute item information includes missing attribute items, attribute identifiers of missing attribute items and at least one potential attribute value associated with the missing attribute item, etc. A missing attribute item refers to an attribute item missing in the input information, and a known attribute item refers to an attribute item contained in the input information. Unless otherwise specified, the missing attribute items and known attribute items referred to in the present invention are all defined in the same way.
[0033] In an embodiment of the present invention, an attribute identifier is used to characterize the importance of a corresponding attribute item. The importance reflects the information value or criticality of the corresponding attribute item in the task instruction. If the information value or criticality of a certain attribute item in the task instruction is higher, the importance of the attribute item in the task instruction is greater, and vice versa. In one example, the attribute identifier "1~3" can be used to indicate the importance of the attribute item, wherein the larger the numerical value corresponding to the attribute identifier is, the higher the importance of the attribute item is. For example, the attribute identifier "1" is used to identify a general attribute item, the attribute identifier "2" is used to identify a secondary attribute item, and the attribute identifier "3" is used to identify a key attribute item. In another example, the attribute identifier "required parameter" and "optional parameter" can be used to indicate the importance of the attribute item. For example, the attribute identifier "required parameter" can be used to identify a key attribute item, and the attribute identifier "optional parameter" can be used to identify a general attribute item, and so on. It should be understood that the attribute identifier is not limited to the above examples, and can be set according to actual needs, and the present invention does not limit this.
[0034] In an embodiment of the present invention, the known attribute item information and the missing attribute item information in each task instruction can be obtained by marking according to the attribute items (for example, including key attribute items and general attribute items) adapted to the task instruction. The attribute items (for example, including key attribute items and general attribute items) adapted to each task instruction can be set according to the business objectives to which the task instruction belongs, wherein the business objectives are used to describe the business scenarios that match the task instruction. For example, when the business objective to which the task instruction "xxx" belongs is "non-agricultural land identification", the attribute items adapted to the task instruction "xxx" can include, for example, key attribute items "land feature category", "spatial range" and "time", etc., as well as general attribute items "spatial resolution" and "star source", etc. According to the above-mentioned attribute items adapted to the task instruction "xxx", the known attribute item information and the missing attribute item information in the task instruction "xxx" can be marked.
[0035] In the embodiment of the present invention, in the preset annotation information, the judgment content about whether the demand intention represented by each task instruction is vague can be determined, for example, based on whether the key attribute item is missing in the task instruction. If a key attribute item is missing in a task instruction, the demand intention represented by the task instruction is marked as vague; otherwise, the demand intention represented by the task instruction is marked as clear.
[0036] Several examples of annotation information corresponding to task instructions are given below to help those skilled in the art understand the technical solution of the present invention.
[0037] In an example, the annotation information corresponding to the task instruction “identify the images of region A in 2023” can be expressed as follows.
[0038] Task instruction annotation information:
[0039] ——The task instruction “Identify the images of region A in 2023” indicates the requirement to identify the remote sensing images of region A in 2023, which is an ambiguous task instruction.
[0040] ——The known mandatory parameters in the mission instructions include spatial scope (region A) and time (2023).
[0041] ——Missing required parameters include feature categories, examples of which include water areas, woodlands, and grasslands.
[0042] ——Missing optional parameters include spatial resolution and star source information. Examples of star sources include star source 1, star source 2, star source 3, etc., and examples of spatial resolution include 1 meter, 2 meters, 10 meters, etc.
[0043] In another example, the annotation information corresponding to the task instruction "High-resolution images from May to August 2020 are required to identify lakes around area C, the star source is star source x, and the spatial resolution is 0.5 meters" can be expressed as follows.
[0044] Task instruction annotation information:
[0045] The task instruction "High-resolution images from May to August 2020 are needed to identify lakes around area C. The star source is star source x and the spatial resolution is 0.5 meters" indicates the intention of obtaining high-resolution images from May to August 2020 to identify lakes around area C. The specified star source is star source x and the spatial resolution is 0.5 meters. It has provided information about the required parameters (spatial range, time and feature category) and the optional parameters (star source, spatial resolution). This is a clear task instruction.
[0046] ——The known mandatory parameters in the mission instructions include spatial scope (C region), time (May to August 2020) and land feature category (lake).
[0047] ——The known optional parameters in the mission instructions include the star source (star source x) and spatial resolution (0.5 meters).
[0048] ——Missing required and optional parameters: None.
[0049] In an embodiment of the present invention, the dialogue sample data constructed based on the preset annotation information may be a dialogue between some users and a large language model. Figure 2 The process of constructing the dialogue sample data is explained. It should be understood that Figure 2 The contents shown in are merely exemplary, and the task instructions, dialog sample data, etc. in the embodiments of the present invention are not limited thereto.
[0050] like Figure 2 As shown, for the task instruction 211 "Identify the images of area A in 2023", the preliminary thinking 201 of constructing a large language model is preliminarily performed using the annotation information 231 corresponding to the task instruction 211 in the preset annotation information 230 to clarify the ambiguity of the demand intention represented by the task instruction, and if the demand intention represented by the task instruction is ambiguous, identify the missing attribute items in the task instruction, the attribute identifiers of the missing attribute items, and at least one potential attribute value associated with the missing attribute items.
[0051] like Figure 2As shown, according to preliminary thinking 201, the requirement intent represented by task instruction 211 "identify the image of region A in 2023" is vague, and the missing attribute items include the required parameter "land feature category" and the optional parameters "spatial resolution" and "star source". Preliminary thinking 201 also gives at least one potential attribute value associated with each missing attribute item, for example, the potential attribute value associated with the required parameter "land feature category" is water area, woodland, grassland, etc.
[0052] Then, the preliminary thinking 201 is represented based on the prompt template to help or guide the large language model to subsequently obtain the attribute values of the missing attribute items in the task instruction 211 "Identify the images of region A in 2023" through dialogue interaction, which is conducive to more efficient completion of the missing information.
