User utterance data processing method and device, equipment and storage medium

Abstract

The invention relates to the technical field of user utterance data processing, in particular to a user utterance data processing method, device and equipment and a storage medium, and the user utterance data processing method comprises the following steps: generating user intention data corresponding to utterance data based on the utterance data input by a target user to a target agent; adjusting a weight corresponding to the user intention according to feedback information of a user using the target agent on the user intention to obtain a target weight corresponding to the user intention; determining a target intention corresponding to the utterance data in the user intention data based on a target weight corresponding to the user intention; and calling an API interface to process the target intention according to the parameter category information of the target intention to obtain a processing result corresponding to the target intention. According to the method and the device, the accuracy of judging the target intention can be improved, and the API interface called according to the parameter category information of the target intention can quickly and accurately process language data containing professional domain knowledge.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of user discourse data processing, and in particular, to a method, device, equipment and storage medium for processing user discourse data. Background Art

[0002] An Agent refers to an agent that can perceive the environment and take actions to achieve specific goals. Agents are widely used in the field of artificial intelligence and are commonly found in automated systems, robots, virtual assistants, and game characters, etc.

[0003] After large language models such as GPT (Generative Pre-Trained Transformer) have improved the ability of natural language models to communicate and understand semantics, agents have demonstrated amazing abilities in arithmetic, language communication, etc.; however, the computing power, manpower, and data requirements for the training of large language models are huge, and large language models do not have the ability to adjust in a timely manner, resulting in agents being unable to quickly and accurately process discourse data containing professional domain knowledge. Therefore, existing agents perform poorly in vertical domains. Summary of the Invention

[0004] The present application provides a method, device, equipment and storage medium for processing user discourse data, which can quickly and accurately determine the target intent corresponding to the discourse data containing professional domain knowledge, so as to improve the accuracy and efficiency of the discourse data processing.

[0005] In a first aspect, a method for processing user discourse data is provided, including: Generating user intent data corresponding to the discourse data based on the discourse data input by the target user to the target agent; the user intent data includes at least one user intent; Adjusting the weight corresponding to the user intent according to the feedback information of the user using the target agent on the user intent to obtain the target weight corresponding to the user intent; the feedback information includes indicating that the user intent is incorrect, indicating that the user intent is correct, and not adopting the user intent; Determining the target intent corresponding to the discourse data in the user intent data based on the target weight corresponding to the user intent; Processing the target intent by calling the API interface according to the parameter category information of the target intent to obtain the processing result corresponding to the target intent.

[0006] In a second aspect, a device for processing user discourse data is provided, including: A user intention generation module, configured to generate user intention data corresponding to the utterance data based on the utterance data input by the target user to the target intelligent agent; the user intention data includes at least one user intention. A weight adjustment module, configured to adjust the weight corresponding to the user intention according to the feedback information of the user using the target intelligent agent on the user intention, so as to obtain the target weight corresponding to the user intention; the feedback information includes indicating that the user intention is incorrect, indicating that the user intention is correct, and not adopting the user intention. A target intention determination module, configured to determine the target intention corresponding to the utterance data in the user intention data based on the target weight corresponding to the user intention. A processing module, configured to process the target intention by calling an API interface according to the parameter category information of the target intention, so as to obtain a processing result corresponding to the target intention.

[0007] In a third aspect, there is provided an electronic device, including: a processor and a memory, where the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to execute the method in the first aspect or its various implementation manners.

[0008] In a fourth aspect, there is provided a computer-readable storage medium, configured to store a computer program, and the computer program causes a computer to execute the method in the first aspect or its various implementation manners.

[0009] Through the technical solution provided by this application, by adjusting the weight corresponding to the user intention according to the feedback information of the user using the target intelligent agent on the user intention, while reducing the computational cost of determining the target intention in the user intention data, it can also improve the accuracy of the target intention determined based on the target weight corresponding to the user intention. Then, according to the API interface called by the parameter category information of the target intention, it can quickly and accurately complete the processing of the target intention, improve the accuracy of the processing result corresponding to the utterance data containing professional domain knowledge, so that the target intelligent agent can accurately and quickly process the utterance data containing professional domain knowledge in the vertical domain based on the solution provided by this application.

[0010] Other features and advantages of the present disclosure will be described in detail in the subsequent specific implementation part. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0012] Figure 1An application scenario diagram provided by an embodiment of the present application; Figure 2 A flowchart of a method for processing user discourse data provided by an embodiment of the present application; Figure 3 A schematic diagram of an initial category recommendation structure provided by an embodiment of the present application; Figure 4 A schematic diagram of an initial category recommendation structure provided by an embodiment of the present application; Figure 5 A schematic diagram of a category recommendation structure to be determined provided by an embodiment of the present application; Figure 6 A schematic diagram of a user discourse data processing device provided by an embodiment of the present application; Figure 7 It is a schematic block diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0013] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0014] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server including a series of steps or units does not necessarily need to be limited to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0015] An agent refers to an entity that can perceive the environment and take actions to achieve specific goals. It can be software, hardware, or a system, and has autonomy, adaptability, and interaction capabilities. An agent perceives changes in the environment, makes judgments and decisions based on the knowledge and algorithms learned by itself, and then executes actions to affect the environment or achieve a predetermined goal. Agents are widely used in the field of artificial intelligence, and are commonly found in automated systems, robots, virtual assistants, and game characters, etc. The core lies in the ability to learn autonomously and evolve continuously to better complete tasks and adapt to complex environments.

[0016] In the prior art, after large language models such as GPT have improved the ability of natural language models to communicate and semantically understand, agents have demonstrated amazing capabilities in arithmetic, language communication, etc. However, due to the huge computing power, labor, and data requirements for training large language models, and the lack of the ability to adjust in a timely manner in large language models, agents are unable to quickly and accurately process discourse data containing professional domain knowledge. Therefore, existing agents perform poorly in vertical domains.

[0017] To solve the above technical problems, the inventive concept of this application is as follows: Generate user intent data corresponding to the discourse data based on the discourse data input by the target user to the target agent; adjust the weight corresponding to the user intent according to the feedback information of the user using the target agent on the user intent to obtain the target weight corresponding to the user intent; based on the target weight corresponding to the user intent, determine the target intent corresponding to the discourse data in the user intent data; call the API (Application Program Interface) interface according to the parameter category information of the target intent to process the target intent and obtain the processing result corresponding to the target intent. By adjusting the weight corresponding to the user intent according to the feedback information of the user using the target agent in this application, while reducing the computational cost of determining the target intent in the user intent data, the accuracy of the target intent determined based on the target weight corresponding to the user intent can be improved. Then, according to the API interface called based on the parameter category information of the target intent, the processing of the target intent can be completed quickly and accurately, improving the accuracy of the processing result corresponding to the discourse data containing professional domain knowledge, so that the target agent can accurately and quickly process the discourse data containing professional domain knowledge in the vertical domain based on the solution provided by this application.

