Information processing method and device of interaction model, vehicle and storage medium

Abstract

The invention discloses an information processing method and device of an interaction model, a vehicle and a storage medium, and relates to the technical field of information interaction, and the method comprises the steps: obtaining multi-source heterogeneous information corresponding to a current interaction task; performing scene level matching on the current interaction task according to the multi-source heterogeneous information to obtain a current task scene; determining a target context compression strategy corresponding to the current interaction task based on the current task scene; and guiding the interaction model to compress the multi-source heterogeneous information through the target context compression strategy to generate target context information corresponding to the current interaction task. According to the method, the target context compression strategy adaptive to the current interaction task is dynamically selected through the current task scene determined after task scene grading based on the multi-source heterogeneous information, and then context adaptive compression is performed based on the target context compression strategy, so that the problem of unreasonable resource allocation caused by an existing unified compression scheme is avoided, and the user experience is improved. And the information processing efficiency of the interaction model is improved.

Description

Technical Field

[0001] This application relates to the field of information interaction technology, and in particular to an information processing method, apparatus, vehicle and storage medium for an interaction model. Background Technology

[0002] With the rapid development of intelligent connected vehicles, intelligent cockpit systems have gradually evolved from single infotainment platforms into complex environments that integrate multimodal interactions. Modern intelligent cockpits integrate multiple interaction methods such as voice, vision, and touch, and the interaction model needs to simultaneously respond to diverse task requirements ranging from basic vehicle control to cross-domain collaboration.

[0003] However, in existing intelligent cockpit systems, the context information processing mechanism of the interaction model mostly adopts a uniform strategy, which has problems such as unreasonable resource allocation and low processing efficiency. Summary of the Invention

[0004] The main objective of this application is to provide an information processing method, apparatus, vehicle, and storage medium for interactive models, aiming to solve the technical problems of unreasonable resource allocation and low processing efficiency in existing interactive model context information processing schemes.

[0005] To achieve the above objectives, this application proposes an information processing method for an interaction model, the method comprising: Obtain multi-source heterogeneous information corresponding to the current interactive task; Based on the multi-source heterogeneous information, the current interactive task is matched at the scene level to obtain the current task scene; Determine the target context compression strategy corresponding to the current interactive task based on the current task scenario; The target context compression strategy guides the interaction model to compress the multi-source heterogeneous information and generate the target context information corresponding to the current interaction task.

[0006] In one embodiment, the step of determining the target context compression strategy corresponding to the current interaction task based on the current task scenario includes: Obtain the task classification confidence level corresponding to the current interactive task; Based on the task classification confidence level and the current task scenario, the target context compression strategy corresponding to the current interactive task is dynamically determined through a dual-path selection mechanism.

[0007] In one embodiment, the step of dynamically determining the target context compression strategy corresponding to the current interaction task based on the task classification confidence level and the current task scenario through a dual-path selection mechanism includes: Obtain the preset threshold; If the confidence level of the task classification is greater than or equal to the preset confidence threshold, the preset compression strategy corresponding to the current task scenario will be used as the target context compression strategy corresponding to the current interactive task.

[0008] In one embodiment, the current task scenario includes: a basic operation scenario, a contextual service scenario, and / or a cross-domain collaboration scenario; The preset compression strategy corresponding to the basic operation scenario is a key information extraction strategy, which is a compression strategy for extracting preset key information contained in multi-source heterogeneous information. The preset compression strategy corresponding to the scenario service is a visual compression strategy, which is a compression strategy that converts the multi-source heterogeneous information into an image to be analyzed and performs visual language analysis on the image to be analyzed. The preset compression strategy corresponding to the cross-domain collaborative scenario is a summary-style text compression strategy, which is a compression strategy that extracts the core semantic information of the multi-source heterogeneous information.

[0009] In one embodiment, the step of dynamically determining the target context compression strategy corresponding to the current interaction task based on the task classification confidence level and the current task scenario through a dual-path selection mechanism further includes: Obtain the preset threshold; When the confidence level of the task classification is less than the preset confidence threshold, the degree of dimension matching is obtained based on the matching between the preset compression strategy and the multi-dimensional scoring indicators. The multi-dimensional scoring indicators include at least one of compression efficiency, information fidelity and resource adaptability. Based on the degree of dimensional matching, the comprehensive compression score corresponding to the preset compression strategy is determined, and the preset compression strategy with the highest comprehensive compression score is taken as the target context compression strategy corresponding to the current interactive task.