[0053] Next, the large language model is used to generate task interaction information 221 based on the preliminary thinking 201. Figure 2 As shown, task interaction information 221 represents a query for all missing attribute items (including the required parameter "ground feature category" and the optional parameters "spatial resolution" and "star source") in task instruction 211 "identify images of region A in 2023", and provides at least one potential attribute value associated with each missing attribute item. These potential attribute values can provide users with corresponding prompts to guide users to feedback task response information that is more suitable for task interaction information 221, thereby reducing the number of dialogue interactions and improving the efficiency of obtaining missing information.
[0054] Next, the task response information 212 of the user for the task interaction information 221 is obtained. The task response information 212 may include the target attribute values corresponding to all missing attribute items. In some embodiments, the task response information 212 may also include additional attribute values for the known attribute items in the task instruction 211 "identify the images of region A in 2023". The additional attribute values of the known attribute items may be used to correct or supplement the attribute values of the known attribute items, so that the corrected or supplemented attribute values of the known attribute items are more in line with the current intention of the user.
[0055] like Figure 2As shown, the task response information 212 provided by the user, "I need to identify the woodland in region A in December 2023, and the spatial resolution is expected to be about 1 meter. It would be better if there is data of star source 1," not only provides the target attribute values corresponding to all the missing attribute items, such as the target attribute value of "land feature category" is "woodland," the target attribute value of "spatial resolution" is "1 meter," and the target attribute value of "star source" is "star source 1," but also provides the additional attribute value "December 2023" for the known attribute item "time." The additional attribute value "December 2023" is used to correct or supplement the attribute value "2023" of the known attribute item "time" in the task instruction 211 "Identify the image of region A in 2023," so that the corrected attribute value of the known attribute item "time" is compatible with the user's current intention.
[0056] It should be noted that the query for all missing attribute items in the task instruction can also be achieved through multiple rounds of dialogue. In each round of dialogue, only one missing attribute item can be queried and a prompt can be given.
[0057] Next, the task instruction 211 , the task interaction information 221 , and the task response information 212 are used to construct a summary thought.
[0058] like Figure 2 As shown, for example, the key information in the task instruction 211, the task interaction information 221, and the task response information 212 (such as the target attribute values corresponding to all missing attribute items, the attribute values of the known attribute items, and the additional attribute values of the known attribute items) can be used to construct the summary thinking 202. The summary thinking 202 summarizes all the information collected so far (such as the ground feature category, spatial resolution, star source, and time information), and explains why the information collected so far is sufficient to clarify the user's intention.
[0059] Next, the summary thinking 202 is represented based on the prompt template, and the task execution information is generated based on the summary thinking 202 using the large language model.
[0060] like Figure 2 As shown, the task execution information 222 "identify the forest land in area A in December 2023, with a spatial resolution of 1 meter and remote sensing image data of star source 1" contains sufficient and necessary information to more comprehensively, accurately and clearly characterize the user's intention.
[0061] By constructing multiple dialogue sample data through the method described above, and using the multiple dialogue sample data to pre-train the large language model, the large language model can judge the ambiguity of the demand intention represented by the input information based on the input information expressed in natural language, and if the demand intention represented by the input information is ambiguous, identify the missing attribute items in the input information, and use at least one information interaction between the large language model and the user to restore and summarize the attribute values of the missing attribute items in the input information, so as to obtain task execution information for representing clear demand intentions.
[0062] Return to reference Figure 1 According to an embodiment of the present invention, if a large language model is used to process input information, and it is determined that the demand intention represented by the input information is clear, that is, there are no missing key attribute items in the input information, then the input information can be used as task execution information for subsequent processing. If a large language model is used to process input information, and it is determined that the demand intention represented by the input information is fuzzy, that is, the missing attribute items in the input information include key attribute items, then at least one information interaction between the large language model and the user can be used to supplement the attribute values of the missing attribute items in the input information to obtain task execution information, and the demand intention represented by the task execution information is clear. The task execution information can be directly used for subsequent tasks. Thereby, the fuzzy intention represented by the input information is transformed into a clear intention, so that subsequent processing can be performed based on the clear intention, thereby providing a basis for obtaining accurate feedback information.
[0063] According to an embodiment of the present invention, the target task model is an operator workflow that matches the target intent represented by the task execution information. The task execution information can provide corresponding configuration parameters for the target task model. A task flow can be generated based on the target task model and the task execution information, and the task flow is adapted to the target intent. Task processing is performed based on the task flow, and an execution result related to the task execution information can be obtained.
[0064] According to an embodiment of the present invention, a first execution result related to the task execution information is determined based on the target task model and the task execution information, a first task flow can be generated through the target task model and the task execution information, and a first execution result can be obtained by performing task processing based on the first task flow. The first task execution result can include information that matches the target intent represented by the task execution information.
[0065] According to an embodiment of the present invention, generating the first feedback information based on the first execution result may include using the first execution result as the first feedback information, or filling a preset configuration template based on the first execution result to obtain the first feedback information. Of course, other methods may also be used to determine the first feedback information, and the present invention is not limited to this.
[0066] According to the technical solution of the present invention, by using a large language model to process input information, it is determined whether the demand intention represented by the input information is vague, and when it is determined that the demand intention represented by the input information is vague, task execution information is generated based on the input information, and the demand intention represented by the task execution information is clear and accurate. In this way, the fuzzy intention represented by the input information is transformed into a clear intention, so that the task execution information can more fully, comprehensively and accurately represent the user's intention, thereby determining the target task model based on the task execution information, and further generating the first feedback information based on the target task model and the task execution information, so that the first feedback information can more fully and accurately meet the user's demand intention, and improve the accuracy and efficiency of the user's acquisition of information.
[0067] According to an embodiment of the present invention, processing input information using a large language model to determine whether the demand intention represented by the input information is ambiguous may include: using a large language model to perform attribute recognition on the input information to determine missing attribute items in the input information and attribute identifiers of the missing attribute items, wherein the attribute identifiers are used to represent the importance of the missing attribute items; in response to determining that there are missing attribute items in the input information, and the attribute identifiers of the missing attribute items indicate that the missing attribute items are key attribute items, determining that the demand intention represented by the input information is ambiguous.