[0018] It should be understood that the technical solution of this application can be applied to the following scenarios, but is not limited to: In some implementable ways, Figure 1 This is an application scenario diagram provided for an embodiment of this application. As Figure 1 shown, this application scenario may include an electronic device 110 and a network device 120. The electronic device 110 can establish a connection with the network device 120 through a wired network or a wireless network.

[0019] Exemplarily, the electronic device 110 may be a desktop computer, a laptop computer, a tablet computer, etc., but is not limited thereto. The network device 120 may be a terminal device or a server, but is not limited thereto. In an embodiment of the present application, the electronic device 110 may send a request message to the network device 120, and the request message may be used to generate user intention data corresponding to the speech data based on the speech data input by the target user to the target intelligent agent. Further, the electronic device 110 may receive a response message sent by the network device 120, and the response message includes obtaining user intention data corresponding to the speech data generated based on the speech data input by the target user to the target intelligent agent.

[0020] In addition, Figure 1 An exemplary electronic device 110 and a network device 120 are given. In fact, other numbers of electronic devices and network devices may be included, and the present application does not limit this.

[0021] In some other implementable manners, the technical solution of the present application may also be executed by the above-mentioned electronic device 110, or the technical solution of the present application may also be executed by the above-mentioned network device 120, and the present application does not limit this.

[0022] After introducing the application scenarios of the embodiments of the present application, the technical solution of the present application will be elaborated in detail below: Figure 2 It is a flowchart of a method for processing user speech data provided for an embodiment of the present application. This method is applied to the server side and may be executed by the electronic device 110 as shown in Figure 1 but is not limited thereto. As shown in Figure 2 the method may include the following steps: S210. Generate user intention data corresponding to the speech data based on the speech data input by the target user to the target intelligent agent.

[0023] Among them, the user intention data includes at least one user intention; here, the user intention data may be obtained by the graph recognition module recognizing the speech data.

[0024] S220. Adjust the weight corresponding to the user intention according to the feedback information of the user using the target intelligent agent on the user intention, and obtain the target weight corresponding to the user intention.

[0025] Among them, the feedback information includes indicating that the user intention is incorrect, indicating that the user intention is correct, and not adopting the user intention; here, the feedback information may include the historical feedback information and the current feedback information of all users using the target intelligent agent on the user intention.

[0026] Here, the feedback information of the user using the target agent for the user intention can be understood as all the feedback information input by all users using the target agent for the user intention in the display interface; for example, the speech data input by the user of the target agent on January 17, 2025 is "What's the weather like today", and the following user intentions are displayed in the display interface: "Predicted rainfall data on January 17, 2025", "Predicted outdoor temperature on January 17, 2025", and "Predicted wind speed data on January 17, 2025"; if the user using the target agent indicates that the "Predicted rainfall data on January 17, 2025" is incorrect, then at the position corresponding to the "Predicted rainfall data on January 17, 2025" in the display interface, confirm that it is incorrect; if the user using the target agent indicates that the "Predicted wind speed data on January 17, 2025" is correct, then at the position corresponding to the "Predicted wind speed data on January 17, 2025" in the display interface, confirm that it is correct; if the user using the target agent does not adopt the "Predicted outdoor temperature on January 17, 2025", then the user does not perform an input action on whether the "Predicted outdoor temperature on January 17, 2025" in the display interface is correct.

[0027] In this step, according to the feedback information of the user using the target agent for the user intention, adjusting the weight corresponding to the user intention can individually adapt the weights corresponding to each user intention according to the feedback actions of the users using the target agent, making the target weight corresponding to the user intention have a high adaptability with the speech data, so as to avoid calling the API interface according to the parameter type information of the user intention when the user using the target agent does not adopt or obviously indicates an error in the user intention; at the same time, by introducing the feedback information of the user using the target agent for the user intention to adjust the weight corresponding to the user intention, the accuracy of the target intention judgment containing professional domain knowledge in step S230 can be improved based on the feedback information of the users of the target agent.

[0028] S230. Determine the target intention corresponding to the speech data in the user intention data based on the target weight corresponding to the user intention.

[0029] In this step, distance calculation can be performed based on the target weight corresponding to the user intention, so as to determine the target intention closest to the speech data in the user intention data according to the calculation result.

[0030] In addition, after determining the target intention corresponding to the speech data in the user intention data, it also includes uploading the target weight corresponding to the target intention to the weight server to update the weight data in the weight database.

[0031] S240. Call the API interface according to the parameter category information of the target intent to process the target intent, and obtain the processing result corresponding to the target intent.

[0032] Here, after calling the API interface according to the parameter category information of the target intent to process the target intent, the obtained processing result corresponding to the target intent is uploaded to the display interface for display.

[0033] By adopting the above method, by adjusting the weight corresponding to the user intent according to the feedback information of the user using the target agent on the user intent, while reducing the calculation cost of determining the target intent in the user intent data, it can also improve the accuracy of the target intent determined based on the target weight corresponding to the user intent. Then, according to the API interface called based on the parameter category information of the target intent, it can quickly and accurately complete the processing of the target intent, improve the accuracy of the processing result corresponding to the discourse data containing professional domain knowledge, so that the target agent can accurately and quickly process the discourse data containing professional domain knowledge in the vertical domain based on the solution provided by this application.

[0034] In some possible implementation embodiments, adjusting the weight corresponding to the user intent according to the feedback information of the user using the target agent on the user intent to obtain the target weight corresponding to the user intent may include the following steps: S310. If the feedback information corresponding to the user intent indicates that the user intent is incorrect, quickly reduce the weight corresponding to the user intent according to the preset penalty mechanism to obtain the target weight corresponding to the user intent.

[0035] Here, the preset penalty mechanism is a rule of linearly reducing the weight corresponding to the user intent with the first preset slope.

[0036] Taking the first preset slope as 0.5 as an example, if the user using the target agent clearly indicates on the display interface that the user intent is incorrect, the weight corresponding to the user intent can be calculated by the following formula:

[0037] where is the target weight of the user intent adjusted at the i-th time when the user using the target agent indicates that the user intent is incorrect at time t, is the target weight of the user intent adjusted at the (i - 1)-th time at time t - 1 according to the feedback information of the user using the target agent on the user intent; where can be understood as the target weight corresponding to the user intent, can be understood as the weight corresponding to the user intent that has not been adjusted according to the feedback information corresponding to the user intent.

