Content reasoning method and device, equipment and medium

Abstract

The embodiment of the invention provides a content reasoning method and device, equipment and a medium. The method comprises the steps of obtaining a task description text associated with a reasoning task; based on the text key content in the task description text, determining a task cognition template adapted to the reasoning task, and based on multiple pieces of template chain information in the task cognition template, determining a reasoning path associated with the reasoning task; when the reasoning task is executed according to the reasoning path, mapping the text key content to analysis class chain information, performing reasoning analysis on the analysis class chain information mapped with the text key content to obtain a content reasoning result, and performing result verification on the content reasoning result through verification class chain information to obtain a verification result; and determining a task reasoning result for the reasoning task according to the content reasoning result and the verification result. By adopting the embodiment of the invention, the inference analysis capability can be improved from the source, and the result output accuracy can be improved through a result verification mode.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to a content inference method, apparatus, device, and medium. Background Art

[0002] At present, an inference model refers to a large language model with reasoning capabilities. When using the large speech model to perform reasoning tasks, the large language model can output the answer to the reasoning task (i.e., the inference result) based on the matching pattern memorized during model training (for example, the question-answer matching pattern).

[0003] However, the inventors have found in practice that when using a large language model to perform reasoning tasks, the large language model often needs to rely on the memorized question-answer matching pattern and output the answer matched according to the question-answer matching pattern as the reasoning result. This means that when the large language model is used to process some complex, untrained reasoning tasks, the reasoning and analysis capabilities of the large language model may be reduced due to the insufficient model generalization ability of the large speech model, which in turn leads to the problem of inaccurate result output of the reasoning results output by the large language model. Therefore, how to improve the reasoning and analysis capabilities and enhance the accuracy of the result output for the currently acquired reasoning tasks is a technical problem that needs to be solved urgently. Summary of the Invention

[0004] The embodiments of the present application provide a content reasoning method, apparatus, device and medium, which can flexibly determine a task cognition template suitable for the current reasoning task based on the key content of the text in the task description text, and then activate the task reasoning capability through multiple template chain information in the task cognition template. When the reasoning path for processing the reasoning task is specifically determined through the task cognition template, the execution of the task can be reasoned step by step according to the reasoning path. In this way, not only can the reasoning analysis capability be improved from the root, but also the accuracy of the result output can be improved through result verification.

[0005] An embodiment of the present application provides a content reasoning method, including:

[0006] Obtaining task description text information associated with the reasoning task;

[0007] Determine a task cognitive template suitable for the reasoning task based on key text content in the task description text information, and determine a reasoning path associated with the reasoning task based on multiple reasoning chain information in the task cognitive template; the multiple reasoning chain information includes at least analysis chain information for performing reasoning analysis and verification chain information for performing reasoning verification;

[0008] When executing the reasoning task according to the reasoning path, the key content of the text is mapped to the analysis chain information, and the analysis chain information mapped with the key content of the text is subjected to reasoning analysis to obtain the content reasoning result. The content reasoning result is verified by the verification chain information to obtain the verification result.

[0009] The task reasoning results for the reasoning task are determined through the content reasoning results and verification results.

[0010] An embodiment of the present application provides a content inference device, including:

[0011] A text acquisition module, used to acquire task description text associated with the reasoning task;

[0012] A template determination module is configured to determine a task cognition template suitable for the reasoning task based on key textual content in the task description text, and to determine a reasoning path associated with the reasoning task based on multiple template chain information in the task cognition template; the multiple template chain information in the task cognition template at least includes analysis chain information for performing reasoning analysis and verification chain information for performing reasoning verification;

[0013] The content analysis module is used to map the key content of the text to the analysis chain information when performing the reasoning task according to the reasoning path, perform reasoning analysis on the analysis chain information mapped with the key content of the text, obtain the content reasoning result, and verify the content reasoning result through the verification chain information to obtain the verification result;

[0014] The result verification module is used to determine the task reasoning result for the reasoning task through the content reasoning result and the verification result.

[0015] The method is executed by a business client, on which a target task reasoning model is deployed;

[0016] Among them, the text acquisition module includes:

[0017] A text acquisition unit, configured to display an input display page associated with the reasoning task;

[0018] The text acquisition unit is further configured to transmit the query data information to the target task reasoning model in response to a confirmation operation on the query data information entered in the input display page;

[0019] The text acquisition unit is further configured to perform information recognition on the query data information by the target task reasoning model to obtain the data information type of the query data information;

[0020] The text acquisition unit is further configured to, if the data information type includes a text data type and a non-text data type, acquire first query data content corresponding to the non-text data type from the query data information, acquire second query data content corresponding to the text data type, perform content conversion processing on the first query data content, and obtain first query conversion content for the first query data content;

[0021] The text acquisition unit is further configured to determine a task description text associated with the reasoning task through the first query conversion content and the second query data content.

[0022] Optionally, the text acquisition module further includes:

[0023] The text acquisition unit is further configured to determine the query data information as a task description text if the data information type includes a text data type.

[0024] Optionally, the text acquisition module further includes:

[0025] The text acquisition unit is further configured to perform content conversion processing on the query data information if the data information type includes a non-text data type, obtain second query conversion content for the query data information, and use the second query conversion content as the task description text.

[0026] Optionally, the text acquisition module further includes:

[0027] The result display unit is used to output the task reasoning results to the result display page of the business client; the result display page includes a first display area and a second display area; the first display area is used to display the content reasoning results; the second display area includes a verification result display area, and the verification result display area is used to display the verified results.

[0028] Optionally, the text acquisition module further includes:

[0029] The result display unit is further configured to display feedback prompt information for providing result feedback on the result display page of the business client; the feedback prompt information includes first-category feedback information;

[0030] The result display unit is further configured to determine the content reasoning result as a task execution result for the reasoning task in response to a feedback triggering operation for the first type of feedback information;

[0031] The result display unit is further used to output the task execution result to the result display page of the business client; the result display page includes a third display area; the third display area is used to display the task execution result.

[0032] Among them, the feedback prompt information includes the second type of feedback information;

[0033] Optionally, the text acquisition module further includes:

[0034] The result display unit is further configured to respond to a feedback triggering operation for the second type of feedback information, rewrite the task reasoning result based on the second type of feedback information, and obtain an optimized task reasoning result;

[0035] The result display unit is further used to determine the optimization task reasoning result as the task execution result for the reasoning task;

[0036] The result display unit is further used to output the task execution result to the result display page of the business client; the result display page includes a third display area; the third display area is used to display the task execution result.

[0037] Optionally, the text acquisition module further includes:

[0038] The result display unit is further configured to display feedback text content for providing result feedback on the result display page of the business client in response to a feedback operation on the verification result; the feedback text content includes first-category feedback content;

[0039] The result display unit is further configured to determine the content reasoning result as a task execution result for the reasoning task based on the first type of feedback content;

[0040] The result display unit is further used to output the task execution result to the result display page of the business client; the result display page includes a third display area; the third display area is used to display the task execution result.

[0041] Among them, the feedback text content includes the second type of feedback content;

[0042] Optionally, the text acquisition module further includes:

[0043] The result display unit is further used to rewrite the task reasoning result based on the second type of feedback content to obtain an optimized task reasoning result;

[0044] The result display unit is further used to determine the optimization task reasoning result as the task execution result for the reasoning task;

[0045] The result display unit is further used to output the task execution result to the result display page of the business client; the result display page includes a third display area; the third display area is used to display the task execution result.

[0046] The template determination module includes:

[0047] A template determination unit is used to perform word segmentation extraction on the task description text to obtain multiple word segmentation texts for the task description text, and to select a target word segmentation text that matches the reasoning task from the multiple word segmentation texts;

[0048] The template determination unit is further configured to use the target segmented text as the text key content, perform content analysis on the text key content, and obtain content association relationships between the segmented texts in the text key content;

[0049] The template determination unit is further configured to determine a task cognitive template adapted to the reasoning task from a plurality of cognitive templates associated with the reasoning task based on the text key content and content association relationships.

[0050] The template determination module further includes:

[0051] The template determination unit is further used to determine the task type of the reasoning task based on the key content and content association relationship of the text;

[0052] The template determination unit is further used to obtain multiple cognitive templates that match the task type of the reasoning task and obtain key content of template analysis in each cognitive template;

[0053] The template determination unit is further configured to perform content matching between the text key content and the template analysis key content in each cognitive template to obtain a content matching degree associated with each cognitive template; a content matching degree is used to represent the degree of content matching between the template analysis key content in a cognitive template and the text key content;

[0054] The template determination unit is further configured to obtain a maximum content matching degree from the content matching degrees, and determine a task cognitive template adapted to the reasoning task through the cognitive template corresponding to the maximum content matching degree.

[0055] The cognitive template corresponding to the maximum content matching degree includes a first cognitive template and a second cognitive template;

[0056] The template determination unit is further configured to determine the template text confidence of the first cognitive template as the first template text confidence, and determine the template text confidence of the second cognitive template as the second template text confidence; the first template text confidence is determined by the content confidence of the first template analysis key content in the first cognitive template; the second template text confidence is determined by the content confidence of the second template analysis key content in the second cognitive template;

[0057] The template determination unit is further configured to obtain a minimum template text confidence from the first template text confidence and the second template text confidence, and use the cognitive template corresponding to the minimum template text confidence as the task cognitive template.

[0058] Optionally, the template determination module further includes:

[0059] The template determination unit is further configured to obtain a general cognitive template when multiple cognitive templates matching the task type of the reasoning task are not obtained, and determine the general cognitive template as the task cognitive template adapted to the reasoning task.

[0060] The template determination module includes:

[0061] A path generation unit, configured to obtain a plurality of template chain information from the task recognition template and obtain an information coding identifier of each template chain information;

[0062] The path generation unit is further configured to generate a link node corresponding to each template chain information according to the information coding identifier of each template chain information;

[0063] The path generation unit is further configured to generate a reasoning path associated with the reasoning task through the link nodes corresponding to each template chain information.

[0064] Optionally, the template determination module further includes:

[0065] A path optimization unit is used to obtain each content text unit in the key content of the text and determine the unit confidence of each content text unit;

[0066] The path optimization unit is further used to determine the text confidence of the task description text through the unit confidence of each content text unit, and to determine the task complexity of the task description text through the text complexity; the value of the task complexity is used to represent the difficulty of querying the task description text;

[0067] The path optimization unit is also used to optimize the reasoning path when the task complexity reaches the complex optimization condition, and obtain the reasoning path after the path optimization;

[0068] The path optimization unit is further configured to execute the reasoning task according to the updated reasoning path when the reasoning path is updated by the reasoning path after the path optimization.

[0069] The template determination module further includes:

[0070] A confidence determination unit, used to obtain a minimum text division unit for performing unit division;

[0071] The confidence determination unit is further used to divide the key text content into units based on the minimum text division unit to obtain each content text unit in the key text content;

[0072] The confidence determination unit is further configured to obtain the sample confidence of the sample text unit that matches each content text unit, and use the sample confidence of each sample text unit as the unit confidence of each content text unit.

[0073] The reasoning path includes a plurality of link nodes; the complex optimization condition includes a first confidence threshold for adding nodes and a second confidence threshold for deleting nodes; the first confidence threshold is less than the second confidence threshold;

[0074] The template determination module further includes:

[0075] The path optimization unit is further configured to, if the value of the task complexity is less than or equal to a first confidence threshold, obtain template-added chain information for link optimization, and add a link node corresponding to the template-added chain information to the multiple link nodes included in the reasoning path to obtain a first reasoning path;

[0076] The path optimization unit is further configured to, if the value of the task complexity is greater than or equal to a second confidence threshold, obtain template deletion chain information for link optimization from the plurality of template chain information, and remove the link node corresponding to the template deletion chain information from the plurality of link nodes included in the reasoning path to obtain a second reasoning path;

[0077] The path optimization unit is further configured to use the first reasoning path or the second reasoning path as the reasoning path after path optimization.

[0078] The analysis chain information includes first analysis chain information and second analysis chain information; the reasoning path includes at least a first link node corresponding to the first analysis chain information and a second link node corresponding to the second analysis chain information;

[0079] The content analysis module includes:

[0080] a content analysis unit, configured to use information content in the first analysis class chain information as first node key content of the first link node when performing the reasoning task according to the reasoning path;

[0081] The content analysis unit is further configured to determine a location of content to be supplemented from the first node key content of the first link node, use the missing content at the location of the content to be supplemented as the first content to be processed, and use the content of the first node key content other than the first content to be processed as the second content to be processed;

[0082] The content analysis unit is further configured to update the first to-be-processed content at the position of the to-be-supplemented content to the key text content when mapping the text key content to the first analysis class chain information, and obtain task mapping content information for the first analysis class chain information through the key text content and the second to-be-processed content;

[0083] The content analysis unit is also used to use the information content in the second analysis class chain information as the second node key content of the second link node, and perform reasoning analysis on the task mapping content information through the task analysis strategy indicated by the second node key content to obtain content reasoning results.