[0068] In an embodiment of the present invention, a large language model can be used to perform attribute recognition on input information to determine known attribute items, attribute values of known attribute items, attribute identifiers of known attribute items, missing attribute items, and attribute identifiers of missing attribute items in the input information, thereby obtaining attribute recognition results. Then, based on the attribute recognition results, it is determined whether the demand intention represented by the input information is ambiguous.
[0069] In one example, if the attribute recognition result indicates that there is no missing attribute item in the input information, or there is a missing attribute item in the input information but the attribute identifier of the missing attribute item indicates that the missing attribute item is not a key attribute item, it is determined that the requirement intent represented by the input information is clear.
[0070] For example, a large language model can be used to perform attribute recognition on the input information "Identify rice in the xxx area in August 2024, the image spatial resolution is 0.5m, and the star source is star source xx", and the attribute recognition result for the input information is obtained: the known attribute items include the required parameter "time", the required parameter "spatial range", the required parameter "land feature category", the optional parameter "spatial resolution" and the optional parameter "star source", among which the attribute value of the required parameter "time" is "August 2024", the attribute value of the required parameter "spatial range" is "xxx area", the attribute value of the required parameter "land feature category" is "rice", the attribute value of the optional parameter "spatial resolution" is "0.5m", and the attribute value of the optional parameter "star source" is "star source xx". Missing attribute items: None. Among them, "required parameters" are used to identify key attribute items, and "optional parameters" are used to identify general attribute items.
[0071] Based on the above attribute recognition results, it can be seen that there is no missing attribute item in the input information "Identify the rice in xxx area in August 2024, the image spatial resolution is 0.5m, and the star source is star source xx". Therefore, the demand intention represented by the input information "Identify the rice in xxx area in August 2024, the image spatial resolution is 0.5m, and the star source is star source xx" is clear.
[0072] For another example, the large language model can be used to perform attribute recognition on the input information "Please provide images with a spatial resolution of less than 10 meters from 2015 to 2020 to identify wetlands and woodlands in the xxx area", and the attribute recognition results for the input information are obtained: the known attribute items include the required parameter "time", the required parameter "spatial range", the required parameter "land feature category" and the optional parameter "spatial resolution", the attribute value of the required parameter "time" is "2015 to 2020", the attribute value of the required parameter "spatial range" is "xxx area", the attribute value of the required parameter "land feature category" is "wetlands and woodlands", and the attribute value of the optional parameter "spatial resolution" is "less than 10 meters". Missing attribute items include the optional parameter "star source".
[0073] Based on the above attribute recognition results, it can be seen that although there is a missing attribute item in the input information "Please provide images with a spatial resolution of less than 10 meters from 2015 to 2020 for identifying wetlands and woodlands in the xxx area", that is, the optional parameter "star source", the missing attribute item is not a key attribute item. Therefore, the demand intention represented by the above input information "Please provide images with a spatial resolution of less than 10 meters from 2015 to 2020 for identifying wetlands and woodlands in the xxx area" is clear.
[0074] In another example, if the attribute recognition result indicates that there is a missing attribute item in the input information, and the attribute identifier of the missing attribute item indicates that the missing attribute item is a key attribute item, it is determined that the requirement intention represented by the input information is ambiguous.
[0075] For example, the large language model can be used to perform attribute recognition on the input information "obtain remote sensing images of non-cultivated agricultural land in xxx area in 2020", and the attribute recognition results for the input information are obtained: the known attribute items include the required parameters "time" and "spatial range", the attribute value of the required parameter "time" is "2020", and the attribute value of the required parameter "spatial range" is "xxx area". The missing attribute items include the required parameter "land feature category", the optional parameter "star source" and the optional parameter "spatial resolution".
[0076] Based on the above attribute recognition results, it can be seen that there are multiple missing attribute items in the input information "Obtain remote sensing images of non-cultivated agricultural land in xxx area in 2020" (including the required parameter "land feature category", the optional parameter "star source" and the optional parameter "spatial resolution"), and the multiple missing attribute items include a key attribute item (that is, the required parameter "land feature category"). Therefore, the demand intention represented by the input information "Obtain remote sensing images of non-cultivated agricultural land in xxx area in 2020" is vague.
[0077] According to an embodiment of the present invention, when it is determined that the demand intention represented by the input information is vague, the attribute values of the missing attribute items in the input information can be obtained through dialogue interaction between the user and the large language model to obtain task execution information representing a clear intention.
[0078] According to an embodiment of the present invention, generating task execution information according to input information may include, for example, the following process.
[0079] First, interactive information is generated based on the input information and then displayed.
[0080] In an embodiment of the present invention, the preset annotation information can be queried according to the missing attribute item in the input information to obtain the first prompt information, and the first prompt information may include the missing attribute item and at least one potential attribute value corresponding to the missing attribute item. Then, the interaction information is generated according to the first prompt information using the large language model. The interaction information includes the query information for the missing attribute item in the input information and at least one potential attribute value corresponding to the missing attribute item. Afterwards, the interaction information can be displayed through the interactive interface so that the user can browse and input the response information related to the interaction information in a timely manner.
[0081] In the embodiment of the present invention, the first prompt information can help the large language model to understand the meaning expressed by the missing attribute item in the input information, so that the large language model can generate interactive information associated with the missing attribute item instead of information unrelated to the missing attribute item.
[0082] In an embodiment of the present invention, the potential attribute values corresponding to the missing attribute items in the interaction information can provide corresponding prompts for the user to guide the user to feedback response information that is compatible with the interaction information, thereby helping to reduce the number of dialogue interactions and improve the efficiency of obtaining missing information.
[0083] Next, response information to the interactive information is obtained.
[0084] In an embodiment of the present invention, the response information is feedback information related to the interactive information provided by the user. The response information may include target attribute values for each missing attribute item in the input information. In addition, in some embodiments, the response information may also include additional attribute values for known attribute items in the input information. The additional attribute values of the known attribute items may be used to correct or supplement the attribute values of the known attribute items, so that the corrected or supplemented attribute values of the known attribute items are more in line with the user's intention.