[0038] From the above formula, it can be seen that when the user expression of the user intention using the target agent is incorrect, the target weight corresponding to the user intention at the current time is half of the target weight corresponding to the user intention at the previous moment; thus, it can be seen that by selecting the preset penalty mechanism as the rule that the weight corresponding to the user intention linearly decreases at the first preset slope, the weight corresponding to the user intention can be quickly reduced to reduce the hit probability of this user intention.

[0039] It should be noted that the initial weight corresponding to the user intention can be determined according to the user intention in the weight database and the weight data corresponding to the user intention.

[0040] S320. If the feedback information corresponding to the user intention indicates that the user intention is correct, then increase the weight corresponding to the user intention according to the preset reward mechanism to obtain the target weight corresponding to the user intention.

[0041] Here, the preset reward mechanism is the rule that the weight corresponding to the user intention linearly increases at the second preset slope.

[0042] Taking the second preset slope as e as an example, if the user using the target agent clearly indicates that the user intention is correct on the display interface, the weight corresponding to the user intention can be calculated by the following formula:

[0043] Among them, is the target weight of the user intention adjusted for the i-th time at time t when the user using the target agent indicates that the user intention is correct, is the target weight of the user intention adjusted for the (i - 1)-th time at time t - 1 according to the feedback information of the user using the target agent on the user intention.

[0044] From the above example, it can be seen that when the user expression of the user intention using the target agent is correct, the target weight corresponding to the user intention at the current time is e times the target weight corresponding to the user intention at the previous moment; thus, it can be seen that by setting the preset reward mechanism as the rule that the weight corresponding to the user intention linearly increases at the second preset slope, the weight corresponding to the user intention can be quickly increased to increase the hit probability of this user intention.

[0045] S330. If the feedback information corresponding to the user intention indicates that the user intention is not adopted, then when the number of times the user using the target agent does not adopt the user intention reaches the first threshold, slowly reduce the weight corresponding to the user intention according to the preset corruption mechanism to obtain the target weight corresponding to the user intention.

[0046] Here, the preset corruption mechanism is the rule that slowly reduces the weight corresponding to the user intention according to the preset logarithmic function.

[0047] It should be noted that the feedback information corresponding to the user intention is that the user intention is not adopted. This may be because the user using the target intelligent agent wants to input other discourse data to the target intelligent agent, resulting in no feedback on this user intention. It may also be that the user using the target intelligent agent believes that the user intention deviates greatly from the discourse data he / she inputs, resulting in no feedback on this user intention. Since when the number of times the user using the target intelligent agent does not adopt the user intention is small, the reason for the user using the target intelligent agent not to adopt the user intention tends to be that he / she wants to input other discourse data to the target intelligent agent. And when the number of times the user using the target intelligent agent does not adopt the user intention is large, the reason for the user using the target intelligent agent not to adopt the user intention tends to be that the user believes that the user intention deviates greatly from the discourse data he / she inputs. Therefore, when the number of times the user using the target intelligent agent does not adopt the user intention reaches the first threshold, it is considered that the reason for the user using the target intelligent agent not to adopt the user intention tends to be that the user believes that the user intention deviates greatly from the discourse data he / she inputs, and it is determined that the weight corresponding to the user intention needs to be adjusted. When the number of times the user using the target intelligent agent does not adopt the user intention does not reach the first threshold, it is considered that the reason for the user using the target intelligent agent not to adopt the user intention tends to be that he / she wants to input other discourse data to the target intelligent agent, and it is determined that the weight corresponding to the user intention does not need to be adjusted.

[0048] Taking the preset logarithmic function as the log function as an example, if the feedback information corresponding to the user intention is that the user intention is not adopted, the weight corresponding to the user intention can be calculated by the following formula:

[0049] Wherein, is the target weight adjusted for the user intention at the i-th time when the user using the target intelligent agent does not adopt the user intention at time t, is the target weight adjusted for the user intention at the (i - 1)-th time according to the feedback information of the user using the target intelligent agent on the user intention at time t - 1.

[0050] In addition, here the value of the target weight corresponding to the user intention is also limited to a maximum of 1, that is, when the target weight adjusted for the user intention at the i-th time by the user using the target intelligent agent at time t according to the feedback information of the user using the target intelligent agent on the user intention is less than 1, 1 is used as the weight value corresponding to the target weight.

[0051] By adopting the above method, it is possible to adjust the weight corresponding to the user intention with different mechanisms according to different feedback information of the user using the target intelligent agent on the user intention, so that the adjusted target weight corresponding to the user intention can be separately adapted to the actual feedback situation of the user using the target intelligent agent, and the target weight corresponding to the user intention has a high adaptability to the discourse data.

[0052] In some possible embodiments, after adjusting the weight corresponding to the user intention according to the feedback information of the user intention by the user using the target agent to obtain the target weight corresponding to the user intention, the following steps may further be included: S410. When the number of users using the target agent is greater than the second threshold, calculate the user weight corresponding to each user using the target agent according to the number of users using the target agent.

[0053] Since the weight corresponding to the user intention is adjusted according to the feedback information of the user intention input by all users using the target agent in the display interface; however, when a certain user using the target agent repeatedly inputs the same feedback information about the user intention on the display interface for some reason, the behavior of this user will affect other users using the target agent, and this user will also dominate the input result of the feedback information of other users about the user intention on the display interface due to inputting a large number of the same feedback information, so that the target weight corresponding to the user intention is easily attacked consciously or unconsciously by users with the above abnormal behavior; therefore, it is necessary to limit the influence of the user's personal target weight of the user using the target agent within the scope of the user's personal, so here, the user weight corresponding to each user using the target agent is calculated, so that the final target weight obtained in step S430 is limited under the behavior of the target user, so as to avoid the interference of the behavior of users with abnormal behavior on the final target weight.

[0054] Here, the calculation formula of the user weight corresponding to each user using the target agent is as follows:

[0055] wherein, is the user weight corresponding to the i-th user using the target agent, and n is the number of users using the target agent.

[0056] S420. Fuse the user weight and the target weight to obtain the final target weight corresponding to the user intention.

[0057] Here, the user weight and the target weight can be fused through the following formula:

[0058] wherein, is the final target weight after fusing the target weight and the user weight adjusted for the i-th time according to the feedback information of the user using the target agent at the t-th moment of the user intention; is the final target weight after fusing the target weight adjusted according to the feedback information of the user using the target agent for the (i-1)-th time at time t-1 and the user weight; n is the number of users using the target agent; is the i-th user; represents that the user intention is incorrect; means that the user intention is not adopted; represents that the user intention is correct. Among them, can be understood as the current final target weight corresponding to the user intention, can be understood as the final target weight corresponding to the user intention at the previous moment.