[0084] The reasoning path also includes a third link node corresponding to the verification class chain information;

[0085] Optionally, the result verification module further includes:

[0086] a result determination unit, configured to use the information content in the verification chain information as the third node key content of the third link node, and determine the analysis strategy indicated by the third node key content as the verification analysis strategy;

[0087] The result determination unit is further configured to, if the verification result indicates that there is an optimizable reasoning result in the content reasoning result, optimize the optimizable reasoning result in the content reasoning result according to the verification analysis strategy to obtain an optimized content reasoning result;

[0088] The result determination unit is further configured to determine the optimized content reasoning result as a task reasoning result for the reasoning task.

[0089] Optionally, the result verification module further includes:

[0090] A comparison relationship table generating unit, configured to generate a content comparison relationship table for comparing content reasoning results and optimizing content reasoning results;

[0091] The comparison relationship table display unit is used to output the content comparison relationship table to the result display page of the business client; the result display page includes a second display area; the second display area includes a comparison relationship table display area; the comparison relationship table display area is used to display the content comparison relationship table.

[0092] Among them, the task reasoning result is obtained by the target task reasoning model according to the reasoning path of the task cognitive template; the target task reasoning model is obtained by training the initial task reasoning model;

[0093] Optionally, the device further comprises:

[0094] A model training module is used to obtain training sample information associated with a target sample reasoning task. The training sample information includes a sample task description text used to train the initial task reasoning model and a reference task reasoning result for the sample task description text. The sample reasoning task corresponding to the sample task description text is the target sample reasoning task.

[0095] The model training module is further configured to determine a sample task cognitive template adapted for a target sample reasoning task based on key content of the sample text in the sample task description text, and to determine a sample reasoning path based on multiple template chain information in the sample task cognitive template; the multiple template chain information at least includes sample analysis chain information for performing reasoning analysis and sample verification chain information for performing reasoning verification;

[0096] The model training module is further used to map the key content of the sample text to the sample analysis class chain information when executing the target sample reasoning task according to the sample reasoning path, perform reasoning analysis on the sample analysis class chain information mapped with the key content of the sample text to obtain the sample content reasoning result, and verify the sample content reasoning result through the sample verification class chain information to obtain the sample verification result;

[0097] The model training module is also used to determine the sample task reasoning results for the target sample reasoning task through the sample content reasoning results and the sample verification results, and to train the initial task reasoning model through the sample task reasoning results and the reference task reasoning results to obtain the target task reasoning model.

[0098] The model training module includes:

[0099] The sample determination unit is used to obtain N to-be-screened task description texts associated with multiple sample reasoning tasks, perform vector representation processing on the N to-be-screened task description texts, and obtain N sample text representation vectors; one sample text representation vector is used to represent one to-be-screened task description text; N is a positive integer;

[0100] The sample determination unit is further configured to calculate the vector similarity between any two sample text representation vectors from the N sample text representation vectors to obtain M vector similarities, screen at least one target vector similarity whose vector similarity is greater than or equal to a vector similarity threshold from the M vector similarities, and determine the two to-be-screened task description texts associated with each screened target vector similarity as target screening task description texts, respectively; M is an integer greater than or equal to N;

[0101] The sample determination unit is further configured to obtain the sample text confidence of each determined target screening task description text, and among the sample text confidences of each target screening task description text, use the target screening task description text corresponding to the minimum sample text confidence as the sample task description text, determine the sample reasoning task corresponding to the sample task description text as the target sample reasoning task, and determine the task reasoning result corresponding to the sample task description text as the reference task reasoning result; a sample text confidence is used to characterize the task difficulty of the sample reasoning task corresponding to a target screening task description text;

[0102] The sample determination unit is further used to construct training sample information associated with the target sample reasoning task based on the sample task description text and the reference task reasoning result.

[0103] In one aspect, an embodiment of the present application provides a computer device, including a memory and a processor, wherein the memory is connected to the processor, the memory is used to store a computer program, and the processor is used to call the computer program so that the computer device executes the method provided in the above aspect of the embodiment of the present application.

[0104] On one hand, an embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. The computer program is suitable for being loaded and executed by a processor, so that a computer device with a processor executes the method provided in the above aspect of the embodiment of the present application.

[0105] According to one aspect of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method provided in the above aspect.

[0106] The embodiment of the present application can obtain the task description text information associated with the reasoning task for any reasoning task, and then can flexibly determine the cognitive template (i.e., task cognitive template) that is suitable for the reasoning task based on the keywords (i.e., text key content) identified from the task text description information. Further, the embodiment of the present application can activate the task reasoning ability through the determined cognitive template (i.e., task cognitive template). For example, the embodiment of the present application can determine the reasoning path associated with the reasoning task based on the multiple reasoning chain information in the cognitive template (i.e., task cognitive template), and then can guide the execution of the reasoning task step by step according to the reasoning path. Among them, it should be noted that the multiple reasoning chain information in the cognitive template (i.e., task cognitive template) can at least include analysis class chain information for performing reasoning analysis and verification class chain information for performing reasoning verification. In other words, the embodiment of the present application can use a cognitive template (i.e., a task cognitive template) to specifically determine the reasoning path for processing the reasoning task. In this way, when executing the reasoning task according to the reasoning path, the keywords of the task description text information (i.e., the key content of the text) can be first mapped to the analysis chain information, and then the analysis chain information mapped with the key content of the text can be reasoned and analyzed to obtain the content reasoning result. Furthermore, in order to determine whether the current content reasoning result is correct, the embodiment of the present application proposes that the content reasoning result can be verified by using the verification chain information to obtain the verification result for the content reasoning result. Then, the embodiment of the present application can finally determine the task reasoning result for the reasoning task through the content reasoning result and the verification result. It can be seen that the embodiment of the present application provides a new way to activate task reasoning capabilities through task cognitive templates, so that when the reasoning path for processing the reasoning task is determined specifically through the task cognitive template, the execution of the reasoning task can be further guided step by step through the reasoning path determined by the task cognitive template. In this way, not only can the reasoning and analysis capabilities be improved from the root, but also the accuracy of the result output can be improved through result verification. BRIEF DESCRIPTION OF THE DRAWINGS

[0107] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0108] Figure 1 This is a schematic diagram of a network architecture provided by an embodiment of the present application;

[0109] Figure 2This is a schematic diagram of an interactive scenario of a content reasoning method provided in an embodiment of the present application;

[0110] Figure 3 This is a flowchart of a content reasoning method provided in an embodiment of the present application;

[0111] Figure 4 This is a schematic diagram of an input display page provided in an embodiment of the present application;

[0112] Figure 5 This is a schematic diagram of a scenario for determining a task description text based on query data information of non-text data types and text data types provided by an embodiment of the present application;

[0113] Figure 6 This is a schematic diagram showing a task reasoning result provided by an embodiment of the present application;

[0114] Figure 7 This is a schematic diagram of a task execution result provided by an embodiment of the present application;

[0115] Figure 8 This is another schematic diagram showing the results of task execution provided by an embodiment of the present application;

[0116] Figure 9 This is a schematic diagram of a scenario for determining the content association relationship of key contents of a text provided by an embodiment of the present application;

[0117] Figure 10 is a schematic diagram of a cognitive template (i.e., a task cognitive template) provided in an embodiment of the present application;

[0118] Figure 11 This is a schematic diagram of determining a cognitive template (i.e., a task cognitive template) provided in an embodiment of the present application;

[0119] Figure 12 This is a schematic diagram of generating a reasoning path provided by an embodiment of the present application;

[0120] Figure 13 This is a schematic diagram of determining unit confidence provided by an embodiment of the present application;

[0121] Figure 14 This is a schematic diagram of a scenario of a task mapping content information generation process provided by an embodiment of the present application;

[0122] Figure 15 This is a schematic diagram of a scenario of a content inference result generation process provided by an embodiment of the present application;

[0123] Figure 16 This is a schematic diagram of a task reasoning result generation process provided by an embodiment of the present application;

[0124] Figure 17 This is a flowchart of a content reasoning method provided in an embodiment of the present application;

[0125] Figure 18 This is a schematic diagram of determining a sample task description text provided in an embodiment of the present application;

[0126] Figure 19 This is another schematic diagram showing the results of task execution provided by an embodiment of the present application;

[0127] Figure 20 This is a schematic diagram of the structure of a content inference device provided in an embodiment of the present application;

[0128] Figure 21 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0129] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0130] 1. Task Reasoning Model. The task reasoning model involved in the embodiments of this application includes an initial task reasoning model and a target task reasoning model. The task reasoning model is a large language model (LLM) with reasoning capabilities. Its core is to generate answers through multi-step reasoning when faced with complex task scenarios.

[0131] In the embodiment of the present application, the task reasoning model that has not been trained with training sample information is called the initial task reasoning model, and is used to determine the cognitive template (i.e., task cognitive template) corresponding to the reasoning task based on the keywords (i.e., text key content) in the task description text information when the task description text information associated with the reasoning task is obtained, and then determine the reasoning path through the cognitive template (i.e., task cognitive template), and further, perform reasoning analysis on the keywords (i.e., text key content) through the reasoning path to obtain content reasoning results, and then perform result verification on the content reasoning results to obtain verification results, so that the task reasoning model that determines the task reasoning results for the reasoning task through the content reasoning results and the verification results is called the target reasoning model.

[0132] Specifically, for example, in a business scenario where the task description text is about planning a travel route, the target task reasoning model obtains the task description text containing the business requirements of the business object (e.g., user U1), for example, planning a four-day, three-night itinerary from location A to location B with a budget of 2,000. The target task reasoning model extracts keywords (i.e., key content of the text), such as location A to location B, budget of 2,000, and four days and three nights. Furthermore, the target task reasoning model obtains a task cognition template of the travel planning type based on the keywords (i.e., key content of the text), and then determines a reasoning path through the task cognition template. The target task reasoning model then performs reasoning analysis on the keywords (i.e., key content of the text) according to the reasoning path, obtaining a four-day, three-night travel route plan from location A to location B with a budget of 2,000 (i.e., content reasoning result). Furthermore, the target task reasoning model verifies the travel route plan (i.e., content reasoning result) to obtain a verification result. Furthermore, the target task reasoning model can determine the final reasonable four-day and three-night travel route planning from location A to location B with a budget of 2,000 (i.e., task reasoning result) through content reasoning results and verification results.

[0133] II. Cognitive Templates. The cognitive templates involved in the embodiments of this application include task cognitive templates, first cognitive templates, second cognitive templates, general cognitive templates, and sample task cognitive templates. A task cognitive template is defined as a model specifically designed for handling complex reasoning tasks. Its core is to generate answers through multi-step reasoning, highly reproducing the human thought process.

[0134] In the embodiment of the present application, the cognitive templates corresponding to the template analysis key contents that have the highest matching degree with the keywords of the task description text (i.e., the key contents of the text) are referred to as the first cognitive template and the second cognitive template, and the cognitive template applicable to the keywords of the task description text that has not been matched with it (i.e., the key contents of the text) is referred to as the general cognitive template, and the cognitive template that matches the keywords of the sample task description text (i.e., the key contents of the sample text) is referred to as the sample task cognitive template.

[0135] Specifically, for example, in a business scenario where the task description text is to formulate a four-day and three-night travel route from location A to location B with a budget of 2,000, the embodiment of the present application obtains keywords (i.e., key content of the text) that match the task description text, for example, multiple task cognitive templates of travel planning types that match location A to location B, a budget of 2,000, and four days and three nights. Furthermore, an inference path for generating a travel route is generated through multiple template chain information of the cognitive template task cognitive template.

[0136] 3. Template chain information. The template chain information involved in the embodiments of this application includes analysis chain information and verification chain information. In the context of task recognition templates, template chain information refers to a series of interconnected or dependent elements, steps, functions, or objects that form a logical sequence.

[0137] In the embodiment of the present application, the template chain information used for reasoning analysis in the cognitive template (i.e., the task cognitive template) is referred to as analysis-class chain information, and the template chain information used for result verification in the cognitive template (i.e., the task cognitive template) is referred to as verification-class chain information. Specifically, for example, in a business scenario where the task description text is to formulate a four-day and three-night travel route from location A to location B with a budget of 2000, the embodiment of the present application generates a reasoning path for generating a travel route through multiple template chain information in the cognitive template (i.e., the task cognitive template). Further, the embodiment of the present application maps location A to location B, a budget of 2000, and four days and three nights (i.e., text key content) to analysis-class chain information (e.g., travel destination, travel time, and budget range), obtains task mapping content information, and then performs reasoning analysis on the task mapping content information to obtain content reasoning results. Further, the embodiment of the present application performs result verification on the content reasoning results according to verification-class chain information (e.g., time verification, path optimization, and funding control) to obtain verification results.