[0085] It should be noted that the query for all missing attribute items in the input information can be implemented through one round of dialogue or through multiple rounds of dialogue. When implemented through multiple rounds of dialogue, each round of dialogue can query and give prompts for only one missing attribute item.
[0086] Next, the large language model is used to generate task execution information based on input information, interaction information, and response information.
[0087] In an embodiment of the present invention, a large language model is used to extract key information (including feedback information for missing attribute items and known attribute items) based on input information and contextual information (including interaction information and response information) in the dialogue interaction process, and to determine whether the key information is sufficient to clarify the user's needs and intentions, and when it is determined that the key information can clarify the user's needs and intentions, task execution information is generated based on the key information.
[0088] According to an embodiment of the present invention, determining a target task model associated with the task execution information based on the task execution information includes: using a large language model to perform intent recognition on the task execution information to obtain a target intent; based on the target intent, obtaining second prompt information associated with the target intent; and based on the target intent and the second prompt information, querying from a plurality of preset task models to obtain the target task model.
[0089] According to an embodiment of the present invention, since the task execution information contains key information related to the user's demand intention, the large language model is used to identify the task execution information, and an accurate and true target intention can be obtained. On this basis, the second prompt information is obtained based on the target intention, and the second prompt information can include one or more of the context information, professional knowledge, etc. related to the target intention, and further according to the target intention and the second prompt information, the target task model is queried from the preset multiple task models, so that the target task model is more adapted to the target intention, thereby improving the adaptability of the final execution result to the user's demand intention.
[0090] According to an embodiment of the present invention, the target task model is an operator workflow that matches the target intent represented by the task execution information. The target task model may, for example, include at least one first operator and at least one execution dependency between the first operators. The execution dependency between the first operators may be represented based on the operator execution topology, the nodes in the operator execution topology may represent the first operators, and the edge relationships between the first operator nodes may represent the operator execution dependencies. Among them, the operator execution dependencies include operator parameter dependencies and data dependencies, the operator parameter dependencies refer to the operator parameter transfer relationships between the first operators, and the data dependencies refer to the data transfer relationships between the first operators. Each first operator in the target task model may be a single operator or a combined operator. Among them, a combined operator refers to an operator combination obtained by arranging at least one operator.
[0091] According to an embodiment of the present invention, each first operator may include a first operator identifier and a first basic parameter. The first operator identifier is used to characterize the association relationship between the first operator and the first resource. The first resource, for example, includes but is not limited to at least one of a plug-in tool, an interface tool, a function tool, and a model tool. The first basic parameter is a preset operator configuration parameter associated with the first operator, which may include but is not limited to an operator parameter, an operator parameter type, and an association between operator parameters.
[0092] According to an embodiment of the present invention, determining a first execution result related to task execution information based on a target task model and task execution information includes: generating a first task flow based on the target task model and task execution information; and calling a first resource associated with at least one first operator to perform task processing based on the first task flow to obtain a first execution result.
[0093] According to an embodiment of the present invention, a first task process is generated based on a target task model and task execution information. Key information can be extracted from the task execution information, and the extracted key information is assigned as a first configuration parameter to a corresponding first operator in the target task model to update part or all of the operator parameters in the corresponding first operator, so that the updated operator parameters are adapted to the current task, thereby obtaining a first task process that is adapted to the target intent.
[0094] According to an embodiment of the present invention, the first resource associated with each first operator can be called through the first operator identifier corresponding to each first operator in the first task process, and task processing is performed based on the first task process to obtain a first execution result. The first task execution result may include information that matches the target intent represented by the task execution information. In an embodiment of the present invention, calling the first resource for task processing through the first operator identifier can realize automatic calling of resources, thereby improving information processing efficiency.
[0095] According to an embodiment of the present invention, before calling the first resources associated with at least one first operator to perform task processing based on the first task flow, the information processing method also includes: displaying the first task flow; in response to obtaining correction information for the first target operator in the first task flow, correcting the first target operator based on the correction information to obtain a second task flow; calling the second resources associated with each second operator in the second task flow to perform task processing based on the second task flow to obtain a second execution result; and generating second feedback information based on the second execution result, and displaying the second feedback information.
[0096] According to an embodiment of the present invention, before calling the first resources associated with at least one first operator to perform task processing based on the first task process, the first task process can be displayed through an interactive interface for user confirmation, thereby improving the degree of match between the first task process and the user's actual intention, and further enabling the obtained feedback information to truly meet the user's needs.
[0097] According to an embodiment of the present invention, the user may also input correction information for the first target operator in the first task flow through the interactive interface. The correction information may be, for example, natural language information indicating that any type of operation, such as a deletion operation, a modification operation, or a replacement operation, is performed on the first target operator or the target operator parameter of the first target operator. Of course, the embodiment of the present invention is not limited thereto, and the correction information may also be, for example, natural language information indicating that the execution dependency associated with the first target operator is corrected.
[0098] After obtaining the correction information input by the user, the first target operator can be corrected based on the correction information to obtain a second task flow adapted to the user's current needs, and then second feedback information matching the user's current needs is obtained based on the second task flow.
[0099] According to an embodiment of the present invention, the second resource associated with each second operator can be called through the second operator identifier corresponding to each second operator in the second task flow, and task processing is performed based on the second task flow to obtain a second execution result. The second task execution result includes information that matches the user's current needs. Then, second feedback information is generated based on the second execution result.
[0100] According to an embodiment of the present invention, the second resource includes, for example, but is not limited to, at least one of a plug-in tool, an interface tool, a function tool, and a model tool. In addition, the method for obtaining the second feedback information is the same or similar to the method for obtaining the first feedback information, and will not be repeated here.