[0059] In addition, here there is also a limit that the value of the final target weight corresponding to the user intention is at most 1, that is, for the calculated according to the above formula, when it is less than 1, 1 is taken as the weight value corresponding to the final target weight.

[0060] By adopting the above method, through fusing the user weight and the target weight, the influence of the user's personal target weight on the user intention of the user using the target agent is restricted within the scope of the user's personal, so as to avoid the target weight corresponding to the user intention being interfered by the behaviors of users with abnormal behaviors.

[0061] In some possible implementation embodiments, based on the target weight corresponding to the user intention, determining the target intention corresponding to the discourse data in the user intention data may include the following steps: S510. For each user intention, calculate the similarity between the user intention and the discourse data according to the target weight corresponding to the user intention.

[0062] Here, when calculating the similarity between the user intention and the discourse data, the similarity calculation result between the user intention and the discourse data can be obtained by calculating the distance of the target weight corresponding to the user intention.

[0063] S520. Determine the target intention in the user intention data according to the similarity calculation result between the user intention and the discourse data.

[0064] Here, when determining the target intention in the user intention data, the user intention with the highest similarity to the discourse data can be used as the target intention.

[0065] In addition, after determining the target intention in the user intention data, the target intention and the target weight corresponding to the target intention can be updated in the weight server to update the target weight corresponding to the user intention corresponding to the target intention in the weight database.

[0066] By using the above method, the similarity calculation result between the user intention and the discourse data calculated according to the target weight corresponding to the user intention can be used to quickly and accurately determine the target intention from the user intention data.

[0067] In some possible embodiments, the parameter category information includes a parameter category and score information corresponding to the parameter category; processing the target intention by calling an API interface according to the parameter category information of the target intention to obtain a processing result of the target intention may include the following steps: S610. When the score information corresponding to the parameter category of the target intention is less than a third threshold, input the parameter category and its corresponding score information into the constructed category recommendation structure to obtain a parameter recommendation category corresponding to the target intention.

[0068] Wherein, before inputting the parameter category of the target intention and the score information corresponding to the parameter category of the target intention into the constructed category recommendation structure to obtain a parameter recommendation category corresponding to the target intention, it further includes: inputting the target intention into a parameter extraction module or a large model to obtain the parameter category and the score information corresponding to the parameter category of the target intention.

[0069] Here, the category recommendation structure is used to recommend the parameter category of the target intention according to the parameter category information of the user intention, and the category recommendation structure is constructed by inputting the parameter category corresponding to the user intention, the score information corresponding to the parameter category, and the historical parameter category feedback by the user on the parameter category into a first Markov chain; wherein, the parameter category corresponding to the user intention, the score information corresponding to the parameter category, and the historical parameter category feedback by the user on the parameter category can all be historical data information stored in the big data.

[0070] It should be noted that the score information corresponding to the parameter category of the target intention being less than the third threshold may be caused by inaccurate extraction of the parameter category information of the target intention by the large model or the parameter extractor, or may be caused by inaccurate determination of the target intention in the user intention data. Therefore, in order to be able to call an accurate API interface, here, by inputting the parameter category and its corresponding score information into the constructed category recommendation structure to obtain a parameter recommendation category corresponding to the target intention, so that the target user has more opportunities to select the API interface for the discourse data input by the target user to the target agent.

[0071] Exemplarily, the parameter categories corresponding to the target intention are A, B, C, and the score information corresponding to the parameter category is 30 points. Then, it is necessary to input the parameter categories A, B, C and the score information of 30 points into the constructed category recommendation structure to obtain the parameter recommendation categories corresponding to the target intention as A, B, D.

[0072] S620. Upload the parameter category and parameter recommendation category corresponding to the target intention to the interaction interface so that the target user can select the parameter category or the parameter recommendation category of the target intention in the interaction interface.

[0073] Continuing with the example in step S610, upload the parameter categories A, B, C corresponding to the target intention and the parameter recommendation categories A, B, D to the interaction interface so that the target user can select parameter recommendation categories A, B, C or parameter recommendation categories A, B, D in the interaction interface.

[0074] It should be noted that when the parameter category and the parameter recommendation category of the target intention are the same, the API interface is directly called according to the parameter category of the target intention.

[0075] S630. If the target user selects the parameter category of the target intention, call the API interface corresponding to the parameter category of the target intention to process the target intention.

[0076] S640. If the target user selects the parameter recommendation category, call the API interface corresponding to the parameter recommendation category to process the target intention.

[0077] Here, the target user's selection of the parameter recommendation category can be understood as changing the target intention. Therefore, when the target user selects the parameter recommendation category here, it is actually re - determining the target intention, thereby improving the accuracy of the intention recognition of the dialogue data.

[0078] Using the above - mentioned method, by uploading the parameter recommendation category and the parameter category corresponding to the target intention obtained by inputting the parameter category corresponding to the target intention and its corresponding score information into the constructed category recommendation structure to the interaction interface for the target user to select the parameter recommendation category and the parameter category corresponding to the target intention according to the dialogue data in the interaction interface, it can not only improve the target user's sense of participation, but also determine the target intention with the highest similarity to the dialogue data through the target user's selection of the parameter recommendation category and the parameter category corresponding to the target intention, enhance the accuracy of the target intention, and further improve the accuracy of the called API interface.

[0079] Furthermore, the category recommendation structure is constructed by the following method: S710. Input the parameter category of the user intention, the score information corresponding to the parameter category, and the historical parameter category information corresponding to the parameter category into the first Markov chain to obtain the initial category recommendation structure.

[0080] Here, the first Markov chain includes at least one first recommendation path, and the first recommendation path includes a first score range, the parameter recommendation category corresponding to the first score range, and the selection probability information corresponding to the parameter recommendation category.

[0081] Among them, the selection probability information corresponding to each parameter recommendation category is determined according to the proportion of the number of each parameter recommendation category corresponding to the first score range. For example, the parameter recommendation categories corresponding to the first score range include parameter recommendation category 1 and parameter recommendation category 2, and the number of parameter recommendation category 1 is 1, and the number of parameter recommendation category 2 is 3. Then the selection probability information corresponding to parameter recommendation category 1 is 25%, and the selection probability information corresponding to parameter recommendation category 2 is 75%.