[0138] IV. Confidence. The confidence levels involved in the embodiments of this application include template text confidence, content confidence, unit confidence, text confidence, sample confidence, and sample text confidence. Confidence level is a probability indicator used in statistics to measure the reliability of parameter estimates or prediction results. It is usually expressed as 1-α (α is the significance level) and reflects the probability that the confidence interval contains the true value of the population parameter.

[0139] In the embodiment of the present application, the confidence of the cognitive template (i.e., the task cognitive template) is referred to as the template text confidence, and the confidence of the template analysis key content of the cognitive template (i.e., the task cognitive template) is referred to as the content confidence, and the confidence of each content text unit in the text key content is referred to as the unit confidence, and the confidence of the text key content is referred to as the text confidence, and the confidence of the sample text unit matching each content text unit is referred to as the sample confidence, and the confidence of the sample text key content is referred to as the sample text confidence.

[0140] Specifically, for example, in a business scenario where the task description text is to formulate a four-day and three-night travel route from location A to location B with a budget of 2,000, the embodiment of the present application determines the template text confidence of the first cognitive template with the greatest match between location A to location B, a budget of 2,000, and four days and three nights (i.e., the key content of the text) as the first template text confidence, and determines the template text confidence of the second cognitive template as the second template text confidence, and then obtains the minimum template text confidence from the first template text confidence and the second template text confidence, and then uses the cognitive template corresponding to the minimum template text confidence as the task cognitive template. Furthermore, the embodiment of the present application divides the unit of location A to location B, the budget of 2000 and four days and three nights (i.e., the key content of the text) into units based on the minimum text division unit, and obtains each content text unit of the key content of the text, and then obtains the sample confidence of the sample text unit that matches each content text unit, and then uses the sample confidence of each sample text unit as the unit confidence of each content text unit, and then determines the text confidence of the task description text through the unit confidence, and then determines the task complexity of the task description text through the text confidence, and then optimizes the reasoning path through the task complexity. Similarly, the sample text confidence is used to determine the task complexity of the sample task description text, and then optimizes the sample reasoning path through the task complexity.

[0141] 5. Link nodes. The link nodes included in the embodiments of the present application include a first link node, a second link node, and a third link node. Link nodes refer to the nodes that make up a link path. A link is a physical channel between nodes, and a node is the endpoint of this channel.

[0142] In the embodiment of the present application, the link node corresponding to the first analysis class chain information is referred to as the first link node, and the link node corresponding to the second analysis class chain information is referred to as the second link node, and the link node corresponding to the verification class chain information is referred to as the third link node. Specifically, for example, in a business scenario where the task description text is to formulate a four-day and three-night travel route from location A to location B with a budget of 2000, the embodiment of the present application uses the first node key content of the first link node as the first node key content of the first link node, and then determines the position of the content to be supplemented from the first node key content. Further, the embodiment of the present application uses the missing content at the position of the content to be supplemented as a pending content, and uses the content other than the first pending content in the first node key content as the second pending content, and then when mapping location A to location B, a budget of 2000 and four days and three nights (i.e., text key content) to the first analysis class chain information (e.g., travel destination, travel time and budget range), the first pending content at the position of the content to be supplemented is updated to the key text content, and then the task mapping content information for the first analysis class chain information is obtained through the text key content and the second pending content. Furthermore, in an embodiment of the present application, the information content in the second analysis chain information (for example, preference types and special needs) is used as the second node key content of the second link node, and then the task mapping content information is reasoned and analyzed through the task analysis strategy indicated by the second node key content to obtain a travel route planning (i.e., a content reasoning result). Furthermore, in an embodiment of the present application, the information content in the verification chain information is used as the third node key content of the third link node, and then the analysis strategy indicated by the third node key content is determined as a verification analysis strategy. Furthermore, if the verification result indicates that there is an optimizable reasoning result in the travel route planning (i.e., the content reasoning result), the optimizable reasoning result in the travel route planning (i.e., the content reasoning result) is optimized according to the verification analysis strategy to obtain an optimized content reasoning result.

[0143] See Figure 1 , Figure 1 This is a schematic diagram of a network architecture provided by an embodiment of the present application. Figure 1 As shown, the network architecture may include a server 10d and a user terminal cluster, and the user terminal cluster may include one or more user terminals, and the number of user terminals is not limited here. Figure 1As shown, the user terminal cluster may specifically include user terminal 10a, user terminal 10b and user terminal 10c, etc. Among them, server 10d may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, as well as basic cloud computing services such as big data and artificial intelligence platforms. User terminal 10a, user terminal 10b and user terminal 10c, etc. may include: smart phones, tablet computers, laptop computers, PDAs, mobile internet devices (mobile internet devices, MIDs), wearable devices (such as smart watches, smart bracelets, etc.), smart voice interaction devices, smart home appliances (such as smart TVs, etc.) and vehicle-mounted devices and other electronic devices with video / image playback functions. As Figure 1 As shown, the user terminal 10a, the user terminal 10b and the user terminal 10c can respectively establish a network connection with the server 10d, so that each user terminal can exchange data with the server 10d through the network connection.

[0144] like Figure 1 As shown, the user terminal 10a, the user terminal 10b and the user terminal 10c can be connected to the server 10d respectively through the network, so that each user terminal can exchange data with the server 10d through the network connection. For example, in the scenario of travel route planning, the target task reasoning model is deployed on the business server (for example, the above Figure 1 For example, the server 10d in the corresponding embodiment is used as an example. An input display page associated with the reasoning task is displayed on the business client (e.g., the user terminal 10a). When the business object (e.g., the user U1) confirms the query data information input on the input display page, the business client (e.g., the user terminal 10a) sends the query data information to the business server (e.g., the above-mentioned Figure 1 The server 10d in the corresponding embodiment).

[0145] Furthermore, the service server (for example, Figure 1After receiving the query data information, the server 10d) in the corresponding embodiment invokes the target task reasoning model to perform information identification on the query data information and obtain the data information type of the query data information. It should be understood that if the data information type includes a text data type and a non-text data type, the query data information obtains a first query data content corresponding to the non-text data type and a second query data content corresponding to the text data type, and then performs content conversion processing on the first query data content to obtain a first query conversion content for the first query data content. Furthermore, the target task reasoning model determines the task description text associated with the reasoning task based on the first query conversion content and the second query data content.

[0146] It is understood that if the data information type includes a text data type, the query data information is determined as the task description text. It is also understood that if the data information type includes a non-text data type, such as video data, voice data, image data, etc., which are not limited here, the query data information is subjected to content conversion processing to obtain a second query conversion content for the query data information, and the second query conversion content is used as the task description text.

[0147] Furthermore, the target task reasoning model determines the task template corresponding to the reasoning task based on the keywords of the task description text (i.e., the key content of the text), and then determines the reasoning path associated with the reasoning task based on the multiple template chain information in the task template. It should be understood that the template chain information at least includes analysis-class chain information for performing reasoning analysis and verification-class chain information for performing reasoning verification. Furthermore, the target task cognitive template maps the key content of the text to the analysis-class chain information according to the reasoning path to obtain the task mapping content information, and then performs reasoning analysis on the task mapping content information to obtain the travel route planning (i.e., the content reasoning result). Furthermore, the target task reasoning model verifies the travel route planning (i.e., the content reasoning result) through the verification-class chain information to obtain the verification result, and then determines the task reasoning result for the reasoning task through the content reasoning result and the verification result.

[0148] Furthermore, the service server (for example, Figure 1The server 10d in the corresponding embodiment sends the task reasoning result to the business client (e.g., user terminal 10a), so that the business client (e.g., user terminal 10a) outputs the task reasoning result to the result display page, and then displays the content reasoning result in the first display area of the result display page, and displays the verification result in the verification result display area of the second display area of the result display page. Furthermore, the business object (e.g., user U1) provides feedback on the task reasoning result, so that the business client (e.g., user terminal 10a) displays feedback prompt information for providing result feedback on the result display page. It should be understood that the feedback prompt information includes first-class feedback prompt information and second-class feedback prompt information. It should be understood that when the feedback prompt information is first-class feedback prompt information, the content reasoning result is directly determined as the task execution result for the reasoning task. When the feedback prompt information is second-class feedback prompt information, the task reasoning result is rewritten based on the second-class feedback prompt information to obtain the optimized task reasoning result, and then the optimized task reasoning result is determined as the task execution result for the reasoning task. Furthermore, the service client (eg, the user terminal 10a) outputs the task execution result to the result display page, and then displays the task execution result in the third display area of the result display page.

[0149] For further understanding, please refer to Figure 2 , Figure 2 This is a schematic diagram of an interactive scenario of a content reasoning method provided in an embodiment of the present application. Figure 2 The user terminal 20a shown in FIG. 2 may be the above-mentioned Figure 1 The user terminal 10a corresponds to the user terminal 10a in the figure. Here, the user terminal 10a and the user terminal 20a are service clients, such as mobile phones, tablets, etc., and there is no excessive restriction here. It should be understood that Figure 2 The server 20d shown in FIG can be the above Figure 1 Corresponding to the server 10d in FIG, the server 10d and the server 20d here are business servers.

[0150] In the embodiment of the present application, the target task reasoning model is deployed on the business server (for example, the above Figure 1 Taking the server 10d in the corresponding embodiment as an example, the business client (e.g., the user terminal 10a) displays the input display page associated with the reasoning task. Further, the business object (e.g., the user U1) enters the query data information 4a on the input display page, and then clicks the confirmation control on the input display page to trigger the query data information 4a, and sends the query data information 4a to the business server (e.g., the above Figure 1 The server 10d in the corresponding embodiment).

[0151] Furthermore, the service server (for example, Figure 1The server 10d) in the corresponding embodiment receives query data information 4a and then invokes the target task reasoning model to determine the task description text 2a associated with the reasoning task using the query data information 4a. Furthermore, the target task reasoning model performs text extraction on the task description text 2a to obtain text key content 2b. Based on the text key content 2b, the target task reasoning model determines the task cognitive template 2c corresponding to the reasoning task. It should be understood that the task cognitive template 2c contains multiple template chain information, including at least analysis chain information 2d for performing reasoning analysis and verification chain information 2e for performing reasoning verification. Furthermore, the target task reasoning model determines the reasoning path 2f associated with the reasoning task based on the multiple template chain information. When executing the reasoning task according to the reasoning path 2f, the target task reasoning model maps the text key content 2b to the analysis chain information 2d, obtaining task mapping content information. Furthermore, the target task reasoning model performs reasoning analysis on the task mapping content information to obtain a content reasoning result 2g. The target task reasoning model then verifies the content reasoning result 2g using the verification chain information 2e, obtaining a verification result 2h. It can be understood that the target task reasoning model determines the task reasoning result 2j for the reasoning task through the content reasoning result 2g and the verification result 2h, so that the business server (for example, the above Figure 1 The server 10d) in the corresponding embodiment returns the task reasoning result 2j to the service client (eg, the user terminal 10a).

[0152] For easier understanding, see Figure 3 , Figure 3 This is a flow chart of a content inference method provided by an embodiment of the present application. It is understandable that the method can be executed by a computer device, where the computer device can be the above-mentioned business terminal or the business server, where the business terminal can be the above-mentioned Figure 1 The corresponding user terminal 10a, the business service here can be the above Figure 1 The corresponding server 10d. Figure 3 As shown, the method may at least include the following steps S101 to S104:

[0153] It should be noted that the target task reasoning model can also be an object deployed on the business server. For ease of understanding, the embodiment of the present application is described with the target task reasoning model as an object integrated on the business client.

[0154] Step S101: Obtain task description text associated with the reasoning task.

[0155] Taking the task description text of developing a three-day cultural and historical tourism plan for Location A with a budget of approximately 2,000 and requiring moderate physical effort as an example, the business client displays an input display page associated with the reasoning task. In response to a confirmation operation on the query data information entered in the input display page, the query data information is transmitted to the target task reasoning model. Furthermore, the target task reasoning model identifies the query data information and obtains the data information type of the query data information. It should be understood that if the data information type includes a text data type and a non-text data type, the first query data content corresponding to the non-text data type is obtained from the query data information, and the second query data content corresponding to the text data type is obtained. The first query data content is then content-converted to obtain a first query conversion content for the first query data content. Furthermore, the target task reasoning model determines the task description text associated with the reasoning task based on the first query conversion content and the second query data content.

[0156] For easier understanding, see Figure 4 , Figure 4 This is a schematic diagram of an input display page provided by an embodiment of the present application. Figure 4 As shown, a business object (e.g., user U1) can enter query data information 4a in the input display page. It should be understood that the data type of the query data information can be a text data type or a non-text data type, such as video data, voice data, image data, etc., which is not limited here. Furthermore, the business object (e.g., user U1) triggers the confirmation control in the input display page to transmit the query data information 4a to the target task inference model.