[0010] In one embodiment, after obtaining the multi-source heterogeneous information corresponding to the current interaction task, the method further includes: When the multi-source heterogeneous information triggers a preset edge scenario, the context compression strategy corresponding to the preset edge scenario is used as the target context compression strategy corresponding to the current interaction task.

[0011] In one embodiment, the step of performing scene-level matching on the current interactive task based on the multi-source heterogeneous information to obtain the current task scene includes: A comprehensive semantic analysis is performed on the multi-source heterogeneous information to obtain multi-dimensional task information corresponding to the current interaction task. The multi-dimensional task information includes scene information and / or operation boundaries. The multidimensional task information is mapped to a preset multidimensional hierarchical index to obtain a task feature vector; The current task scenario is determined based on the degree of matching between the preset scenario classification standard and the task feature vector.

[0012] Furthermore, to achieve the above objectives, this application also proposes an information processing device for an interactive model, which includes: The information acquisition module is used to acquire multi-source heterogeneous information corresponding to the current interactive task; The scene analysis module is used to perform scene-level matching on the current interactive task based on the multi-source heterogeneous information to obtain the current task scene; The strategy selection module is used to determine the target context compression strategy corresponding to the current interaction task based on the current task scenario. The information processing control module is used to guide the interaction model to compress the multi-source heterogeneous information through the target context compression strategy, and generate the target context information corresponding to the current interaction task.

[0013] In addition, to achieve the above objectives, this application also proposes a vehicle, which includes: a memory, a processor, and an information processing program for an interactive model stored in the memory and executable on the processor, the information processing program for the interactive model being configured to implement the steps of the information processing method for the interactive model as described above.

[0014] In addition, to achieve the above objectives, this application also provides a storage medium storing a program for implementing an information processing method of an interaction model, wherein the program for implementing the information processing method of the interaction model is executed by a processor to implement the steps of the information processing method of the interaction model as described above.

[0015] This application provides an information processing method, apparatus, vehicle, and storage medium for an interaction model. The method includes: acquiring multi-source heterogeneous information corresponding to the current interaction task; performing scene-level matching on the current interaction task based on the multi-source heterogeneous information to obtain the current task scene; determining a target context compression strategy corresponding to the current interaction task based on the current task scene; and guiding the interaction model to compress the multi-source heterogeneous information through the target context compression strategy to generate target context information corresponding to the current interaction task.

[0016] This application proposes an adaptive context information hierarchical processing scheme based on task scenarios. Specifically, this application hierarchically classifies task scenarios based on multi-source heterogeneous information to achieve differentiated identification of different tasks. Then, it dynamically selects a target context compression strategy that is suitable for the current task scenario. Finally, it performs context adaptive compression of the multi-source heterogeneous information based on the target context compression strategy. This avoids the problem of unreasonable resource allocation caused by existing unified compression schemes and improves the context information processing efficiency of the interaction model. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the first embodiment of the information processing method for the interactive model of this application; Figure 2 This is a flowchart illustrating the second embodiment of the information processing method for the interactive model of this application; Figure 3 This is a simplified schematic diagram of the information processing method of the interaction model in this application; Figure 4 This is a schematic diagram of the module structure of the information processing device for the interactive model in an embodiment of this application; Figure 5 This is a schematic diagram of the device structure of the hardware operating environment involved in the information processing method of the interaction model in the embodiments of this application.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0022] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0023] The main solution of this application is: to obtain multi-source heterogeneous information corresponding to the current interaction task; to perform scene-level matching on the current interaction task based on the multi-source heterogeneous information to obtain the current task scene; to determine the target context compression strategy corresponding to the current interaction task based on the current task scene; and to guide the interaction model to compress the multi-source heterogeneous information through the target context compression strategy to generate the target context information corresponding to the current interaction task.

[0024] Existing intelligent cockpit interaction models employ a unified context information processing mechanism when handling diverse tasks, which can easily lead to problems such as low processing efficiency and unreasonable resource allocation.