[0082] Taking the parameter categories of the user intention, the score information corresponding to the parameter categories, and the historical parameter category information in the following table as examples, the parameter categories, score information, and historical parameter category information of the target intention in the following table are input into the first Markov chain to obtain the initial category recommendation structure as shown in Figure 3 shown.

[0083]

[0084] Since the parameter categories of the user intention are two groups, namely A, B, C and A, D, the initial category recommendation structure constructed here includes two first Markov chains, namely the first Markov chain corresponding to the parameter categories A, B, C of the user intention, and the first Markov chain corresponding to the parameter categories A, D of the user intention.

[0085] S720. Determine whether the first recommendation path needs to be split according to the number of parameter recommendation categories recommended by the first recommendation path.

[0086] Here, when the number of parameter recommendation categories recommended by the first recommendation path is greater than 1, it can be determined that the first recommendation path needs to be split.

[0087] Continuing with the example in step S720, in the first Markov chain corresponding to A, B, C, it can be seen that the number of parameter recommendation categories recommended by the first recommendation path corresponding to the first score range greater than 45 points is 2, so it is determined that this first recommendation path needs to be split; in the first Markov chain corresponding to A, D, it can be seen that the number of parameter recommendation categories recommended by the first recommendation path corresponding to the first score range greater than 50 points is 1, and the number of parameter recommendation categories recommended by the first recommendation path corresponding to the first score range less than 50 points is also 1, so it is determined that the first recommendation paths in this first Markov chain do not need to be split S730. If a first recommendation path needs to be split, calculate the original information entropy corresponding to the initial category recommendation structure according to the selection probability information corresponding to each parameter recommendation category recommended by the initial category recommendation structure.

[0088] Among them, the original information entropy corresponding to the initial category recommendation structure is calculated through the following formula:

[0089] In the formula, is the original information entropy, is the selection probability information corresponding to the recommended category of the i-th parameter, and n is the number of parameter recommended categories recommended by the initial category recommendation structure.

[0090] Continuing with the example where the parameter categories of the user intention in step S710 are A, B, and C, the original information entropy corresponding to this initial category recommendation structure can be calculated by the above formula as: entroy = -1.0 * log21 - 0.4 * log20.4 - 0.6 * log20.6 = 0.97.

[0091] S740. For the first recommendation path that needs to be split, divide the first score range corresponding to the first recommendation path and the parameter recommended category corresponding to the first score range according to a preset score to obtain a category recommendation structure to be determined.

[0092] Here, the category recommendation structure to be determined includes the first recommendation path that does not need to be split and the second recommendation path split from the first recommendation path that needs to be split according to the preset score; the second recommendation path includes a second score range, the parameter recommended sub-categories corresponding to the second score range, and the selection probability information corresponding to the parameter recommended sub-categories.

[0093] It should be noted that the value of the preset score is within the first score range.

[0094] Continuing with the example where the parameter categories of the user intention in step S730 are A, B, and C, if the preset score is set to 49 points here, the second score range divided from the first score range where the score is greater than 45 points includes the score range between 45 points and 49 points and the score range where the score is greater than 49 points. And when the second score range is between 45 points and 49 points, the parameter recommended sub-categories are A, B, E, and the selection probability information corresponding to A, B, E is 100%; when the second score range is greater than 49 points, the parameter recommended sub-categories are A, B, C, and the selection probability information corresponding to A, B, C is 100%. Then, based on the second score range corresponding to the above second recommendation path, the parameter recommended sub-categories corresponding to the second score range, and the selection probability information corresponding to the parameter recommended sub-categories, complete the splitting of the first recommendation path where the first score range is greater than 45 points, and obtain Figure 4 the category recommendation structure to be determined as shown.

[0095] S750. Calculate the current information entropy corresponding to the category recommendation structure to be determined according to the selection probability information corresponding to the parameter recommended categories and parameter recommended sub-categories recommended by the category recommendation structure to be determined.

[0096] Here, the current information entropy corresponding to the category recommendation structure to be determined can be calculated based on the calculation formula in step S730. Taking the category recommendation structure to be determined in the example of step S740 as an example, the current information entropy is entopy = -1.0 * log21 - 1 * log21 - 1 * log21 = 0.

[0097] S760. Determine whether the category recommendation structure to be determined is the final category recommendation structure according to the original information entropy and the current information entropy.

[0098] Here, it can be set that when the current information entropy is not greater than half of the original information entropy, it is determined that the category recommendation structure to be determined corresponding to this current information entropy is the final category recommendation structure.

[0099] Taking the example of step S750, since the original information entropy corresponding to the initial category recommendation structure is 0.97 and the current information entropy corresponding to the category recommendation structure to be determined is 0, it can be determined that the category recommendation structure to be determined is the final category recommendation structure.

[0100] S770. If not, divide the first score range corresponding to the first recommendation path according to the new preset score, and repeat the above steps until it is determined that the category recommendation structure to be determined is the final category recommendation structure.

[0101] Among them, when setting the new preset score, it can be set in sequence based on the preset score set in step S740 and the first score range corresponding to the first recommendation path that needs to be split. For example, when the preset score set in step S740 is 48, the new preset score set in this step can be 49. In this way, by setting the scores within the first score range corresponding to the first recommendation path that needs to be split as the preset scores in sequence, the final category recommendation structure can be accurately determined.

[0102] Using the above method, the constructed category recommendation structure can accurately recommend the parameter recommendation category corresponding to the target intention based on the parameter category of the target intention and its corresponding score information, so that the selection result between the parameter recommendation category recommended by the category recommendation structure and the parameter category corresponding to the target intention can accurately determine the target intention with the highest similarity to the discourse data, thereby improving the accuracy of the called API interface.

[0103] In some possible implementation embodiments, calling the API interface according to the parameter category information of the target intention to process the target intention, and obtaining the processing result corresponding to the target intention may include the following steps: S810. Input the parameter category information of the target intention into the constructed parameter value recommendation structure to obtain a set of recommended parameter values.

[0104] Since when extracting the parameter category information of the target intention, if the large model or the parameter extractor extracts inaccurately, or the parameter category information extracted by the large model or the parameter extractor is missing, it is likely to cause the inaccuracy of the called API interface. Therefore, by inputting the parameter category information of the target intention into the constructed parameter value recommendation structure to obtain a set of recommended parameter values, it is possible to automatically complete the recommended parameter value set by leveraging the previous behavior accumulation of the users using the target intelligent agent, so as to avoid the inaccuracy of the called API interface caused by the inaccurate extraction of the large model or the parameter extractor, or the missing of the extracted parameter category information.