[0157] For further understanding, please see Figure 5 , Figure 5 This is a schematic diagram of a scenario for determining a task description text based on query data information of non-text data types and text data types provided by an embodiment of the present application. Figure 5 As shown, the target task reasoning model obtains the first query data content corresponding to the non-text data type and the second query data content corresponding to the text data type from the query data information 4a, and then performs content conversion processing on the first query data content to obtain the first query conversion content for the first query data content. Furthermore, the task description text 2a associated with the reasoning task is determined based on the first query conversion content and the second query data content.

[0158] It should be understood that if the data information type includes a text data type, the query data information is determined as the task description text. Similarly, if the data information type includes a non-text data type, the query data information is subjected to content conversion processing to obtain a second query conversion content for the query data information, and the second query conversion content is used as the task description text.

[0159] Optionally, the target task reasoning model outputs the task reasoning results to a result display page of the business client. The result display page includes a first display area and a second display area. It should be understood that the first display area is used to display the content reasoning results, and the second display area includes a verification result display area, which is used to display the verified results.

[0160] For easier understanding, see Figure 6 , Figure 6 This is a diagram showing a task reasoning result provided by an embodiment of the present application. Figure 6 As shown, the result display page 6a includes a first display area and a second display area. The first display area is used to display the three-day cultural and historical tour plan of location A (i.e., the content inference result), wherein the specific content is as follows:

[0161] Day 1: Theme 1

[0162] morning

[0163] 08:30-11:00 Location B (free, recommended 2 hours)

[0164] → Walk 10 minutes to location C

[0165] 11:10-14:00 Location C (Ticket 60 yuan, including lunch time, recommended 2.5 hours)

[0166] afternoon

[0167] 14:30-16:00 Location D (Ticket 2 yuan, overlooking the panoramic view of Location C, recommended 1 hour)

[0168] → 20 minutes by bus to location E

[0169] 16:30-18:30 Location E (Free, Hutong Culture + Snacks)

[0170] night

[0171] 19:00-20:30 Dinner at Location F (estimated 80 yuan / person)

[0172] Backup option: If the reservation for location C fails, replace it with "Museum"

[0173] Day 2: Theme 2

[0174] morning

[0175] 08:00-11:30 Location G (Ticket 30 yuan, recommended 3 hours)

[0176] afternoon

[0177] 12:30-15:00 Location H (Ticket 25 yuan, including lunch time, recommended 2 hours)

[0178] →40 minutes by subway to Location I

[0179] 15:40-17:30 Location J (Free, Exterior Location I / Location K)

[0180] Flexible time: If you have enough energy, you can add a "University L" tour

[0181] Day 3: Theme 3

[0182] morning

[0183] 09:00-11:00 Location O (Ticket 15 yuan, recommended 2 hours)

[0184] afternoon

[0185] 12:00-14:30 Location P (free, pedicab hutong tour + lunch, estimated 100 yuan)

[0186] 15:00-17:00 Location Q (Ticket 25 yuan, historical building)

[0187] Furthermore, the second display area is used to display the optimization results (i.e., verification results) of the three-day cultural and historical tour plan for location A, and the specific contents are as follows:

[0188] 1. Time verification:

[0189] a. Day 1: Location C → Jingshan → Location E: The route is smooth, and walking / bus time accounts for 18% (reasonable).

[0190] b. Problem solution: On Day 2, you need to transfer to the subway from Location H to Location I. It is recommended to use a taxi app (about 35 yuan).

[0191] 2. Path optimization:

[0192] a. The route from Location O to Location P on Day 3 is relatively long (6 km apart). It is recommended to adjust it to:

[0193] Optimization plan: Location O → (38 minutes by subway) → Location Q → (15 minutes on foot) → Location P.

[0194] 3. Consumption control:

[0195] a. Estimated total cost for three days: 157 yuan for tickets + 120 yuan for transportation + 450 yuan for food and beverages = 727 yuan per person (much lower than expected).

[0196] b. Consumer Tips: The snacks at Location E are heavily overpriced. It is recommended to choose the "Snacks at Location S" branch.

[0197] Furthermore, optionally, an embodiment of the present application displays feedback prompt information for providing result feedback on the result display page of the business client. It should be understood that the feedback prompt information includes the first type of feedback information. Furthermore, the business client, in response to the feedback trigger operation for the first type of feedback information, determines the task reasoning result as the task execution result for the reasoning task, and then outputs the task execution result to the result display page of the business client. It should be understood that the result display page includes a third display area, and the third display area is used to display the task execution result.

[0198] For easier understanding, see Figure 7 , Figure 7 This is a diagram of a task execution result provided by an embodiment of the present application. Figure 7 As shown, the result display page 6a includes a third display area, wherein, when the business object (for example, user U1) triggers the first type of feedback information in the feedback prompt information, the target task reasoning model directly determines the task reasoning result as the task execution result of the reasoning task, and then outputs the task execution result to the result display page, so that the business client displays the task execution result in the third display area of the result display page.

[0199] The specific contents of the task execution results are as follows:

[0200] Day 1: Theme 1

[0201] morning

[0202] 08:30-11:00 Location B (free, recommended 2 hours)

[0203] → Walk 10 minutes to location C

[0204] 11:10-14:00 Location C (Ticket 60 yuan, including lunch time, recommended 2.5 hours)

[0205] afternoon

[0206] 14:30-16:00 Location D (Ticket 2 yuan, overlooking the panoramic view of Location C, recommended 1 hour)

[0207] → 20 minutes by bus to location E

[0208] 16:30-18:30 Location E (Free, Hutong Culture + Snacks)

[0209] night

[0210] 19:00-20:30 Dinner at Location F (estimated 80 yuan / person)

[0211] Backup option: If the reservation for location C fails, replace it with "Museum"

[0212] Day 2: Theme 2

[0213] morning

[0214] 08:00-11:30 Location G (Ticket 30 yuan, recommended 3 hours)

[0215] afternoon

[0216] 12:30-15:00 Location H (Ticket 25 yuan, including lunch time, recommended 2 hours)

[0217] →Take a taxi to location I

[0218] 15:40-17:30 Location J (Free, Exterior Location I / Location K)

[0219] Flexible time: If you have enough energy, you can add a "University L" tour

[0220] Day 3: Theme 3

[0221] morning

[0222] 09:00-11:00 Location O (Ticket 15 yuan, recommended 2 hours)

[0223] afternoon

[0224] 12:00-14:30 Location Q (Ticket 25 yuan, historical building)

[0225] 15:00-17:00 Location P (free, pedicab hutong tour + lunch, estimated 100 yuan)

[0226] It should be understood that the feedback prompt information includes the second type of feedback information. In response to the feedback trigger operation for the second type of feedback information, the business client rewrites the task reasoning result based on the second type of feedback information to obtain an optimized task reasoning result, and then determines the optimized task reasoning result as the task execution result for the reasoning task. Furthermore, the target task reasoning model outputs the task execution result to the result display page of the business client, and then displays the task execution result in the third display area of the result display page.

[0227] For easier understanding, see Figure 8 , Figure 8 This is another task execution result display diagram provided by the embodiment of the present application. Figure 8As shown, the result display page 6a includes a second display area and a third display area. When a business object (e.g., user U1) triggers the second type of feedback information in the feedback prompt information, the target task reasoning model rewrites the task reasoning result according to the second type of feedback information to obtain an optimized task reasoning result, which is then determined as the task execution result for the reasoning task. Furthermore, the target task reasoning model outputs the task execution result to the result display page, causing the business client to display the task execution result in the third display area of the result display page.

[0228] The specific contents of the task execution results are as follows:

[0229] Day 1: Theme 1

[0230] morning

[0231] 08:30-11:00 Location B (free, recommended 2 hours)

[0232] → Walk 10 minutes to location C

[0233] 11:10-14:00 Location C (Ticket 60 yuan, including lunch time, recommended 2.5 hours)

[0234] afternoon

[0235] 14:30-16:00 Location D (Ticket 2 yuan, overlooking the panoramic view of Location C, recommended 1 hour)

[0236] → 20 minutes by bus to location E

[0237] 16:30-18:30 Location E (Free, Hutong Culture + Snacks)

[0238] night

[0239] 19:00-20:30 Dinner at Location F (estimated 80 yuan / person)

[0240] Backup option: If the reservation for location C fails, replace it with "Museum"

[0241] Day 2: Theme 2

[0242] morning

[0243] 08:00-11:30 Location G (Ticket 30 yuan, recommended 3 hours)

[0244] afternoon

[0245] 12:30-15:00 Location H (Ticket 25 yuan, including lunch time, recommended 2 hours)

[0246] →Take a taxi to location I

[0247] 15:40-17:30 Location J (Free, Exterior Location I / Location K)

[0248] Day 3: Theme 3

[0249] morning

[0250] 09:00-11:00 Location O (Ticket 15 yuan, recommended 2 hours)

[0251] →Take a taxi to location Q

[0252] afternoon

[0253] 12:00-14:30 Location Q (Ticket 25 yuan, historical building)

[0254] 15:00-17:00 Location P (free, pedicab hutong tour + lunch, estimated 100 yuan)

[0255] Step S102: Based on the key text content in the task description text information, determine a task cognitive template adapted to the reasoning task, and determine a reasoning path associated with the reasoning task based on multiple reasoning chain information in the task cognitive template.

[0256] The multiple template chain information includes at least analysis chain information for performing reasoning analysis and verification chain information for performing reasoning verification.

[0257] The target task reasoning model performs word segmentation extraction on the task description text to obtain multiple word segmentation texts for the task description text, and then selects the target word segmentation text that matches the reasoning task from the multiple word segmentation texts. Furthermore, the target task reasoning model uses the target word segmentation text as the text key content and then performs content analysis on the text key content to obtain the content association relationship between the various word segmentation texts in the text key content. Furthermore, based on the text key content and content association relationship, the target task reasoning model determines a task cognitive template suitable for the reasoning task from multiple cognitive templates associated with the reasoning task.

[0258] For easier understanding, see Figure 9 , Figure 9 This is a schematic diagram of a scenario for determining the content association relationship of key text contents provided by the embodiment of the present application. Figure 9 As shown, multiple segmented texts are extracted from the task description text 2a, wherein the multiple segmented texts include segmented text 1, segmented text 2, segmented text 3 and segmented text 4. Further, the target segmented text that matches the reasoning task is screened from the multiple segmented texts, and then the target segmented text is used as the text key content 2b. Figure 9As shown, the target segmented text includes target segmented text 1, target segmented text 2, and target segmented text 3. It should be understood that segmented text 1 corresponds to target segmented text 1, segmented text 3 corresponds to target segmented text 2, and segmented text 4 corresponds to target segmented text 3. Furthermore, the target task reasoning model performs content analysis on the text key content 2b to obtain content association relationships between the various segmented texts in the text key content 2b (i.e., target segmented text 1, target segmented text 2, and target segmented text 3).

[0259] Among them, the target task reasoning model determines the task type of the reasoning task based on the text key content and content association relationship. Furthermore, the target task reasoning model obtains multiple cognitive templates that match the task type of the reasoning task, obtains the template analysis key content in each cognitive template, and then content matches the text key content with the template analysis key content in each cognitive template to obtain the content matching degree associated with each cognitive template. It should be understood that a content matching degree is used to characterize the degree of content matching between the template analysis key content and the text key content in a cognitive template. Furthermore, the target task cognitive template obtains the maximum content matching degree from the content matching degree, and determines the task cognitive template corresponding to the reasoning task through the cognitive template corresponding to the maximum content matching degree.

[0260] The cognitive template corresponding to the maximum content matching degree includes a first cognitive template and a second cognitive template.

[0261] Among them, the target task reasoning model determines the template text confidence of the first cognitive template as the first template text confidence, and determines the template text confidence of the second cognitive template as the second template text confidence. It should be understood that the first template text confidence is determined by the content confidence of the key content of the first template analysis in the first cognitive template, and the second template text confidence is determined by the content confidence of the key content of the second template analysis in the second cognitive template. Furthermore, the target task reasoning model obtains the minimum template text confidence from the first template text confidence and the second template text confidence, and uses the cognitive template corresponding to the minimum template text confidence as the task cognitive template.

[0262] For easier understanding, see Figure 10 , Figure 10 This is a schematic diagram of a cognitive template (i.e., a task cognitive template) provided in an embodiment of the present application. Figure 10As shown, the cognitive template (i.e., task cognitive template 2c) includes four template chain information: clear requirements, route planning, rationality check, and user feedback and iteration. When the target task reasoning model obtains the task cognitive template 2c that matches the reasoning task, it generates a reasoning path for the reasoning task through the multiple template chain information in the task cognitive template 2c.