[0101] According to an embodiment of the present invention, before calling the first resources associated with at least one first operator to perform task processing based on the first task process, the information processing method also includes: displaying the first task process; in response to a selection operation on a second target operator in the first task process, checking whether the first target resource associated with the second target operator includes dependency information required to run the second target operator, and obtaining a verification result; and in response to the verification result indicating that the first target resource includes dependency information, running the second target operator based on the dependency information to obtain an execution result corresponding to the second target operator; generating third feedback information based on the execution result corresponding to the second target operator, and displaying the third feedback information.
[0102] According to an embodiment of the present invention, before calling the first resources associated with at least one first operator to perform task processing based on the first task flow, the user can also specify to run some operators in the first task flow, thereby controlling the execution of the task flow according to the user's current needs and improving the flexibility of the task flow execution.
[0103] For example, when a user selects a second target operator in a first task flow or triggers a run control associated with the second target operator, it is possible to check whether the first target resource associated with the second target operator includes dependency information required to run the second target operator, and obtain a check result. Here, the dependency information may refer to at least one of input data and configuration parameters required to run the second target operator.
[0104] If the inspection result indicates that the first target resource includes dependency information, it means that the conditions for running the second target operator are now met. The second target operator can be run based on the dependency information to obtain the execution result corresponding to the second target operator, and the third feedback information can be generated and displayed accordingly.
[0105] If the inspection result indicates that the first target resource does not include dependency information, it means that the conditions for running the second target operator are not met at this time. In this case, the second target operator can be stopped from running, or the first resource associated with at least one first operator can be called to perform task processing based on the first task flow to obtain a first execution result, and first feedback information is generated based on the first execution result, and the first feedback information is displayed.
[0106] According to an embodiment of the present invention, in the process of obtaining each execution result, after obtaining each execution result, each execution result can also be displayed through an interactive interface. In some embodiments, each execution result and the corresponding task flow can also be marked and stored in a database for subsequent use.
[0107] Figure 3A and Figure 3B It is an application scenario diagram of the information processing method based on a large language model according to an embodiment of the present invention.
[0108] like Figure 3A As shown, in the application scenario 300, the user can input the input information 311 "view the distribution of land types" into the large language model, so as to use the large language model to process the input information 311 to determine whether the demand intention represented by the input information 311 is ambiguous.
[0109] For example, by performing attribute recognition on the input information 311 based on the large language model, it can be determined that there are missing attribute items in the input information 311: the required parameter "spatial range", the required parameter "time range" and the required parameter "land feature category". Since these missing attribute items are all key attribute items, it can be determined that the demand intention represented by the input information 311 is ambiguous.
[0110] In the case where it is determined that the demand intention represented by the input information 311 is ambiguous, the preset annotation information can be queried according to the missing attribute items in the input information 311 to obtain the first prompt information, which includes each missing attribute item and at least one potential attribute value corresponding to each missing attribute item. For example, the first prompt information includes the missing attribute items: the required parameter "spatial range", the required parameter "time range" and the required parameter "land feature category", multiple potential attribute values corresponding to the required parameter "spatial range" (for example, "Beijing City", "Shandong Province", "Henan Province", etc.), multiple potential attribute values corresponding to the required parameter "time range" (for example, "the first quarter of 2023", "the first half of 2023", "March 2023", etc.) and multiple potential attribute values corresponding to the required parameter "land feature category" (for example, "forest", "building", "road", etc.).
[0111] Next, the large language model may be used to generate interaction information 321 according to the first prompt information, and the interaction information 321 may be displayed through an interactive interface. Figure 3A As shown, the interaction information 321 includes query information for each missing attribute item in the input information 311 and at least one potential attribute value corresponding to each missing attribute item. These information are used to guide the user to feedback response information adapted to the interaction information 321.
[0112] like Figure 3A As shown, the user can input the response information 312 "wheat" through the interactive interface. After obtaining the response information 312 input by the user, the large language model is used to extract key information based on the input information 311, the interactive information 321, and the response information 312 to determine whether the key information is sufficient to clarify the user's needs and intentions. By analyzing this information using the large language model, it can be determined that the key information currently collected only includes the target attribute value "wheat" corresponding to the required parameter "land feature category", and the attribute values corresponding to the required parameters "spatial range" and "time range" are still missing. The key information collected is not sufficient to clarify the user's intentions.
[0113] In this case, the large language model can be used to generate the next round of interaction information based on the remaining missing attribute items (including the required parameter "spatial range" and the required parameter "time range") and at least one potential attribute value corresponding to the missing attribute items, so as to continue to guide the user to give corresponding response information. If the next round of dialogue interaction still does not collect key information that is sufficient to clarify the user's intention, the large language model is continued to be used to generate a new round of interaction information based on the remaining missing attribute items and at least one potential attribute value corresponding to the missing attribute items, so as to continue to guide the user to give corresponding response information. Repeat the above operations until it is determined based on the large language model that the collected key information is sufficient to clarify the user's intention, or until the user's confirmation information for the interaction information that can indicate the user's clear intention is obtained.
[0114] like Figure 3A As shown, after multiple rounds of dialogue interaction between the user and the large language model (for example, including input information 311, interactive information 321 to interactive response information 313), the large language model determines that the collected key information is sufficient to clarify the user's intention. At this time, task execution information 323 "Based on the information you provided, what you want is 'the distribution of wheat in Wuhan in March 2023'?" can be generated based on these key information, and task execution information 323 is displayed for user confirmation. The user can also input adjustment information 314 "No, I want rice" for task execution information 323 through the interactive interface, so as to adaptively adjust the task execution information according to the user's current needs. After receiving the adjustment information 314 input by the user, the large language model can be used to generate new task execution information 324 "Based on the information you provided, what you want is 'the distribution of rice in Wuhan in March 2023'?" based on task execution information 323 and adjustment information 314, and the task execution information 324 is displayed for user confirmation. After receiving the user's confirmation information 315 for the task execution information 324, the large language model can be used to identify the intent of the task execution information 324 to obtain the target intent. Next, the second prompt information is obtained according to the target intent. The second prompt information may include one or more of the context information, professional knowledge, etc. related to the target intent, and further according to the target intent and the second prompt information, the target task model is queried from the preset multiple task models. Next, the key information is extracted from the task execution information 324, and the extracted key information is assigned as the first configuration parameter to the corresponding first operator in the target task model to update part or all of the operator parameters in the corresponding first operator, so that the updated operator parameters are adapted to the current task, thereby obtaining the first task process adapted to the target intent. Then, the first task process is displayed through the interactive interface for user confirmation.