[0025] To address the aforementioned issues, this application proposes a hierarchical and adaptive context information processing framework. The specific processing flow is as follows: First, comprehensive multi-source heterogeneous information corresponding to the current interaction task is collected. Then, the task is classified into scenarios based on this multi-source heterogeneous information, achieving differentiated recognition of different task scenarios. Next, the most suitable compression strategy for the current interaction task is dynamically selected based on the task scenario recognition results. That is, the target context compression strategy guides the interaction model (such as a large language model) to efficiently process the multi-source heterogeneous information. Therefore, this application can achieve refined and differentiated context processing through task classification and strategy matching, thereby avoiding resource waste caused by using complex strategies for simple tasks and preventing the loss of key information due to over-compression in complex tasks. This also avoids the unreasonable resource allocation problems caused by existing unified compression schemes, significantly improving the overall efficiency of context information processing in the interaction model and enhancing system resource utilization.

[0026] It should be noted that the execution subject in this embodiment can be an intelligent cockpit controller connected to the interaction model, or a computing service device equipped with an interaction model and possessing data processing, network communication, and program execution functions, such as a vehicle, or an information processing device capable of implementing the aforementioned functions. This embodiment does not specifically limit this. The interaction model can be applied to the intelligent cockpit system and is an intelligent processing model capable of receiving multimodal user input, processing contextual information, executing task decisions, and outputting responses. Its core function is to realize user-cockpit interaction and task implementation, such as completing navigation and air conditioning control based on voice commands. The information processing method proposed in this embodiment can be a series of standardized operations for collecting, classifying, compressing, and processing the contextual information of intelligent cockpit tasks. The core objective is to optimize processing efficiency and improve task response speed. The following uses an intelligent cockpit controller (hereinafter referred to as the controller) as the execution subject as an example to describe this embodiment and the following embodiments.

[0027] Based on this, embodiments of this application provide an information processing method for an interaction model, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the information processing method for the interactive model of this application.

[0028] In this embodiment, the information processing method for the interaction model includes steps S10 to S40: Step S10: Obtain multi-source heterogeneous information corresponding to the current interactive task; Step S20: Perform scene-level matching on the current interactive task based on the multi-source heterogeneous information to obtain the current task scene; It is easy to understand that the aforementioned current interaction task can represent the specific interaction request initiated by the user to the intelligent cockpit system, which may include voice interaction tasks, cockpit interaction tasks, and / or central control screen interaction tasks. The aforementioned multi-source heterogeneous information can be a collection of information from multiple sources and of multiple types related to the current interaction task, in order to conduct as comprehensive an information analysis of the current interaction task as possible, and subsequently make accurate task scenario judgments based on the multi-source heterogeneous information.

[0029] For example, in this embodiment, the multi-source heterogeneous information may include user input (e.g., voice commands, text commands, and touch operation commands input through the human-machine interface), historical interaction records (e.g., historical dialogue texts and historical operation logs obtained from the session storage unit), vehicle status data (including but not limited to operating status (vehicle speed, engine / motor speed, gear), environmental adaptation status (tire pressure, tire temperature, in-vehicle temperature), resource status (remaining range / fuel, remaining battery power, window / door lock status), and location context (latitude and longitude, altitude, whether in a tunnel / underground parking garage) collected through the vehicle bus), and a set of available atomic capabilities (a list of functional units that the current system can execute, consisting of independent functional modules (i.e., atomic functions) of the cockpit, such as navigation, music playback, and web (web page) retrieval).

[0030] At this point, the controller can analyze the current interactive task based on a preset hierarchical system and multi-source heterogeneous information, and determine the preset task scenario to which it belongs, that is, to perform the above-mentioned task scenario matching. The core of this process is to achieve accurate task classification through semantic parsing and scenario feature matching, and the preset task scenario that matches the current interactive task is the above-mentioned current task scenario.

[0031] Therefore, in order to clarify the standardized process of task scenario matching, in a feasible implementation, step S20 in this embodiment may include steps A1 to A3: Step A1: Perform comprehensive semantic analysis on the multi-source heterogeneous information to obtain multi-dimensional task information corresponding to the current interaction task. The multi-dimensional task information includes scene information and / or operation boundaries. It is important to understand that the aforementioned comprehensive semantic analysis can refer to the process by which the controller comprehensively analyzes multi-source heterogeneous information, integrates information such as user intent, historical logic, vehicle state association, and atomic capability boundaries, and forms a holistic understanding of the current task.