[0263] For easier understanding, see Figure 11 , Figure 11 This is a schematic diagram of determining a cognitive template (i.e., a task cognitive template) provided in an embodiment of the present application. Figure 11 As shown, multiple cognitive templates that match the reasoning task are obtained, wherein the multiple cognitive templates include cognitive template 1, cognitive template 2, and cognitive template 3. Further, the text key content 2b is content-matched with the template analysis key content of each cognitive template to obtain the content matching degree associated with each cognitive template. It should be understood that the template analysis key content includes template analysis key content 1 corresponding to cognitive template 1, template analysis key content 2 corresponding to cognitive template 2, and template analysis key content 3 corresponding to cognitive template 3, and the content matching degree includes content matching degree 1 between cognitive template 1 and text key content 2b, content matching degree 2 between cognitive template 2 and text key content 2b, and content matching degree 3 between cognitive template 3 and text key content 2b. Further, the target task reasoning model obtains the cognitive template corresponding to the maximum content matching degree among the content matching degrees. Among them, the cognitive template corresponding to the maximum content matching degree includes the first cognitive template and the second cognitive template. It should be understood that the target task reasoning model determines the template text confidence of the first cognitive template as the first template text confidence, and determines the template text confidence of the second cognitive template as the second template text confidence, and then obtains the minimum template text confidence from the first template text confidence and the second template text confidence, and uses the cognitive template corresponding to the minimum template text confidence as the task cognitive template.

[0264] It should be noted that, optionally, when multiple cognitive templates matching the task type of the reasoning task are not obtained, a general cognitive template is obtained, and the general cognitive template is determined as the task cognitive template adapted to the reasoning task.

[0265] Furthermore, the target task reasoning model obtains multiple template chain information from the task cognition template and obtains the information encoding identifier of each template chain information. Then, according to the information encoding identifier of each template chain information, it generates a link node corresponding to each template chain information. Furthermore, the target task reasoning model generates a reasoning path associated with the reasoning task through the link node corresponding to each template chain information.

[0266] For easier understanding, see Figure 12 , Figure 12This is a schematic diagram of generating a reasoning path provided by an embodiment of the present application. Figure 12 As shown, multiple template chain information is obtained from the task cognitive template 2c, and the multiple template chain information is information coding identifiers, wherein the multiple template chain information includes template chain information 1, template chain information 2, and template chain information 3, and the information coding identifiers include information coding identifier 1 corresponding to template chain information 1, information coding identifier 2 corresponding to template chain information 2, and information coding identifier 3 corresponding to template chain information 3. Furthermore, the target task reasoning model generates a link node corresponding to each template chain information according to the information coding identifier of each template chain information, wherein the link nodes include link node 1 corresponding to template chain information 1, link node 2 corresponding to template chain information 2, and link node 3 corresponding to template chain information 3. Furthermore, the target task reasoning model generates a reasoning path 2f associated with the reasoning task through the link nodes corresponding to each template chain information.

[0267] Among them, the target task reasoning model obtains each content text unit in the key content of the text, determines the unit confidence of each content text unit, and then determines the text confidence of the task description text through the unit confidence of each content text unit. It should be understood that the text confidence of the task description text can be determined by averaging or weighted calculation of the unit confidence, which is not limited here. Further, the target task reasoning model determines the task complexity of the task description text through text complexity. It should be understood that the value of task complexity is used to characterize the query difficulty when performing text query on the task description text. Further, when the task complexity reaches the complex optimization condition, the target task reasoning model optimizes the reasoning path to obtain the reasoning path after path optimization, and then when the reasoning path is updated through the reasoning path after path optimization, the reasoning task is executed according to the updated reasoning path.

[0268] The target task reasoning model obtains the minimum text division unit for unit division, and then divides the key text content into units based on the minimum text division unit, obtaining each content text unit in the key text content. Furthermore, the target task reasoning model obtains the sample confidence of the sample text unit that matches each content text unit, and uses the sample confidence of each sample text unit as the unit confidence of each content text unit.

[0269] For easier understanding, see Figure 13 , Figure 13 This is a schematic diagram of determining unit confidence provided by an embodiment of the present application. Figure 13As shown, the target task reasoning model divides the text key content 2b (for example, four days and three nights, 2000 yuan and location A) into units by the minimum text division unit, and obtains each content text unit of the text key content 2b (for example, four days, three nights, 2000 yuan and location A). Further, the target task reasoning model obtains the sample confidence of the sample text unit that matches each content text unit. It should be understood that the obtained sample confidence includes sample confidence 1 matching four days, sample confidence 2 matching three nights, sample confidence 3 matching 2000 yuan and sample confidence 4 matching location A, and then uses sample confidence 1, sample confidence 2, sample confidence 3 and sample confidence 4 as the unit confidence of each content text unit, wherein the unit confidence includes unit confidence 1 corresponding to sample confidence 1, unit confidence 2 corresponding to sample confidence 2, unit confidence 3 corresponding to sample confidence 3 and unit confidence 4 corresponding to sample confidence 4.

[0270] The reasoning path includes multiple link nodes; the complex optimization condition includes a first confidence threshold for adding nodes and a second confidence threshold for deleting nodes; and the first confidence threshold is smaller than the second confidence threshold.

[0271] It should be noted that if the value of the task complexity is less than or equal to the first confidence threshold, the template addition chain information for link optimization is obtained, and then the link node corresponding to the template addition chain information is added to the multiple link nodes included in the reasoning path to obtain the first reasoning path. Similarly, if the value of the task complexity is greater than or equal to the second confidence threshold, the template deletion chain information for link optimization is obtained from the multiple template chain information, and then the link node corresponding to the template deletion chain information is removed from the multiple link nodes included in the reasoning path to obtain the second reasoning path. Furthermore, the target task reasoning model uses the first reasoning path or the second reasoning path as the reasoning path after path optimization, and then updates the reasoning path through the reasoning path after path optimization.

[0272] Step S103: When executing the reasoning task according to the reasoning path, the text key content is mapped to the analysis class chain information, and the analysis class chain information mapped with the text key content is reasoned and analyzed to obtain the content reasoning result, and the content reasoning result is verified by the verification class chain information to obtain the verification result.

[0273] Among them, the analysis class chain information includes first analysis class chain information and second analysis class chain information; the reasoning path includes at least a first link node corresponding to the first analysis class chain information and a second link node corresponding to the second analysis class chain information.

[0274] Among them, when the target task reasoning model performs the reasoning task according to the reasoning path, it uses the information content in the first analysis class chain information as the first node key content of the first link node, and then determines the position of the content to be supplemented from the first node key content of the first link node. Furthermore, the target task reasoning model uses the missing content at the position of the content to be supplemented as the first content to be processed, and uses the content of the first node key content other than the first content to be processed as the second content to be processed. Furthermore, when the target task reasoning model maps the text key content to the first analysis class chain information, it updates the first content to be processed at the position of the content to be supplemented to the key text content, and then obtains the task mapping content information for the first analysis class chain information through the key text content and the second content to be processed.

[0275] It should be noted that the second content to be processed is replaced by a preset reasonable hypothesis, and then the task mapping content information for the first analysis type chain information is obtained through the replaced reasonable hypothesis and the text key content.

[0276] For easier understanding, see Figure 14 , Figure 14 This is a scene diagram of a task mapping content information generation process provided by an embodiment of the present application. Figure 14 As shown, the target task reasoning model takes the information content of the first analysis class chain information as the first node key content of the first link node, and then determines the position of the content to be supplemented from the first node key content of the first link node. Furthermore, the target task reasoning model takes the missing content at the position of the content to be supplemented as the first content to be processed, and takes the content of the first node key content other than the first content to be processed as the second content to be processed. Furthermore, when the target task reasoning model maps the text key content 2b to the first analysis class chain information, the first content to be processed at the position of the content to be supplemented is updated to the text key content 2b, and then the task mapping content information for the first analysis class chain information is obtained through the text key content 2b and the second content to be processed.

[0277] Furthermore, the target task reasoning model uses the information content in the second analysis class chain information as the second node key content of the second link node, and performs reasoning analysis on the task mapping content information through the task analysis strategy indicated by the second node key content to obtain the content reasoning result.

[0278] For easier understanding, see Figure 15 , Figure 15 This is a scene diagram of the content inference result generation process provided by the embodiment of this application. Figure 15As shown, the target task reasoning model takes the information content in the second analysis class chain information as the second node key content of the second link node, and then performs reasoning analysis on the task mapping content information through the task analysis strategy indicated in the second node key content to obtain the content reasoning result 2g.

[0279] It should be understood that the reasoning path also includes a third link node corresponding to the verification-type chain information.

[0280] Step S104: Determine the task reasoning result for the reasoning task through the content reasoning result and the verification result.

[0281] Furthermore, the target task reasoning model uses the information content in the verification chain information as the third node key content of the third link node, and then determines the analysis strategy indicated by the third node key content as the verification analysis strategy. It is understood that if the verification result indicates that there are optimizable reasoning results in the content reasoning results, then the optimizable reasoning results in the content reasoning results are optimized according to the verification analysis strategy to obtain an optimized content reasoning result, and then determine the optimized content reasoning result as the task reasoning result for the reasoning task.

[0282] For easier understanding, see Figure 16 , Figure 16 This is a schematic diagram of a task reasoning result generation process provided by an embodiment of the present application. Figure 16 As shown, the target task reasoning model uses the information content in the verification chain information 2e as the third node key content of the third link node. It then optimizes the optimizable reasoning results in the content reasoning result 2g using the verification analysis strategy indicated in the third node key content, resulting in an optimized content reasoning result. Furthermore, the target task reasoning result determines the optimized content reasoning result as the task reasoning result for the reasoning task.

[0283] Optionally, the target task reasoning model generates a content comparison relationship table for comparing and optimizing the content reasoning results. Furthermore, the target task reasoning model outputs the content comparison relationship table to a result display page of the business client, causing the business client to display the content comparison relationship table in a comparison relationship table display area in a second display area of the result display page.

[0284] For details on the content comparison relationship table, please see the following table:

[0285] Table 1 Content comparison relationship table

[0286]

[0287] The embodiment of the present application can obtain the task description text information associated with the reasoning task, and then determine the cognitive template (i.e., task cognitive template) corresponding to the reasoning task based on the keywords (i.e., text key content) in the task text description information. Further, the embodiment of the present application determines the reasoning path associated with the reasoning task based on the multiple reasoning chain information in the cognitive template (i.e., task cognitive template). It should be understood that the multiple reasoning chain information at least includes analysis class chain information for performing reasoning analysis and verification class chain information for performing reasoning verification. Further, the embodiment of the present application maps the keywords (i.e., text key content) of the task description text information to the analysis class chain information according to the reasoning path, and then performs reasoning analysis on the analysis class chain information mapped with the text key content to obtain content reasoning results. Furthermore, the embodiment of the present application verifies the content reasoning result by verifying the class chain information, obtains the verification result for the content reasoning result, and then determines the task reasoning result for the reasoning task through the content reasoning result and the verification result. It can be understood that the embodiment of the present application flexibly determines the task cognition template suitable for the current reasoning task based on the key content of the text in the task description text, and then activates the task reasoning ability through multiple template chain information in the task cognition template. When the reasoning path for processing the reasoning task is specifically determined through the task cognition template, the execution of the task can be reasoned step by step according to the reasoning path. In this way, not only can the reasoning and analysis ability of the target task reasoning model be improved from the root, but also the accuracy of the target task reasoning model result output can be improved through result verification.

[0288] For easier understanding, see Figure 17 , Figure 17 This is a flow chart of a content inference method provided by an embodiment of the present application. It is understandable that the method can be executed by a computer device, where the computer device can be the above-mentioned business terminal or the business server, where the business terminal can be the above-mentioned Figure 1 The corresponding user terminal 10a, the business service here can be the above Figure 1 The corresponding server 10d. Figure 17 As shown, the method may at least include the following steps S201 to S208:

[0289] It should be noted that the target task reasoning model can also be an object deployed on the business server. For ease of understanding, the embodiment of the present application is described with the target task reasoning model as an object integrated on the business client.

[0290] Step S201: Acquire training sample information associated with the target sample inference task.

[0291] Among them, the training sample information associated with the target sample reasoning task includes the sample task description text used to train the initial task reasoning model and the reference task reasoning result for the sample task description text; the sample reasoning task corresponding to the sample task description text is the target sample reasoning task.

[0292] Among them, the initial task reasoning model obtains N to-be-screened task description texts associated with multiple sample reasoning tasks, and then performs vector representation processing on the N to-be-screened task description texts to obtain N sample text representation vectors. It should be understood that one sample text representation vector is used to represent one to-be-screened task description text, and N is a positive integer. Further, the initial task reasoning model calculates the vector similarity between any two sample text representation vectors in the N sample text representation vectors to obtain M vector similarities, and then selects at least one target vector similarity whose vector similarity is greater than or equal to the vector similarity threshold from the M vector similarities. Further, the initial task reasoning model determines the two to-be-screened task description texts associated with each screened target vector similarity as the target screening task description texts. It can be understood that M is an integer greater than or equal to N.