[0105] Here, the set of recommended parameter values includes at least two sets of recommended parameter value sets corresponding to the parameter category information; the parameter value recommendation structure is used to recommend the parameter value set of the target intention according to the parameter category information of the target intention; the parameter value recommendation structure is obtained by inputting the parameter category information corresponding to the user intention and the historical parameter values feedback by the user on the parameter category information into the second Markov chain; among them, the parameter category information corresponding to the user intention and the historical parameter values feedback by the user on the parameter category information can both be historical data information stored in the big data.

[0106] Taking the parameter category information corresponding to the user intention in the following table and the historical parameter values feedback by the user on the parameter category information as an example, input the parameter category information and the historical parameter values in the following table into the second Markov chain to obtain as Figure 5 shown in the parameter value recommendation structure.

[0107]

[0108] S820. Upload the set of recommended parameter values to the interaction interface so that the target user can select the target parameter value set corresponding to the parameter category information from the set of recommended parameter values in the interaction interface.

[0109] Here, uploading the set of recommended parameter values to the interaction interface for the target user to select the target parameter values in the set of recommended parameter values can improve the user experience of selecting the recommended parameter values.

[0110] Continuing with the example in step S810, when uploading the set of recommended parameter values to the interaction interface, the set of recommended parameter values includes [A = 10, B = 13, C = 1], [A = 10, B = 5, C = 2], and [A = 10, B = 5, C = 1], so that the target user can select the target parameter value set corresponding to the parameter category information from the three parameter value sets in the set of recommended parameter values on the interaction interface.

[0111] S830. According to the target parameter value set corresponding to the parameter category information, call the API interface to process the target intent.

[0112] Using the above method, by allowing the target user to select from the set of recommended parameter values recommended by the parameter value recommendation structure on the interaction interface, not only can the user experience of selecting recommended parameter values be improved, but also the set of recommended parameter values can be automatically completed with the help of the parameter value recommendation structure, so as to avoid the inaccurate invocation of the API interface caused by inaccurate extraction by the large model or the parameter extractor, or the missing parameter category information extracted.

[0113] Figure 6 It is a schematic diagram of a user discourse data processing device 900 according to an embodiment provided by the present invention. The device 900 is applied to a collection terminal, such as Figure 6 As shown, the device 900 includes: A user intent generation module 910, configured to generate user intent data corresponding to the discourse data based on the discourse data input by the target user to the target intelligent agent; the user intent data includes at least one user intent identified from the discourse data; A weight adjustment module 920, configured to adjust the weight corresponding to the user intent according to the feedback information of the user using the target intelligent agent for the user intent, to obtain the target weight corresponding to the user intent; the feedback information includes indicating that the user intent is incorrect, indicating that the user intent is correct, and not adopting the user intent; A target intent determination module 930, based on the target weight corresponding to the user intent, determines the target intent corresponding to the discourse data in the user intent data; A processing module 940, configured to call the API interface to process the target intent according to the parameter category information of the target intent, to obtain the processing result corresponding to the target intent.

[0114] In some implementable manners, the weight adjustment module 920 includes: A first target weight adjustment unit, configured to, if the feedback information corresponding to the user intent indicates that the user intent is incorrect, quickly reduce the weight corresponding to the user intent according to a preset penalty mechanism, to obtain the target weight corresponding to the user intent; the preset penalty mechanism is a rule of linearly reducing the weight corresponding to the user intent at a first preset slope; The second target weight adjustment unit is configured to, if the feedback information corresponding to the user intention indicates that the user intention is correct, increase the weight corresponding to the user intention according to a preset reward mechanism to obtain the target weight corresponding to the user intention; the preset reward mechanism is a rule of linearly increasing the weight corresponding to the user intention at a second preset slope; The third target weight adjustment unit is configured to, if the feedback information corresponding to the user intention does not adopt the user intention, when the number of times the user who uses the target agent does not adopt the user intention reaches a first threshold, slowly reduce the weight corresponding to the user intention according to a preset corruption mechanism to obtain the target weight corresponding to the user intention; the preset corruption mechanism is a rule of slowly reducing the weight corresponding to the user intention according to a preset logarithmic function.

[0115] In some implementable ways, it further includes: The user weight calculation module is configured to, when the number of users using the target agent is greater than a second threshold, calculate the user weight corresponding to each user using the target agent according to the number of users using the target agent; The target weight generation module is configured to fuse the user weight and the target weight to obtain the final target weight corresponding to the user intention.

[0116] In some implementable ways, the target intention determination module 930 includes: The similarity calculation unit is configured to, for each user intention, calculate the similarity between the user intention and the discourse data according to the target weight corresponding to the user intention; The target intention determination unit is configured to determine the target intention in the user intention data according to the similarity calculation result between the user intention and the discourse data.

[0117] In some implementable ways, the parameter category information includes a parameter category and score information corresponding to the parameter category; the processing module includes: The category recommendation unit is configured to, when the score information corresponding to the parameter category of the target intention is less than a third threshold, input the parameter category and its corresponding score information into the constructed category recommendation structure to obtain the parameter recommended category corresponding to the target intention; the category recommendation structure is used to recommend the parameter category of the target intention according to the parameter category information of the target intention, and the category recommendation structure is constructed by inputting the parameter category corresponding to the user intention, the score information corresponding to the parameter category, and the historical parameter category feedback by the user on the parameter category into a first Markov chain; The parameter category upload unit is configured to upload the parameter category and the parameter recommended category corresponding to the target intention to the interaction interface so that the target user can select the parameter category or the parameter recommended category of the target intention in the interaction interface; A first processing unit, configured to, if a target user selects a parameter category of a target intent, call an API interface corresponding to the parameter category of the target intent to process the target intent; A second processing unit, configured to, if the target user selects a parameter recommendation category, call an API interface corresponding to the parameter recommendation category to process the target intent.