[0032] At this point, the multidimensional task information obtained through comprehensive semantic analysis can be the core information obtained through comprehensive semantic analysis, including scenario information determined by analyzing the semantics of the user's current query, historical dialogues, and vehicle status, i.e., the application scenario involved in the current interaction task, such as navigation and entertainment; and the operation boundary determined by analyzing the atomic capability list, i.e., the range of atomic capabilities that the current interaction task needs to call.

[0033] Step A2: Map the multi-dimensional task information to a preset multi-dimensional hierarchical index to obtain a task feature vector; Step A3: Determine the current task scenario based on the degree of matching between the preset scenario classification standard and the task feature vector.

[0034] It is easy to understand that the aforementioned preset multi-dimensional grading indicators can be the three core dimensions involved in task grading in this embodiment, namely, task complexity, real-time requirements, and interaction depth. Among them, task complexity can refer to the number of atomic capability calls, the number of cross-domain / scenario calls, and the logical depth of decision steps required to execute the current interactive task, which can be comprehensively determined based on scenario information and / or operation boundaries; real-time requirements can be the maximum permissible response delay that the user can perceive, which can be determined based on scenario information; and interaction depth can be the expected number of dialogue rounds for the current interactive task or the frequency of information exchange required to complete the current interactive task, which can be determined based on scenario information.

[0035] Accordingly, the aforementioned task feature vector can be vector data formed by mapping multi-dimensional task information to preset grading indicators. In this case, the controller can map the multi-dimensional task information to three dimensions corresponding to the preset multi-dimensional grading indicators: task complexity, real-time requirements, and interaction depth, forming a task feature vector. Specifically, regarding task complexity, the level (low / medium / high) can be determined based on the number of atomic capabilities to be invoked and the number of domains involved; regarding real-time requirements, the level (high / medium / low) can be determined based on the user's expected response speed; and regarding interaction depth, the level (shallow / medium / deep) can be determined based on the estimated number of dialogue turns. Finally, the mapping results of the three dimensions are combined to form the task feature vector.

[0036] For example, after comprehensive semantic analysis, the controller infers that the scenario information of the current interaction task can be travel, energy management and service booking. The operation boundary requires the linkage of three capabilities: "navigation", "POI (Point of Interest) search" and "voice interaction" (corresponding to high task complexity); the travel scenario needs to meet the user's need to get an answer before the next turn (corresponding to medium real-time); the service booking scenario may need to confirm the gas station brand and lunch flavor (corresponding to medium interaction depth). The corresponding generated task feature vector can be (task complexity: high, real-time: medium, interaction depth: medium).

[0037] It's important to understand that the aforementioned preset scenario classification criteria can be critical value standards for distinguishing different task scenarios (consistent with the classification system mentioned earlier). The aforementioned matching degree can be the degree of fit between the task feature vector and the preset scenario classification criteria, which is the core basis for determining the task scenario. That is, the controller can then output the task scenario with the highest matching degree as the current task scenario based on the degree of fit between the task feature vector and the preset scenario classification criteria.

[0038] For example, suppose the pre-defined classification scenarios include scenario A and scenario B. The pre-defined scenario classification criteria for scenario A are (task complexity: low / medium, real-time performance: low / medium, interaction depth: low), and the pre-defined scenario classification criteria for scenario B are (task complexity: high, real-time performance: medium / high, interaction depth: medium / high). Then, when the task feature vector corresponding to the current interaction task is (task complexity: low, real-time performance: medium, interaction depth: medium), the matching degree with scenario A has three dimensions, and the matching degree with scenario B has one dimension. Therefore, the current task scenario corresponding to the current interaction task can be selected as scenario A.

[0039] In this implementation, a standardized three-step classification process (comprehensive semantic analysis → feature vector mapping → matching degree judgment) can be used. A standardized task classification system can be constructed based on dimensions such as task complexity, real-time requirements, and interaction depth to ensure the accuracy and consistency of task scenario classification results. This provides a solid foundation for subsequent strategy selection based on the current task scenario with high reliability, and ensures the effectiveness of the overall information processing process.

[0040] Step S30: Determine the target context compression strategy corresponding to the current interaction task based on the current task scenario; Step S40: Guide the interaction model to compress the multi-source heterogeneous information through the target context compression strategy to generate the target context information corresponding to the current interaction task.