[0293] Furthermore, the initial task reasoning model obtains the sample text confidence of each target screening task description text, and then, among the sample text confidences of each target screening task description text, the target screening task description text corresponding to the minimum sample text confidence is used as the sample task description text, and the sample reasoning task corresponding to the sample task description text is determined as the target sample reasoning task, and the task reasoning result corresponding to the sample task description text is determined as the reference task reasoning result. It should be understood that a sample text confidence is used to characterize the task difficulty of the sample reasoning task corresponding to a target screening task description text. Furthermore, the initial task reasoning model constructs training sample information associated with the target sample reasoning task based on the sample task description text and the reference task reasoning result.

[0294] For easier understanding, see Figure 18 , Figure 18 This is a sample task description text determination diagram provided by the embodiment of this application. Figure 18As shown, the N task description texts to be screened obtained by the initial task reasoning model include task description text 1 to be screened, task description text 2 to be screened, and task description text 3 to be screened. It can be understood that at this time, N=3. Further, the initial task reasoning model obtains N sample text representation vectors of the N task description texts to be screened, wherein the N sample text representation vectors include sample text representation vector 1, sample text representation vector 2, and sample text representation vector 3, and then calculates the vector similarity between any two sample text representation vectors in the N sample text representation vectors to obtain M vector similarities. As shown Figure 18 As shown, the M vector similarities include vector similarity 1, vector similarity 2, and vector similarity 3, where M=3. Furthermore, the initial task reasoning model selects at least one target vector similarity from the M vector similarities whose vector similarity is greater than or equal to the vector similarity threshold. Furthermore, the initial task reasoning model determines the two task description texts to be screened associated with each screened target vector similarity as target screening task description texts. It should be understood that the target screening task description texts here are task description text 1 to be screened and task description text 3 to be screened. Furthermore, the sample text confidence of the target screening task description text is obtained. It should be understood that the sample text confidence here includes the sample text confidence 1 corresponding to the target screening task description text 1 (i.e., task description text 1 to be screened) and the sample text confidence 2 corresponding to the target screening task description text 2 (i.e., task description text 3 to be screened), and then the target screening task description text corresponding to the minimum sample text confidence between the sample text confidence 1 and the sample text confidence 2 is determined as the sample task description text.

[0295] Step S202: Based on the key content of the sample text in the sample task description text, determine a sample task cognitive template adapted to the target sample reasoning task, and determine a sample reasoning path based on multiple template chain information in the sample task cognitive template.

[0296] The multiple template chain information includes at least sample analysis chain information for performing reasoning analysis and sample verification chain information for performing reasoning verification.

[0297] Among them, the initial task reasoning model determines the sample task cognitive template adapted to the target sample reasoning task based on the key content of the sample text in the sample task description text, and then determines the sample reasoning path based on multiple template chain information in the sample task cognitive template.

[0298] It is understandable that the process of the initial task reasoning model generating a sample reasoning path based on the sample task description text is the same as the process of the target task reasoning model generating a reasoning path based on the task description text. The specific implementation method can be found in the above Figure 3The description of obtaining and generating the reasoning path in the corresponding embodiment will not be repeated here.

[0299] Step S203: When executing the target sample reasoning task according to the sample reasoning path, the key content of the sample text is mapped to the sample analysis class chain information, and the sample analysis class chain information mapped with the key content of the sample text is reasoned and analyzed to obtain the sample content reasoning result, and the sample content reasoning result is verified through the sample verification class chain information to obtain the sample verification result.

[0300] When executing the target sample reasoning task along the sample reasoning path, the initial task reasoning model maps the key content of the sample text to the sample analysis chain information, and then performs reasoning analysis on the sample analysis chain information mapped with the key content of the sample text to obtain the sample content reasoning result. Furthermore, the initial task reasoning model verifies the sample content reasoning result using the sample verification chain information to obtain the sample verification result.

[0301] It can be understood that the process of the initial task reasoning model generating sample content reasoning results and sample verification results through the sample reasoning path is the same as the process of the target task reasoning model generating content reasoning results and verification results through the reasoning path. The specific implementation method can be found in the above Figure 3 The description of obtaining and generating verification results in the corresponding embodiment will not be repeated here.

[0302] Step S204: Determine the sample task reasoning result for the target sample reasoning task through the sample content reasoning result and the sample verification result, train the initial task reasoning model through the sample task reasoning result and the reference task reasoning result, and obtain the target task reasoning model.

[0303] Among them, the initial task reasoning model determines the sample task reasoning result for the target sample reasoning task through the sample content reasoning result and the sample verification result, and then trains the initial task reasoning result through the sample task reasoning result and the reference task reasoning result to obtain the target task reasoning model. Among them, the process of the initial task reasoning model obtaining the task reasoning result through the sample content reasoning result and the sample verification result is the same as the process of the target task reasoning model obtaining the task reasoning result through the sample content reasoning result and the sample verification result. The specific implementation method can be found in the above Figure 3 The description of obtaining the inference result of the generation task in the corresponding embodiment will not be repeated here.

[0304] It should be understood that the embodiment of the present application calculates the loss function of the sample task reasoning result and the task reasoning result. When the loss function reaches convergence, it indicates that the training of the target task reasoning model is completed.

[0305] It is understandable that the embodiments of this application adopt the concept of meta-learning, which enables the initial task reasoning model to gradually learn the optimal reasoning framework from a small number of training tasks. Unlike traditional machine learning methods, meta-learning focuses on how to enable the model to summarize general reasoning methods from a small number of examples, rather than just optimizing for a specific task.

[0306] Step S205: Obtain the task description text associated with the reasoning task.

[0307] Optionally, the target task reasoning model, in response to a feedback operation regarding the verification result, displays feedback text content for providing result feedback on the result display page of the business client. It should be understood that the feedback text content includes first-category feedback content, and based on the first-category feedback content, the task reasoning result is determined as the task execution result for the reasoning task. Furthermore, the target task reasoning model outputs the task execution result to the result display page of the business client, so that the business client displays the task execution result in the third display area of the result display page.

[0308] It should be understood that the feedback text content includes the second type of feedback content, which in turn causes the target task reasoning model to rewrite the task reasoning result based on the second type of feedback content, thereby obtaining an optimized task reasoning result. Furthermore, the target task reasoning model determines the optimized task reasoning result as the task execution result for the reasoning task, and then outputs the task execution result to the result display page of the business client, causing the business client to display the task execution result in the third display area of the result display page.

[0309] For easier understanding, see Figure 19 , Figure 19 This is another task execution result display diagram provided by the embodiment of the present application. Figure 19 As shown, the second type of feedback is a desire to reduce walking, which causes the target task reasoning model to optimize the task reasoning result based on the second type of feedback, obtaining an optimized task reasoning result. Furthermore, the target task reasoning model determines the optimized task reasoning result as the task execution result for the reasoning task.

[0310] The specific implementation of steps S206 to S208 can be found in the above Figure 3 The description of the specific process of step S102 to step S104 in the corresponding embodiment will not be repeated here.

[0311] It should be understood that, in addition to being applicable to the scenario of a designated tourist route, the embodiment of the present application can also be applied to the scenario of a mathematical operation. The target task reasoning model receives a task description text associated with the mathematical reasoning operation, and then obtains a task cognition template adapted to the mathematical reasoning operation through the keywords (i.e., text key content) in the task description text, and then obtains multiple template chain information of the task cognition template to generate a reasoning path for the mathematical reasoning operation. It should be understood that the task cognition template here also includes at least analysis class chain information for reasoning analysis and verification class chain information for verification analysis. Further, when the target task reasoning model performs a mathematical reasoning operation through the reasoning path, it maps the text key content to the analysis class chain information, and then performs reasoning analysis on the analysis class chain information mapped with the text key information to obtain a content reasoning result, and then performs verification analysis on the content reasoning result through the verification class chain information to obtain a verification result. Further, the target task reasoning model obtains a task reasoning result for the mathematical reasoning operation through the obtained content reasoning result and verification result.

[0312] It can be understood that the application scenarios of the embodiments of the present application can be, but are not limited to, formulating tourist routes and mathematical reasoning operations. The specific implementation methods of other application scenarios can refer to the above-mentioned specific implementation steps for formulating tourist routes and mathematical reasoning operations, which will not be repeated here.

[0313] The embodiment of the present application obtains a small amount of training sample information to perform model training on the initial task reasoning model, and obtains a target task reasoning model for task reasoning. Among them, the training sample information is obtained by obtaining the sample text confidence of multiple task description texts to be screened whose vector similarity is greater than or equal to the vector similarity threshold, and taking the task description text to be screened with the lowest sample text confidence as the sample task description text, and then determining the reference task reasoning result through the sample task description text. It can be understood that the embodiment of the present application trains the initial task reasoning model through a small amount of training sample information, which stimulates the reasoning ability of the initial task reasoning model and improves the efficiency of model training.

[0314] Further, see Figure 20 , Figure 20 1 is a schematic diagram of the structure of a content inference device provided in an embodiment of the present application. The content inference device can be a computer program (including program code) running on a computer device. For example, the content inference device 1 can be an application software. The content inference device 1 can be used to execute the corresponding steps of the method provided in an embodiment of the present application. The content inference device 1 can include: a text acquisition module 11, a template determination module 12, a content analysis module 13, and a result verification module 14.

[0315] A text acquisition module 11 is used to acquire a task description text associated with the reasoning task;

[0316] A template determination module 12 is configured to determine a task cognition template corresponding to the reasoning task based on key textual content in the task description text, and to determine a reasoning path associated with the reasoning task based on multiple template chain information in the task cognition template; the multiple template chain information in the task cognition template at least includes analysis chain information for performing reasoning analysis and verification chain information for performing reasoning verification;

[0317] The content analysis module 13 is used to map the key content of the text to the analysis chain information according to the reasoning path, perform reasoning analysis on the analysis chain information mapped with the key content of the text to obtain a content reasoning result, and verify the content reasoning result through the verification chain information to obtain a verification result;

[0318] The result verification module 14 is used to determine the task reasoning result for the reasoning task through the content reasoning result and the verification result.

[0319] The specific implementation of the text acquisition module 11, the template determination module 12, the content analysis module 13, and the result verification module 14 can be found in the above Figure 3 In the corresponding embodiment, steps S101 to S104, Figure 17 The description of steps S201 to S208 in the corresponding embodiment will not be repeated here.

[0320] The text acquisition module 11 includes: a text acquisition unit 111 and a result display unit 112 .

[0321] A text acquisition unit 111, configured to display an input display page associated with the reasoning task;

[0322] The text acquisition unit 111 is further configured to transmit the query data information to the target task reasoning model in response to a confirmation operation on the query data information entered in the input display page;

[0323] The text acquisition unit 111 is further configured to perform information recognition on the query data information using the target task reasoning model to obtain the data information type of the query data information;

[0324] The text acquisition unit 111 is further configured to, if the data information type includes a text data type and a non-text data type, acquire first query data content corresponding to the non-text data type from the query data information, acquire second query data content corresponding to the text data type, perform content conversion processing on the first query data content, and obtain first query conversion content for the first query data content;

[0325] The text acquisition unit 111 is further configured to determine a task description text associated with the reasoning task through the first query conversion content and the second query data content.

[0326] The text acquiring unit 111 is further configured to determine the query data information as a task description text if the data information type includes a text data type.

[0327] The text acquisition unit 111 is further configured to perform content conversion processing on the query data information if the data information type includes a non-text data type, obtain second query conversion content for the query data information, and use the second query conversion content as the task description text.

[0328] The result display unit 112 is used to output the task reasoning result to the result display page of the business client; the result display page includes a first display area and a second display area; the first display area is used to display the content reasoning result; the second display area includes a verification result display area, and the verification result display area is used to display the verified result.

[0329] The result display unit 112 is further configured to display feedback prompt information for providing result feedback on the result display page of the business client; the feedback prompt information includes first-category feedback information;

[0330] The result display unit 112 is further configured to determine the content reasoning result as the task execution result for the reasoning task in response to the feedback triggering operation for the first type of feedback information;

[0331] The result display unit 112 is further configured to output the task execution result to a result display page of the business client; the result display page includes a third display area; the third display area is configured to display the task execution result.

[0332] The result display unit 112 is further configured to respond to a feedback triggering operation for the second type of feedback information, rewrite the task reasoning result based on the second type of feedback information, and obtain an optimized task reasoning result;

[0333] The result display unit 112 is further configured to determine the optimization task reasoning result as the task execution result for the reasoning task;

[0334] The result display unit 112 is further configured to output the task execution result to a result display page of the business client; the result display page includes a third display area; the third display area is configured to display the task execution result.

[0335] The result display unit 112 is further configured to display, in response to a feedback operation on the verification result, a feedback text content for providing result feedback on the result display page of the business client; the feedback text content includes the first type of feedback content;

[0336] The result display unit 112 is further configured to determine the content reasoning result as a task execution result for the reasoning task based on the first type of feedback content;

[0337] The result display unit 112 is further configured to output the task execution result to a result display page of the business client; the result display page includes a third display area; the third display area is configured to display the task execution result.