[0118] In some implementable manners, it further includes a category recommendation structure construction module, and the category recommendation structure construction module includes: An initial category recommendation structure construction unit, configured to input the parameter category of the target intent, the score information corresponding to the parameter category, and the historical parameter category information corresponding to the parameter category into a first Markov chain to obtain an initial category recommendation structure; the first Markov chain includes at least one first recommendation path, and the first recommendation path includes a first score range, a parameter recommendation category corresponding to the first score range, and selection probability information corresponding to the parameter recommendation category; A recommendation path splitting judgment unit, configured to judge whether the first recommendation path needs to be split according to the number of parameter recommendation categories recommended by the first recommendation path; An original information entropy calculation unit, configured to, if a first recommendation path needs to be split, calculate the original information entropy corresponding to the initial category recommendation structure according to the selection probability information corresponding to each parameter recommendation category recommended by the initial category recommendation structure; A first structure splitting unit, configured to, for the first recommendation path that needs to be split, divide the first score range corresponding to the first recommendation path and the parameter recommendation category corresponding to the first score range according to a preset score to obtain a category recommendation structure to be determined; the category recommendation structure to be determined includes a first recommendation path that does not need to be split and a second recommendation path split from the first recommendation path that needs to be split according to the preset score; the second recommendation path includes a second score range, a parameter recommendation sub-category corresponding to the second score range, and selection probability information corresponding to the parameter recommendation sub-category; A current information entropy calculation unit, configured to calculate the current information entropy corresponding to the category recommendation structure to be determined according to the selection probability information corresponding to the parameter recommendation category and the parameter recommendation sub-category recommended by the category recommendation structure to be determined; A structure judgment unit, configured to judge whether the category recommendation structure to be determined is the final category recommendation structure according to the original information entropy and the current information entropy; A second structure splitting unit, configured to, if not, divide the first score range corresponding to the first recommendation path according to a new preset score, and repeat the above steps until it is judged that the category recommendation structure to be determined is the final category recommendation structure.

[0119] In some implementable manners, the processing module includes: A recommended parameter value set obtaining unit is configured to input parameter category information of a target intention into a constructed parameter value recommendation structure to obtain a recommended parameter value set; the recommended parameter value set includes at least two sets of recommended parameter value sets corresponding to the parameter category information; the parameter value recommendation structure is used to recommend a parameter value set of the target intention according to the parameter category information of the target intention; the parameter value recommendation structure is obtained by inputting parameter category information corresponding to a user intention and historical parameter values fed back by the user for the parameter category information into a second Markov chain. A recommended parameter value set uploading unit is configured to upload the recommended parameter value set to an interaction interface, so that a target user can select a target parameter value set corresponding to the parameter category information from the recommended parameter value set in the interaction interface. A processing unit is configured to call an API interface to process the target intention according to the target parameter value set corresponding to the parameter category information.

[0120] It should be understood that the embodiments of the user discourse data processing device and the embodiments of the user discourse data processing method can correspond to each other, and similar descriptions can refer to the embodiments of the user discourse data processing method. To avoid repetition, details are not described here. Specifically, Figure 6 The illustrated device 900 can execute the above-mentioned embodiments of the user discourse data processing method, and the foregoing and other operations and / or functions of each module in the device 900 respectively implement the corresponding processes in the above-mentioned user discourse data processing method. For the sake of brevity, details are not described here.

[0121] In the foregoing, the device 900 of the embodiments of the present invention has been described from the perspective of functional modules in conjunction with the accompanying drawings. It should be understood that the functional modules can be implemented in the form of hardware, can also be implemented by instructions in software form, and can also be implemented by a combination of hardware and software modules. Specifically, each step of the user discourse data processing method and the detection method embodiments in the embodiments of the present invention can be completed by the integrated logic circuit in hardware in a processor and / or instructions in software form. Combining the steps of the user discourse data processing method and the detection method disclosed in the embodiments of the present invention can be directly embodied as being executed and completed by a hardware decoding processor, or executed and completed by a combination of hardware and software modules in the decoding processor. Optionally, the software module can be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory, and the processor reads the information in the memory and combines its hardware to complete the steps in the above-mentioned user discourse data processing method and detection method embodiments.

[0122] Figure 7 is a schematic block diagram of an electronic device 110 according to an embodiment provided by the present invention.

[0123] AsFigure 7 As shown, the electronic device 110 may include: A memory 111 and a processor 112. The memory 111 is used to store a computer program and transmit the program code to the processor 112. In other words, the processor 112 can call and run the computer program from the memory 111 to implement the method in the embodiments of the present invention.

[0124] For example, the processor 112 can be used to execute the above method embodiments according to the instructions in the computer program.

[0125] In some embodiments of the present invention, the electronic device 110 may include but is not limited to: A general-purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and so on.

[0126] In some embodiments of the present invention, the memory 111 includes but is not limited to: Volatile memory and / or non-volatile memory. Among them, the non-volatile memory can be Read-Only Memory (ROM), Programmable ROM (PROM), Erasable PROM (EPROM), Electrically Erasable PROM (EEPROM), or flash memory. The volatile memory can be Random Access Memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double DataRate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synch link DRAM (SLDRAM), and Direct Rambus RAM (DR RAM).

[0127] In some embodiments of the present invention, the computer program can be divided into one or more modules, which are stored in the memory 111 and executed by the processor 112 to complete the method provided by the present invention. The one or more modules can be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the controller.

[0128] As Figure 7 shown, the electronic device 110 may further include: A transceiver 113, which can be connected to the processor 112 or the memory 111.

[0129] Among them, the processor 112 can control the transceiver 113 to communicate with other devices. Specifically, it can send information or data to other devices, or receive information or data sent by other devices. The transceiver 113 can include a transmitter and a receiver. The transceiver 113 may further include an antenna, and the number of antennas can be one or more.

[0130] It should be understood that the various components in the electronic device are connected through a bus system. Among them, the bus system includes, in addition to the data bus, a power bus, a control bus, and a status signal bus.

[0131] The present invention also provides a computer storage medium, on which a computer program is stored. When the computer program is executed by a computer, the computer can execute the methods in the above method embodiments. Or rather, an embodiment of the present invention also provides a computer program product containing instructions. When the instructions are executed by a computer, the computer executes the methods in the above method embodiments.

[0132] When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from a website, computer, server, or data center to another website, computer, server, or data center in a wired manner (such as coaxial cable, optical fiber, Digital Subscriber Line (DSL)) or a wireless manner (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that the computer can access, or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a Digital Video Disc (DVD)), or a semiconductor medium (such as a Solid State Disk (SSD)), etc.

[0133] Those of ordinary skill in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0134] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or modules can be in electrical, mechanical, or other forms.

[0135] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical modules, that is, they can be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. For example, in each embodiment of the present application, the various functional modules can be integrated in a processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0136] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for processing user speech data, characterized in that: include: Based on the speech data input by the target user to the target agent, generating user intention data corresponding to the speech data; The user intention data includes at least one user intention; According to feedback information of the user using the target agent on the user intention, the weight corresponding to the user intention is adjusted to obtain the target weight corresponding to the user intention; The feedback information includes one of indicating that the user intention is incorrect, indicating that the user intention is correct, and not adopting the user intention; Based on the target weight corresponding to the user intention, determining the target intention corresponding to the speech data in the user intention data; The target intent is processed by calling an API interface according to the parameter category information of the target intent to obtain a processing result corresponding to the target intent.