[0115] like Figure 3AAs shown, the interactive interface displays the first task flow 3251. The first task flow 3251 is an operator workflow composed of multiple first operators, and the operator parameters of the corresponding first operators in the first task flow 3251 have been updated based on the task execution information. By performing task processing based on the first task flow 3251, an execution result that matches the user's target intention can be obtained.
[0116] like Figure 3A As shown, the interactive interface also provides a "start running" control, a "task debugging" control, and a "rearrange" control. If the user does not need to modify the displayed first task flow 3251, the first execution result corresponding to the first task flow 3251 can be obtained by triggering the "start running" control. When the user needs to modify the displayed first task flow 3251, the user can debug the first task flow 3251 online by triggering (including clicking or touching) the "task debugging" control, so that the debugged task flow matches the user's current needs. The user can also redefine the task flow that matches the target intention by triggering the "rearrange" control.
[0117] In some embodiments, the user can also modify the first task process 3251 through dialogue interaction so that the modified task process matches the user's current needs.
[0118] like Figure 3B As shown, when the user needs to modify the first target operator (e.g., operator 3) in the displayed first task flow 3251, the user can input the modification information 316 for the first target operator through the interactive interface, "Operator 3 does not meet my needs, please replace it with operator 6". By analyzing the modification information 316 using the large language model, the intention expressed by the modification information 316 can be determined. Afterwards, based on this intention, operator 6 can be called and used to replace the first target operator (e.g., operator 3), and a second task flow 3261 that is adapted to the user's current needs is obtained, and the second task flow 3261 is displayed for the user to confirm.
[0119] like Figure 3B As shown, the user can also input correction information 317 of the target operator parameters of the third target operator (e.g., operator 1) in the second task flow 3261 through the interactive interface, “Please change parameter a of operator 1 to parameter b”. Similarly, the target operator parameters of the third target operator (e.g., operator 1) in the second task flow 3261 can be corrected based on the correction information 317 to obtain a second task flow 3271 that is adapted to the user's current needs, and the second task flow 3271 is displayed for the user to confirm.
[0120] Next, for example, in response to a user triggering operation (such as a click operation, a touch operation, etc.) on the "Start Running" control on the interactive interface, the second resource associated with each second operator (including operator 1, operator 2, and operator 4-6) in the second task flow 3271 can be called through the second operator identifier corresponding to each second operator, and task processing is performed based on the second task flow 3271 to obtain a second execution result. The second task execution result includes information that matches the user's current needs. Then, second feedback information is generated based on the second execution result, and the second feedback information is displayed.
[0121] like Figure 3B As shown, the second feedback information 328 may include, for example, a result viewing entry corresponding to the second execution result. The result viewing entry corresponding to the second execution result may be a hyperlink, or may be a control or component. When the result viewing entry is triggered, the second execution result is displayed. For example, after the result viewing entry is displayed, the user may perform a trigger operation (such as a click operation or a touch operation) on the result viewing entry to trigger the result viewing entry, thereby displaying the second execution result on the interactive interface.
[0122] In some embodiments, the second feedback information 328 may also include query information on whether the second execution result meets the user's needs. The user may input feedback information for the query information through the interactive interface, so that the large language model can reflect and adjust in time based on the feedback information input by the user, so that the adjusted content can meet the user's needs and intentions.
[0123] Figure 4 is a block diagram of an information processing apparatus based on a large language model according to an embodiment of the present invention.
[0124] like Figure 4 As shown, the information processing device 400 based on the large language model includes: a first processing module 410 , a generation module 420 , a first determination module 430 , a second determination module 440 and a first feedback module 450 .
[0125] The first processing module 410 is used to process the input information using a large language model in response to obtaining the input information, so as to determine whether the demand intention represented by the input information is ambiguous.
[0126] The generating module 420 is used to generate task execution information according to the input information in response to determining that the demand intention represented by the input information is vague, and the demand intention represented by the task execution information is clear.
[0127] The first determination module 430 is used to determine a target task model associated with the task execution information according to the task execution information.
[0128] The second determination module 440 is used to determine a first execution result related to the task execution information according to the target task model and the task execution information.
[0129] The first feedback module 450 is used to generate first feedback information based on the first execution result and display the first feedback information.
[0130] According to an embodiment of the present invention, the first processing module 410 includes an attribute identification unit and a determination unit. The attribute identification unit is used to perform attribute identification on the input information using a large language model to determine the missing attribute items and the attribute identifiers of the missing attribute items in the input information, wherein the attribute identifiers are used to characterize the importance of the missing attribute items; the determination unit is used to determine that the demand intention represented by the input information is ambiguous in response to determining that there are missing attribute items in the input information and the attribute identifiers of the missing attribute items indicate that the missing attribute items are key attribute items.
[0131] According to an embodiment of the present invention, the generation module 420 includes a first generation unit, a first acquisition unit, and a second generation unit. The first generation unit is used to generate interaction information according to input information and display the interaction information, the interaction information includes query information for missing attribute items in the input information and at least one potential attribute value corresponding to the missing attribute item; the first acquisition unit is used to obtain response information for the interaction information, the response information includes a target attribute value for the missing attribute item, or a target attribute value for the missing attribute item and an additional attribute value for a known attribute item in the input information; the second generation unit is used to generate task execution information according to the input information, interaction information, and response information using a large language model.
[0132] According to an embodiment of the present invention, the first generation unit includes an acquisition subunit and a generation subunit. The acquisition subunit is used to query preset annotation information according to the missing attribute item in the input information to obtain first prompt information, the first prompt information including the missing attribute item and at least one potential attribute value corresponding to the missing attribute item; the generation subunit is used to generate interaction information according to the first prompt information using a large language model.