[0338] The result display unit 112 is further used to rewrite the task reasoning result based on the second type of feedback content to obtain an optimized task reasoning result;

[0339] The result display unit 112 is further configured to determine the optimization task reasoning result as the task execution result for the reasoning task;

[0340] The result display unit 112 is further configured to output the task execution result to a result display page of the business client; the result display page includes a third display area; the third display area is configured to display the task execution result.

[0341] The specific implementation of the text acquisition unit 111 and the result display unit 112 can be found in the above Figure 3 In the corresponding embodiment, steps S101 to S104, Figure 17 The description of steps S201 to S208 in the corresponding embodiment will not be repeated here.

[0342] The template determination module 12 includes: a template determination unit 121 , a path generation unit 122 , a path optimization unit 123 , and a confidence determination unit 124 .

[0343] The template determination unit 121 is used to perform word segmentation extraction on the task description text to obtain multiple word segmentation texts for the task description text, and select a target word segmentation text that matches the reasoning task from the multiple word segmentation texts;

[0344] The template determination unit 121 is further configured to use the target segmented text as the text key content, perform content analysis on the text key content, and obtain content association relationships between the segmented texts in the text key content;

[0345] The template determining unit 121 is further configured to determine a task cognitive template adapted to the reasoning task from a plurality of cognitive templates associated with the reasoning task based on the text key content and content association relationships.

[0346] The template determination unit 121 is further configured to determine a task type of the reasoning task based on the key content of the text and the content association relationship;

[0347] The template determination unit 121 is further configured to obtain a plurality of cognitive templates that match the task type of the reasoning task, and obtain key content of template analysis in each cognitive template;

[0348] The template determination unit 121 is further configured to perform content matching between the text key content and the template analysis key content in each cognitive template to obtain a content matching degree associated with each cognitive template; a content matching degree is used to represent the degree of content matching between the template analysis key content in a cognitive template and the text key content;

[0349] The template determination unit 121 is further configured to obtain a maximum content matching degree from the content matching degrees, and determine a task cognitive template adapted to the reasoning task using the cognitive template corresponding to the maximum content matching degree.

[0350] The template determination unit 121 is further configured to determine the template text confidence of the first cognitive template as the first template text confidence, and determine the template text confidence of the second cognitive template as the second template text confidence; the first template text confidence is determined by the content confidence of the first template analysis key content in the first cognitive template; the second template text confidence is determined by the content confidence of the second template analysis key content in the second cognitive template;

[0351] The template determining unit 121 is further configured to obtain a minimum template text confidence from the first template text confidence and the second template text confidence, and use the cognitive template corresponding to the minimum template text confidence as the task cognitive template.

[0352] The template determining unit 121 is further configured to obtain a general cognitive template when multiple cognitive templates matching the task type of the reasoning task are not obtained, and determine the general cognitive template as the task cognitive template adapted to the reasoning task.

[0353] The path generation unit 122 is used to obtain multiple template chain information from the task recognition template and obtain the information coding identifier of each template chain information;

[0354] The path generation unit 122 is further configured to generate a link node corresponding to each template chain information according to the information coding identifier of each template chain information;

[0355] The path generation unit 122 is further configured to generate a reasoning path associated with the reasoning task through the link nodes corresponding to each template chain information.

[0356] The path optimization unit 123 is used to obtain each content text unit in the text key content and determine the unit confidence of each content text unit;

[0357] The path optimization unit 123 is further configured to determine the text confidence of the task description text based on the unit confidence of each content text unit, and determine the task complexity of the task description text based on the text complexity; the value of the task complexity is used to represent the difficulty of performing a text query on the task description text;

[0358] The path optimization unit 123 is further configured to optimize the reasoning path when the task complexity reaches a complex optimization condition, and obtain an optimized reasoning path;

[0359] The path optimization unit 123 is further configured to execute the reasoning task according to the updated reasoning path when the reasoning path is updated using the reasoning path after the path optimization.

[0360] The path optimization unit 123 is further configured to, if the value of the task complexity is less than or equal to the first confidence threshold, obtain template-added chain information for link optimization, and add a link node corresponding to the template-added chain information to the multiple link nodes included in the reasoning path to obtain a first reasoning path;

[0361] The path optimization unit 123 is further configured to, if the value of the task complexity is greater than or equal to a second confidence threshold, obtain template deletion chain information for link optimization from the plurality of template chain information, and remove the link node corresponding to the template deletion chain information from the plurality of link nodes included in the reasoning path to obtain a second reasoning path;

[0362] The path optimization unit 123 is further configured to use the first reasoning path or the second reasoning path as the reasoning path after path optimization.

[0363] A confidence determination unit 124, configured to obtain a minimum text division unit for performing unit division;

[0364] The confidence determination unit 124 is further configured to divide the key text content into units based on the minimum text division unit to obtain each content text unit in the key text content;

[0365] The confidence determination unit 124 is further configured to obtain the sample confidence of the sample text unit that matches each content text unit, and use the sample confidence of each sample text unit as the unit confidence of each content text unit.

[0366] The specific implementation of the template determination unit 121, the path generation unit 122, the path optimization unit 123, and the confidence determination unit 124 can be found in the above Figure 3 In the corresponding embodiment, steps S101 to S104, Figure 17 The description of steps S201 to S208 in the corresponding embodiment will not be repeated here.

[0367] The content analysis module 13 includes: a content analysis unit 131 .

[0368] The content analysis unit 131 is configured to use the information content in the first analysis class chain information as the first node key content of the first link node when performing the reasoning task according to the reasoning path;

[0369] The content analysis unit 131 is further configured to determine a location of content to be supplemented from the first node key content of the first link node, use the missing content at the location of the content to be supplemented as the first content to be processed, and use the content of the first node key content other than the first content to be processed as the second content to be processed;

[0370] The content analysis unit 131 is further configured to update the first to-be-processed content at the position of the to-be-supplemented content to the key text content when mapping the text key content to the first analysis class chain information, and obtain task mapping content information for the first analysis class chain information through the key text content and the second to-be-processed content;

[0371] The content analysis unit 131 is also used to use the information content in the second analysis class chain information as the second node key content of the second link node, and perform inference analysis on the task mapping content information through the task analysis strategy indicated by the second node key content to obtain a content inference result.

[0372] The specific implementation of the content analysis unit 131 can be found in the above Figure 3 In the corresponding embodiment, steps S101 to S104, Figure 17 The description of steps S201 to S208 in the corresponding embodiment will not be repeated here.

[0373] The result verification module 14 includes: a result determination unit 141 , a comparison relationship table generation unit 142 , and a comparison relationship table display unit 143 .

[0374] The result determination unit 141 is configured to use the information content in the verification chain information as the third node key content of the third link node, and determine the analysis strategy indicated by the third node key content as the verification analysis strategy;

[0375] The result determination unit 141 is further configured to, if the verification result indicates that there is an optimizable reasoning result in the content reasoning result, optimize the optimizable reasoning result in the content reasoning result according to the verification analysis strategy to obtain an optimized content reasoning result;

[0376] The result determination unit 141 is further configured to determine the optimized content reasoning result as a task reasoning result for the reasoning task.

[0377] A comparison relationship table generating unit 142 is used to generate a content comparison relationship table for comparing content reasoning results and optimizing content reasoning results;

[0378] The comparison relationship table display unit 143 is used to output the content comparison relationship table to the result display page of the business client; the result display page includes a second display area; the second display area includes a comparison relationship table display area; the comparison relationship table display area is used to display the content comparison relationship table.

[0379] The specific implementation of the result determination unit 141, the comparison relationship table generation unit 142, and the comparison relationship table display unit 143 can be found in the above Figure 3 In the corresponding embodiment, steps S101 to S104, Figure 17 The description of steps S201 to S208 in the corresponding embodiment will not be repeated here.

[0380] Optionally, the device 1 further includes: a model training module 15.

[0381] The model training module 15 is used to obtain training sample information associated with the target sample reasoning task, where the training sample information includes a sample task description text used to train the initial task reasoning model and a reference task reasoning result for the sample task description text; the sample reasoning task corresponding to the sample task description text is the target sample reasoning task;

[0382] The model training module 15 is further configured to determine a sample task cognitive template adapted to the target sample reasoning task based on key content of the sample text in the sample task description text, and to determine a sample reasoning path based on multiple template chain information in the sample task cognitive template; the multiple template chain information at least includes sample analysis chain information for performing reasoning analysis and sample verification chain information for performing reasoning verification;

[0383] The model training module 15 is further configured to map the key content of the sample text to the sample analysis class chain information when executing the target sample reasoning task according to the sample reasoning path, perform reasoning analysis on the sample analysis class chain information mapped with the key content of the sample text to obtain a sample content reasoning result, and verify the sample content reasoning result using the sample verification class chain information to obtain a sample verification result;

[0384] The model training module 15 is also used to determine the sample task reasoning results for the target sample reasoning task through the sample content reasoning results and the sample verification results, and to train the initial task reasoning model through the sample task reasoning results and the reference task reasoning results to obtain the target task reasoning model.

[0385] The specific implementation of the model training module 15 can be found in the above Figure 3 In the corresponding embodiment, steps S101 to S104, Figure 17 The description of steps S201 to S208 in the corresponding embodiment will not be repeated here.

[0386] The model training module 15 includes: a sample determination unit 151.

[0387] The sample determination unit 151 is configured to obtain N to-be-screened task description texts associated with a plurality of sample reasoning tasks, perform vector representation processing on the N to-be-screened task description texts, and obtain N sample text representation vectors; one sample text representation vector is used to represent one to-be-screened task description text; N is a positive integer;

[0388] The sample determination unit 151 is further configured to calculate the vector similarity between any two sample text representation vectors from the N sample text representation vectors to obtain M vector similarities, select at least one target vector similarity whose vector similarity is greater than or equal to a vector similarity threshold from the M vector similarities, and respectively determine the two to-be-screened task description texts associated with each screened target vector similarity as target screening task description texts; M is an integer greater than or equal to N;

[0389] The sample determination unit 151 is further configured to obtain the sample text confidence of each determined target screening task description text, and among the sample text confidences of each target screening task description text, use the target screening task description text corresponding to the minimum sample text confidence as the sample task description text, determine the sample reasoning task corresponding to the sample task description text as the target sample reasoning task, and determine the task reasoning result corresponding to the sample task description text as the reference task reasoning result; a sample text confidence is used to characterize the difficulty of the sample reasoning task corresponding to a target screening task description text;

[0390] The sample determination unit 151 is further configured to construct training sample information associated with the target sample reasoning task based on the sample task description text and the reference task reasoning result.

[0391] The specific implementation of the sample determination unit 151 can be found in the above Figure 3 In the corresponding embodiment, steps S101 to S104, Figure 17 The description of steps S201 to S208 in the corresponding embodiment will not be repeated here.

[0392] The description of the beneficial effects obtained by adopting the same method will not be further repeated here.

[0393] Further, see Figure 21 , Figure 21 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. Figure 20 As shown, the computer device 1000 may be a user terminal, for example, Figure 1 The user terminal 10a in the corresponding embodiment may also be a server, for example, Figure 1 The server 10d in the corresponding embodiment will not be limited here. For ease of understanding, this application takes the computer device as an example of a user terminal. The computer device 1000 may include: a processor 1001, a network interface 1004 and a memory 1005. In addition, the computer device 1000 may also include: a user interface 1003, and at least one communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication between these components. Among them, the user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or it may be a non-volatile memory (non-volatile memory), such as at least one disk storage. The memory 1005 may optionally also be at least one storage device located away from the aforementioned processor 1001. As Figure 20 As shown, the memory 1005 as a computer-readable storage medium may include an operating system, a network communication module, a user interface module, and a device control application.

[0394] The network interface 1004 in the computer device 1000 can also provide a network communication function, and the optional user interface 1003 can also include a display screen and a keyboard. Figure 13 In the computer device 1000 shown, the network interface 1004 can provide network communication functions; the user interface 1003 is mainly used to provide an input interface for the user; and the processor 1001 can be used to call the device control application stored in the memory 1005 to execute the above Figure 3 or Figure 17 The description of the content reasoning method in the corresponding embodiment can also be performed as described above. Figure 20 The description of the content inference device 1 in the corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of adopting the same method will not be repeated here either.

[0395] In addition, it should be pointed out that: the embodiment of the present application also provides a computer-readable storage medium, and the computer-readable storage medium stores the computer program executed by the content inference device 1 mentioned above, and the computer program includes computer instructions. When the processor executes the computer instructions, it can execute the above-mentioned content inference device 1. Figure 3 or Figure 17 The description of the content reasoning method in the corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of adopting the same method will not be repeated. For technical details not disclosed in the computer-readable storage medium embodiment involved in this application, please refer to the description of the method embodiment of this application. As an example, computer instructions can be deployed on a computer device for execution, or on multiple computer devices located in one location, or on multiple computer devices distributed in multiple locations and interconnected through a communication network. Multiple computer devices distributed in multiple locations and interconnected through a communication network can constitute a blockchain system.