2. The method according to claim 1, characterized in that The step of adjusting the weight corresponding to the user intention according to feedback information of the user using the target agent on the user intention to obtain the target weight corresponding to the user intention includes: If the feedback information corresponding to the user intention indicates that the user intention is incorrect, the weight corresponding to the user intention is quickly reduced according to a preset penalty mechanism to obtain a target weight corresponding to the user intention; the preset penalty mechanism is a rule that linearly reduces the weight corresponding to the user intention at a first preset slope; If the feedback information corresponding to the user intention indicates that the user intention is correct, the weight corresponding to the user intention is increased according to a preset reward mechanism to obtain a target weight corresponding to the user intention; the preset reward mechanism is a rule that linearly increases the weight corresponding to the user intention at a second preset slope; If the feedback information corresponding to the user intention is that the user intention is not adopted, then when the number of times that the user using the target intelligent agent does not adopt the user intention reaches a first threshold, the weight corresponding to the user intention is slowly reduced according to a preset corruption mechanism to obtain a target weight corresponding to the user intention; the preset corruption mechanism is a rule that slowly reduces the weight corresponding to the user intention according to a preset logarithmic function.

3. The method according to claim 1, characterized in that After adjusting the weight corresponding to the user intention according to the feedback information of the user using the target agent on the user intention to obtain the target weight corresponding to the user intention, the method further includes: When the number of users using the target agent is greater than a second threshold, calculating a user weight corresponding to each user using the target agent according to the number of users using the target agent; The user weight and the target weight are merged to obtain a final target weight corresponding to the user intention.

4. The method according to claim 1, characterized in that: The determining, based on the target weight corresponding to the user intention, the target intention corresponding to the speech data in the user intention data includes: For each of the user intentions, calculating the similarity between the user intention and the speech data according to the target weight corresponding to the user intention; The target intention is determined in the user intention data according to a similarity calculation result between the user intention and the speech data.

5. The method according to claim 1, characterized in that The parameter category information includes parameter categories and score information corresponding to the parameter categories; the calling of the API interface according to the parameter category information of the target intent to process the target intent and obtain the processing result of the target intent includes: When the score information corresponding to the parameter category of the target intent is less than a third threshold, the parameter category and the score information corresponding to the parameter category are input into a constructed category recommendation structure to obtain a parameter recommendation category corresponding to the target intent; the category recommendation structure is used to recommend the parameter category of the target intent according to the parameter category information of the target intent, and the category recommendation structure is constructed by inputting the parameter category corresponding to the user intent, the score information corresponding to the parameter category, and the historical parameter category of the user's feedback on the parameter category into a first Markov chain; Uploading the parameter category and the parameter recommendation category corresponding to the target intent to an interactive interface, so that the target user can select the parameter category of the target intent or the parameter recommendation category in the interactive interface; If the target user selects the parameter category of the target intent, an API interface corresponding to the parameter category of the target intent is called to process the target intent; If the target user selects the parameter recommendation category, the API interface corresponding to the parameter recommendation category is called to process the target intention.

6. The method according to claim 5, characterized in that The category recommendation structure is constructed by the following method: Inputting the parameter category of the user's intention, the score information corresponding to the parameter category, and the historical parameter category information corresponding to the parameter category into the first Markov chain to obtain an initial category recommendation structure; the first Markov chain includes at least one first recommendation path, the first recommendation path includes a first score range, a parameter recommendation category corresponding to the first score range, and selection probability information corresponding to the parameter recommendation category; determining, according to the number of parameter recommendation categories recommended by the first recommendation path, whether the first recommendation path needs to be split; If the first recommendation path needs to be split, calculating the original information entropy corresponding to the initial category recommendation structure according to the selection probability information corresponding to each of the parameter recommendation categories recommended by the initial category recommendation structure; For the first recommended path that needs to be split, a first score range corresponding to the first recommended path and a parameter recommendation category corresponding to the first score range are divided according to a preset score to obtain a category recommendation structure to be determined; The category recommendation structure to be determined includes the first recommendation path that does not need to be split, and the second recommendation path that is split according to the preset score from the first recommendation path that needs to be split; the second recommendation path includes a second score range, a parameter recommendation subcategory corresponding to the second score range, and selection probability information corresponding to the parameter recommendation subcategory; Calculating the current information entropy corresponding to the category recommendation structure to be determined according to the selection probability information corresponding to the parameter recommendation category and the parameter recommendation subcategory recommended by the category recommendation structure to be determined; According to the original information entropy and the current information entropy, determining whether the category recommendation structure to be determined is a final category recommendation structure; If not, the first score range corresponding to the first recommended path is divided according to the new preset score, and the above steps are repeatedly performed until it is determined that the category recommendation structure to be determined is the final category recommendation structure.

7. The method according to any one of claims 1 to 6, characterized in that: The calling of the API interface according to the parameter category information of the target intent to process the target intent and obtain a processing result corresponding to the target intent includes: Input the parameter category information of the target intent into the constructed parameter value recommendation structure to obtain a recommended parameter value set; the recommended parameter value set includes at least two sets of recommended parameter value sets corresponding to the parameter category information; the parameter value recommendation structure is used to recommend the parameter value set of the target intent according to the parameter category information of the target intent; the parameter value recommendation structure is obtained by inputting the parameter category information corresponding to the user intent and the historical parameter value of the user's feedback on the parameter category information into the second Markov chain; Uploading the recommended parameter value set to an interactive interface, so that the target user can select a target parameter value set corresponding to the parameter category information from the recommended parameter value set in the interactive interface; According to the target parameter value set corresponding to the parameter category information, the API interface is called to process the target intent.

8. A user speech data processing device, characterized in that: include: A user intention generation module, for generating user intention data corresponding to the speech data based on the speech data input by the target user to the target intelligent agent; The user intention data includes at least one user intention; A weight adjustment module, used to adjust the weight corresponding to the user intention according to the feedback information of the user who uses the target agent on the user intention, so as to obtain the target weight corresponding to the user intention; The feedback information includes information indicating that the user intention is incorrect, information indicating that the user intention is correct, and information indicating that the user intention is not adopted; a target intention determination module, which determines the target intention corresponding to the speech data in the user intention data based on the target weight corresponding to the user intention; A processing module is used to call an API interface to process the target intent according to the parameter category information of the target intent, and obtain a processing result corresponding to the target intent.

9. An electronic device, characterized in that: include: A processor and a memory, the memory being used to store a computer program, and the processor being used to call and run the computer program stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein the computer program enables a computer to execute the method according to any one of claims 1 to 7.

Images