[0133] According to an embodiment of the present invention, the first determination module 430 includes: an intention recognition unit, a second acquisition unit and a query unit. The intention recognition unit is used to use a large language model to perform intention recognition on the task execution information to obtain a target intention; the second acquisition unit is used to obtain second prompt information associated with the target intention according to the target intention; the query unit is used to query and obtain a target task model from a plurality of preset task models according to the target intention and the second prompt information.
[0134] According to an embodiment of the present invention, the target task model includes at least one first operator and at least one execution dependency relationship between the first operators; the second determination module 440 includes: a third generation unit and a processing unit. The third generation unit is used to generate a first task flow according to the target task model and the task execution information; the processing unit is used to call the first resource associated with each of the at least one first operator to perform task processing based on the first task flow to obtain a first execution result.
[0135] According to an embodiment of the present invention, the information processing device 400 based on the large language model also includes: a first display module, a correction module, a second processing module, and a second feedback module. The first display module is used to display the first task flow before calling the first resource associated with at least one first operator to perform task processing based on the first task flow; the correction module is used to respond to the correction information obtained for the first target operator in the first task flow, and correct the first target operator based on the correction information to obtain the second task flow; the second processing module is used to call the second resource associated with each second operator in the second task flow to perform task processing based on the second task flow to obtain the second execution result; the second feedback module is used to generate second feedback information based on the second execution result, and display the second feedback information.
[0136] According to an embodiment of the present invention, the information processing device 400 based on the large language model also includes: a second display module, a verification module, an operation module and a third feedback module. The second display module is used to display the first task process before calling the first resource associated with at least one first operator to perform task processing based on the first task process; the verification module is used to respond to the selection operation of the second target operator in the first task process, check whether the first target resource associated with the second target operator includes the dependency information required to run the second target operator, and obtain the verification result; the operation module is used to respond to the verification result indicating that the first target resource includes the dependency information, and run the second target operator based on the dependency information to obtain the execution result corresponding to the second target operator; the third feedback module is used to generate third feedback information based on the execution result corresponding to the second target operator, and display the third feedback information.
[0137] According to an embodiment of the present invention, the first resource or the second resource includes at least one of a plug-in tool, an interface tool, a function tool, and a model tool.
[0138] It should be noted that the implementation methods, technical problems solved, functions realized, and technical effects achieved of each module in the device part embodiment are the same or similar to the implementation methods, technical problems solved, functions realized, and technical effects achieved of each corresponding step in the method part embodiment, and will not be repeated here.
[0139] Figure 5 The block diagram schematically shows an electronic device suitable for implementing an information processing method based on a large language model according to an embodiment of the present invention.
[0140] like Figure 5 As shown, the electronic device 500 according to an embodiment of the present invention includes a processor 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage part 508 to a random access memory (RAM) 503. The processor 501 may include, for example, a general-purpose microprocessor (such as a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (for example, an application-specific integrated circuit (ASIC)), etc. The processor 501 may also include an onboard memory for caching purposes. The processor 501 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present invention.
[0141] In RAM 503, various programs and data required for the operation of electronic device 500 are stored. Processor 501, ROM 502 and RAM 503 are connected to each other via bus 504. Processor 501 performs various operations of the method flow according to the embodiment of the present invention by executing the program in ROM 502 and / or RAM 503. It should be noted that the program can also be stored in one or more memories other than ROM 502 and RAM 503. Processor 501 can also perform various operations of the method flow according to the embodiment of the present invention by executing the program stored in the one or more memories.
[0142] According to an embodiment of the present invention, the electronic device 500 may further include an input / output (I / O) interface 505, which is also connected to the bus 504. The electronic device 500 may further include one or more of the following components connected to the I / O interface 505: an input portion 506 including a keyboard, a mouse, etc.; an output portion 507 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 508 including a hard disk, etc.; and a communication portion 509 including a network interface card such as a LAN card, a modem, etc. The communication portion 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as needed. A removable medium 511, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 510 as needed, so that a computer program read therefrom is installed into the storage portion 508 as needed.
[0143] The present invention also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiment; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the information processing method based on the large language model according to the embodiment of the present invention is implemented.
[0144] According to an embodiment of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present invention, the computer-readable storage medium may include the ROM 502 and / or RAM 503 described above and / or one or more memories other than ROM 502 and RAM 503.
[0145] The embodiment of the present invention also includes a computer program product, which includes a computer program, and the computer program includes a program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the information processing method based on the large language model provided by the embodiment of the present invention.
[0146] The computer program executes the above functions defined in the system / device of the embodiment of the present invention when it is executed by the processor 501. According to the embodiment of the present invention, the system, device, module, unit, etc. described above can be implemented by a computer program module.
[0147] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 509, and / or installed from the removable medium 511. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.
[0148] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 509, and / or installed from the removable medium 511. When the computer program is executed by the processor 501, the above functions defined in the system of the embodiment of the present invention are performed. According to the embodiment of the present invention, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.
[0149] According to an embodiment of the present invention, the program code for executing the computer program provided by the embodiment of the present invention can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).
[0150] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0151] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present invention may be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present invention. In particular, without departing from the spirit and teachings of the present invention, the features described in the various embodiments of the present invention may be combined and / or combined in various ways. All of these combinations and / or combinations fall within the scope of the present invention.
[0152] The embodiments of the present invention are described above. However, these embodiments are only for the purpose of illustration, and are not intended to limit the scope of the present invention. Although each embodiment is described above, it does not mean that the measures in each embodiment cannot be used in combination advantageously. Without departing from the scope of the present invention, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present invention.