[0396] In addition, it should be noted that: the embodiment of the present application also provides a computer program product or computer program, which may include computer instructions, which may be stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor may execute the computer instructions, so that the computer device performs the above Figure 3 or Figure 17 The description of the content reasoning method in the corresponding embodiment will not be repeated here. In addition, the description of the beneficial effects of using the same method will not be repeated here. For technical details not disclosed in the computer program product or computer program embodiment involved in this application, please refer to the description of the method embodiment of this application.

[0397] It should be noted that for the aforementioned various method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application.

[0398] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0399] The steps in the method of the embodiment of the present application can be adjusted in order, combined and deleted according to actual needs.

[0400] The modules in the device of the embodiment of the present application can be merged, divided and deleted according to actual needs.

[0401] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0402] The above disclosure is only a preferred embodiment of the present application, and certainly cannot be used to limit the scope of rights of the present application. Therefore, equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A content reasoning method, characterized in that: include: Obtaining the task description text associated with the reasoning task; Determining a task cognitive template adapted to the reasoning task based on key text content in the task description text, and determining a reasoning path associated with the reasoning task based on multiple template chain information in the task cognitive template; the multiple template chain information at least including analysis chain information for performing reasoning analysis and verification chain information for performing reasoning verification; When executing the reasoning task according to the reasoning path, mapping the text key content to the analysis chain information, performing reasoning analysis on the analysis chain information mapped with the text key content to obtain a content reasoning result, and performing result verification on the content reasoning result through the verification chain information to obtain a verification result; A task reasoning result for the reasoning task is determined based on the content reasoning result and the verification result.

2. The method according to claim 1, characterized in that The method is executed by a business client, on which a target task reasoning model is deployed; The obtaining of the task description text associated with the reasoning task includes: displaying an input display page associated with the reasoning task; In response to a confirmation operation on the query data information entered in the input display page, transmitting the query data information to the target task reasoning model; The target task reasoning model performs information recognition on the query data information to obtain the data information type of the query data information; If the data information type includes a text data type and a non-text data type, obtaining a first query data content corresponding to the non-text data type in the query data information, and obtaining a second query data content corresponding to the text data type, performing content conversion processing on the first query data content, and obtaining a first query conversion content for the first query data content; A task description text associated with the reasoning task is determined through the first query conversion content and the second query data content.

3. The method according to any one of claims 1-2, characterized in that The method further comprises: The task reasoning result is output to the result display page of the business client; the result display page includes a first display area and a second display area; the first display area is used to display the content reasoning result; the second display area includes a verification result display area, and the verification result display area is used to display the verified result.

4. The method according to claim 3, characterized in that The method further comprises: Displaying feedback prompt information for providing result feedback on a result display page of the business client; the feedback prompt information includes first-category feedback information; In response to a feedback triggering operation for the first type of feedback information, determining the task reasoning result as a task execution result for the reasoning task; The task execution result is output to a result display page of the business client; the result display page includes a third display area; the third display area is used to display the task execution result.

5. The method according to claim 4, characterized in that The feedback prompt information includes the second type of feedback information; The method further comprises: In response to a feedback triggering operation for the second type of feedback information, rewriting the task reasoning result based on the second type of feedback information to obtain an optimized task reasoning result; Determining the optimization task reasoning result as a task execution result for the reasoning task; The task execution result is output to a result display page of the business client; the result display page includes a third display area; the third display area is used to display the task execution result.

6. The method according to claim 1, characterized in that The determining of a task cognitive template adapted to the reasoning task based on key text content in the task description text includes: Performing word segmentation extraction on the task description text to obtain multiple word segmentation texts for the task description text, and screening a target word segmentation text that matches the reasoning task from the multiple word segmentation texts; Taking the target segmented text as the text key content, performing content analysis on the text key content to obtain content association relationships between the segmented texts in the text key content; Based on the text key content and the content association relationship, the task cognitive template adapted to the reasoning task is determined from a plurality of cognitive templates associated with the reasoning task.

7. The method according to claim 6, characterized in that The step of determining the task cognitive template adapted to the reasoning task from a plurality of cognitive templates associated with the reasoning task based on the key content of the text and the content association relationship includes: Determining a task type of the reasoning task based on the key content of the text and the content association relationship; Acquire multiple cognitive templates that match the task type of the reasoning task, and acquire key content of template analysis in each cognitive template; Performing content matching on the text key content and the template analysis key content in each cognitive template to obtain a content matching degree associated with each cognitive template; a content matching degree is used to represent the degree of content matching between the template analysis key content in a cognitive template and the text key content; A maximum content matching degree is obtained from the content matching degrees, and a task cognitive template adapted to the reasoning task is determined through a cognitive template corresponding to the maximum content matching degree.

8. The method according to claim 7, characterized in that The method further comprises: When multiple cognitive templates matching the task type of the reasoning task are not obtained, a general cognitive template is obtained, and the general cognitive template is determined as the task cognitive template adapted for the reasoning task.

9. The method according to claim 1, characterized in that The determining of the reasoning path associated with the reasoning task based on the plurality of template chain information in the task cognition template includes: Acquire multiple template chain information from the task recognition template, and acquire an information coding identifier of each template chain information; Generate a link node corresponding to each template chain information according to the information coding identifier of each template chain information; The reasoning path associated with the reasoning task is generated through the link nodes corresponding to each template chain information.

10. The method according to claim 1 or 9, characterized in that The method further comprises: Obtaining each content text unit in the key content of the text, and determining the unit confidence of each content text unit; The text confidence of the task description text is determined by the unit confidence of each content text unit, and the task complexity of the task description text is determined by the text complexity; the value of the task complexity is used to represent the difficulty of querying the task description text when performing a text query; When the task complexity reaches a complex optimization condition, performing path optimization on the reasoning path to obtain an optimized reasoning path; When the reasoning path is updated by the reasoning path after the path optimization, the reasoning task is executed according to the updated reasoning path.

11. The method according to claim 10, characterized in that The obtaining of each content text unit in the key content of the text and determining the unit confidence of each content text unit includes: Get the minimum text division unit used for unit division; Dividing the key text content into units based on the minimum text division unit to obtain each content text unit in the key text content; The sample confidences of the sample text units that match the respective content text units are obtained, and the sample confidences of the respective sample text units are used as the unit confidences of the respective content text units.

12. The method according to claim 10, characterized in that The reasoning path includes a plurality of link nodes; the complex optimization condition includes a first confidence threshold for adding nodes and a second confidence threshold for deleting nodes; the first confidence threshold is less than the second confidence threshold; When the task complexity reaches the complex optimization condition, the reasoning path is optimized to obtain the optimized reasoning path, including: If the value of the task complexity is less than or equal to the first confidence threshold, obtaining template-added chain information for link optimization, and adding a link node corresponding to the template-added chain information to the multiple link nodes included in the reasoning path to obtain a first reasoning path; If the value of the task complexity is greater than or equal to the second confidence threshold, obtaining template deletion chain information for link optimization from the multiple template chain information, and removing the link node corresponding to the template deletion chain information from the multiple link nodes included in the reasoning path to obtain a second reasoning path; The first reasoning path or the second reasoning path is used as the reasoning path after the path optimization.

13. The method according to claim 1, wherein The analysis chain information includes first analysis chain information and second analysis chain information; the reasoning path includes at least a first link node corresponding to the first analysis chain information and a second link node corresponding to the second analysis chain information; When executing the reasoning task according to the reasoning path, mapping the text key content to the analysis class chain information, performing reasoning analysis on the analysis class chain information mapped with the text key content, and obtaining a content reasoning result, including: When executing the reasoning task according to the reasoning path, using the information content in the first analysis class chain information as the first node key content of the first link node; Determine a location of content to be supplemented from the first node key content of the first link node, use the missing content at the location of content to be supplemented as first content to be processed, and use the content of the first node key content other than the first content to be processed as second content to be processed; When mapping the text key content to the first analysis chain information, updating the first to-be-processed content at the to-be-supplemented content position to the key text content, and obtaining task mapping content information for the first analysis chain information through the key text content and the second to-be-processed content; The information content in the second analysis class chain information is used as the second node key content of the second link node, and the task mapping content information is reasoned and analyzed through the task analysis strategy indicated by the second node key content to obtain the content reasoning result.

14. The method according to any one of claims 1 to 13, characterized in that The reasoning path also includes a third link node corresponding to the verification chain information; The method further comprises: Using the information content in the verification chain information as the third node key content of the third link node, and determining the analysis strategy indicated by the third node key content as the verification analysis strategy; If the verification result indicates that there is an optimizable reasoning result in the content reasoning result, optimizing the optimizable reasoning result in the content reasoning result according to the verification analysis strategy to obtain an optimized content reasoning result; The optimized content reasoning result is determined as a task reasoning result for the reasoning task.

15. The method according to claim 1, wherein The task reasoning result is obtained by reasoning the target task reasoning model according to the reasoning path of the task cognitive template; the target task reasoning model is obtained by training the initial task reasoning model; The method further comprises: Acquire training sample information associated with a target sample reasoning task, the training sample information including a sample task description text used to train the initial task reasoning model and a reference task reasoning result for the sample task description text; the sample reasoning task corresponding to the sample task description text is the target sample reasoning task; Determining a sample task cognitive template adapted for the target sample reasoning task based on key content of the sample text in the sample task description text, and determining a sample reasoning path based on multiple template chain information in the sample task cognitive template; the multiple template chain information at least includes sample analysis chain information for performing reasoning analysis and sample verification chain information for performing reasoning verification; When executing the target sample reasoning task according to the sample reasoning path, mapping the key content of the sample text to the sample analysis class chain information, performing reasoning analysis on the sample analysis class chain information mapped with the key content of the sample text to obtain a sample content reasoning result, and verifying the sample content reasoning result through the sample verification class chain information to obtain a sample verification result; The sample task reasoning result for the target sample reasoning task is determined through the sample content reasoning result and the sample verification result. The initial task reasoning model is trained through the sample task reasoning result and the reference task reasoning result to obtain the target task reasoning model.

16. The method according to claim 15, characterized in that The target sample reasoning task is a sample reasoning task that can participate in model training and is selected from multiple sample reasoning tasks; The obtaining of training sample information associated with the target sample reasoning task includes: Obtain N to-be-screened task description texts associated with the plurality of sample reasoning tasks, perform vector representation processing on the N to-be-screened task description texts to obtain N sample text representation vectors; one sample text representation vector is used to represent one to-be-screened task description text; N is a positive integer; Calculating the vector similarity between any two sample text representation vectors from the N sample text representation vectors to obtain M vector similarities, screening at least one target vector similarity whose vector similarity is greater than or equal to a vector similarity threshold from the M vector similarities, and determining the two to-be-screened task description texts associated with each screened target vector similarity as target screening task description texts, respectively; M is an integer greater than or equal to N; Obtaining the determined sample text confidence of each target screening task description text, among the sample text confidences of each target screening task description text, taking the target screening task description text corresponding to the minimum sample text confidence as the sample task description text, determining the sample reasoning task corresponding to the sample task description text as the target sample reasoning task, and determining the task reasoning result corresponding to the sample task description text as the reference task reasoning result; a sample text confidence is used to characterize the task difficulty of the sample reasoning task corresponding to a target screening task description text; Based on the sample task description text and the reference task reasoning result, the training sample information associated with the target sample reasoning task is constructed.

17. A content inference device, characterized in that: include: A text acquisition module, used to acquire task description text associated with the reasoning task; a template determination module for determining a task cognition template adapted for the reasoning task based on key textual content in the task description text, and determining a reasoning path associated with the reasoning task based on multiple template chain information in the task cognition template; the multiple template chain information in the task cognition template at least including analysis chain information for performing reasoning analysis and verification chain information for performing reasoning verification; a content analysis module for mapping the text key content to the analysis chain information when executing the reasoning task according to the reasoning path, performing reasoning analysis on the analysis chain information mapped with the text key content to obtain a content reasoning result, and performing result verification on the content reasoning result through the verification chain information to obtain a verification result; A result verification module is used to determine a task reasoning result for the reasoning task through the content reasoning result and the verification result.

18. A computer device, characterized in that: including memory and processor; The memory is connected to the processor, the memory is used to store a computer program, and the processor is used to call the computer program so that the computer device executes the method according to any one of claims 1 to 16.

19. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which is suitable for being loaded and executed by a processor, so that a computer device having the processor executes the method according to any one of claims 1 to 16.

20. A computer program product, characterized in that The method comprises a computer program / instruction, which implements the method according to any one of claims 1 to 16 when executed by a processor.

Images