Claims
1. An information processing method based on a large language model, characterized in that: include: In response to obtaining the input information, the input information is processed using a large language model to determine whether the demand intention represented by the input information is ambiguous; In response to determining that the demand intention represented by the input information is ambiguous, generating task execution information according to the input information, wherein the demand intention represented by the task execution information is clear; Determining a target task model associated with the task execution information according to the task execution information; Determining a first execution result related to the task execution information according to the target task model and the task execution information; generating first feedback information based on the first execution result, and displaying the first feedback information; Wherein, determining the target task model associated with the task execution information according to the task execution information includes: using the large language model to perform intent recognition on the task execution information to obtain the target intent; obtaining second prompt information associated with the target intent according to the target intent; and obtaining the target task model by querying from a plurality of preset task models according to the target intent and the second prompt information; The target task model includes at least one first operator and an execution dependency relationship between at least one of the first operators; determining a first execution result related to the task execution information according to the target task model and the task execution information includes: generating a first task flow according to the target task model and the task execution information; calling a first resource associated with each of the at least one first operator to perform task processing based on the first task flow to obtain the first execution result; Before calling the first resource associated with at least one of the first operators to perform task processing based on the first task flow, the method further includes: displaying the first task flow; in response to obtaining the correction information for the first target operator in the first task flow, modifying the first target operator based on the correction information to obtain a second task flow; calling the second resource associated with each second operator in the second task flow to perform task processing based on the second task flow to obtain a second execution result; generating second feedback information based on the second execution result, and displaying the second feedback information; Before calling the first resource associated with at least one of the first operators to perform task processing based on the first task process, the method also includes: displaying the first task process; in response to a selection operation on a second target operator in the first task process, checking whether the first target resource associated with the second target operator includes dependency information required to run the second target operator, and obtaining a verification result; in response to the verification result indicating that the first target resource includes the dependency information, running the second target operator based on the dependency information to obtain an execution result corresponding to the second target operator; generating third feedback information based on the execution result corresponding to the second target operator, and displaying the third feedback information.
2. The method according to claim 1, characterized in that The step of processing the input information by using a large language model to determine whether the demand intention represented by the input information is ambiguous includes: Using the large language model to perform attribute recognition on the input information to determine missing attribute items in the input information and attribute identifiers of the missing attribute items, wherein the attribute identifiers are used to represent the importance of the missing attribute items; In response to determining that there is a missing attribute item in the input information, and the attribute identifier of the missing attribute item indicates that the missing attribute item is a key attribute item, it is determined that the demand intention represented by the input information is ambiguous.
3. The method according to claim 2, characterized in that Generating task execution information according to the input information includes: Generate interaction information according to the input information, and display the interaction information, wherein the interaction information includes query information for missing attribute items in the input information and at least one potential attribute value corresponding to the missing attribute item; Acquire response information for the interaction information, the response information including a target attribute value for the missing attribute item, or a target attribute value for the missing attribute item and an additional attribute value for a known attribute item in the input information; The large language model is used to generate task execution information according to the input information, the interaction information and the response information.
4. The method according to claim 3, characterized in that Generating interaction information according to the input information includes: Querying preset annotation information according to the missing attribute item in the input information to obtain first prompt information, where the first prompt information includes the missing attribute item and at least one potential attribute value corresponding to the missing attribute item; The interaction information is generated according to the first prompt information using the large language model.
5. The method according to claim 1, characterized in that The first resource or the second resource includes at least one of a plug-in tool, an interface tool, a function tool, and a model tool.
6. An information processing device based on a large language model, characterized in that: include: A processing module, configured to, in response to obtaining input information, process the input information using a large language model to determine whether the demand intention represented by the input information is ambiguous; A generating module, configured to generate task execution information according to the input information in response to determining that the demand intention represented by the input information is ambiguous, wherein the demand intention represented by the task execution information is clear; A first determining module, configured to determine a target task model associated with the task execution information according to the task execution information; A second determination module, configured to determine a first execution result related to the task execution information according to the target task model and the task execution information; A feedback module, configured to generate first feedback information based on the first execution result, and display the first feedback information; The first determination module includes: an intention recognition unit, used to use the large language model to perform intention recognition on the task execution information to obtain a target intention; a second acquisition unit, used to obtain second prompt information associated with the target intention according to the target intention; a query unit, used to query and obtain the target task model from a plurality of preset task models according to the target intention and the second prompt information; The target task model includes at least one first operator and an execution dependency relationship between at least one of the first operators; the second determination module includes: a third generation unit, configured to generate a first task flow according to the target task model and the task execution information; a processing unit, configured to call a first resource base associated with each of the at least one first operator; Performing task processing in the first task flow to obtain the first execution result; The device further includes: a first display module, used for displaying the first task flow before calling the first resource associated with at least one of the first operators to perform task processing based on the first task flow; a correction module, used for, in response to obtaining correction information for the first target operator in the first task flow, correcting the first target operator based on the correction information to obtain a second task flow; a second processing module, used for calling the second resources associated with each second operator in the second task flow to perform task processing based on the second task flow to obtain a second execution result; a second feedback module, used for generating second feedback information based on the second execution result, and displaying the second feedback information; The device also includes: a second display module, which is used to display the first task process before calling the first resource associated with at least one of the first operators to perform task processing based on the first task process; a verification module, which is used to respond to the selection operation of the second target operator in the first task process, verify whether the first target resource associated with the second target operator includes the dependency information required to run the second target operator, and obtain a verification result; an operation module, which is used to respond to the verification result indicating that the first target resource includes the dependency information, and run the second target operator based on the dependency information to obtain an execution result corresponding to the second target operator; and a third feedback module, which is used to generate third feedback information based on the execution result corresponding to the second target operator, and display the third feedback information.
7. An electronic device, characterized in that: include: one or more processors; a memory for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are caused to execute the method according to any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that: Executable instructions are stored thereon, and when the instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 5.
9. A computer program product, characterized in that comprising a computer program, said The computer program implements the method according to any one of claims 1 to 5 when executed by a processor.
Citation Information
Patent Citations
Task processing method and dialogue task processing method
CN115934275A
Intelligent dialogue processing method and device, equipment and storage medium
CN116737910A
Information interaction method and device, electronic equipment and storage medium
CN117690002A
Task processing method, resource registration method, task updating method and electronic equipment
CN117950822A
Task-based dialogue method, storage medium and computer equipment
CN118377886A