[0041] Understandably, the controller dynamically selects a data compression strategy that adapts to the current task scenario, while retaining the core information required by the current task scenario, to reduce the amount of data and remove redundant content. This is the aforementioned target context compression strategy, which aims to improve data processing efficiency and reduce system resource consumption.

[0042] Therefore, the controller can aggregate the selected target context compression strategy and multi-source heterogeneous information into the interaction model, enabling the interaction model to adapt to and efficiently process the complex data corresponding to the current interaction task, generating target context information. Subsequently, the interaction model can make driving decisions based on the input information and the compressed target context information, generating and executing atomic capability call sequences to complete the processing of the current interaction task.

[0043] In summary, this embodiment establishes a task scenario hierarchy based on multiple dimensions such as task complexity, real-time requirements, and interaction depth. It divides the current interactive task into different task scenarios and implements differentiated context information processing strategies for each scenario. This ensures immediate response for simple tasks while reserving sufficient context resources for complex tasks. Therefore, compared to traditional unified context processing mechanisms, this embodiment can achieve accurate identification of different task scenarios through a task scenario hierarchy, thereby optimizing resource allocation based on context compression strategies adapted to different task scenarios and improving information processing efficiency.

[0044] This embodiment provides an information processing method for an interaction model. The method includes: acquiring multi-source heterogeneous information corresponding to the current interaction task; performing comprehensive semantic analysis on the multi-source heterogeneous information to obtain multi-dimensional task information corresponding to the current interaction task, whereby the multi-dimensional task information includes scene information and / or operation boundaries; mapping the multi-dimensional task information to preset multi-dimensional hierarchical indicators to obtain task feature vectors; determining the current task scene based on the matching degree between the preset scene hierarchical standards and the task feature vectors; determining a target context compression strategy corresponding to the current interaction task based on the current task scene; and guiding the interaction model to compress the multi-source heterogeneous information using the target context compression strategy to generate target context information corresponding to the current interaction task.

[0045] This embodiment establishes a task scenario hierarchy system based on multiple dimensions such as task complexity, real-time requirements, and interaction depth. It divides the current interactive task into different task scenarios and implements differentiated context information processing strategies for each scenario. This ensures immediate response for simple tasks while reserving sufficient context resources for complex tasks. Therefore, compared to traditional unified context processing mechanisms, this embodiment can achieve accurate identification of different task scenarios through the task scenario hierarchy system, thereby optimizing resource allocation based on context compression strategies adapted to different task scenarios and improving information processing efficiency.

[0046] Based on the first embodiment of this application, in the second embodiment of this application, the same or similar content as the first embodiment described above can be referred to the above description, and will not be repeated hereafter.

[0047] It is understandable that existing intelligent cockpits lack a fine distinction between task characteristics and resource requirements when processing contextual information for diverse tasks, and the accuracy of information processing also needs improvement. Therefore, based on the first embodiment, please refer to... Figure 2 , Figure 2 This is a flowchart illustrating the second embodiment of the information processing method for the interactive model of this application. In this embodiment, step S30 includes steps S31 to S32: Step S31: Obtain the task classification confidence level corresponding to the current interactive task; Step S32: Based on the task classification confidence level and the current task scenario, dynamically determine the target context compression strategy corresponding to the current interactive task through a dual-path selection mechanism.

[0048] It is understandable that the aforementioned task-level confidence score can be the reliability score (0-1 range) output by the controller for the current task scenario after task classification. This task-level confidence score reflects the classification result, i.e., the credibility of the current task scenario, and can be determined by factors such as the clarity of feature matching, the completeness of input information, and the presence of ambiguity during the classification process. For example, if the task features corresponding to the current interaction task completely match a certain task scenario threshold and the input information is unambiguous, the confidence score can be 0.95; if the task features partially match multiple scenarios and the input information is highly complete but ambiguous, the confidence score may be 0.8.

[0049] The aforementioned dual-path selection mechanism refers to the controller, after determining the current task scenario based on task classification confidence, employing two different strategies to select the target context compression strategy corresponding to the current interaction task. In this embodiment, the controller can address the poor adaptability of a single strategy selection method when the classification result is uncertain by introducing task classification confidence and a dual-path selection mechanism, thereby balancing the efficiency and accuracy of strategy selection.

[0050] For example, in this embodiment, the controller can match the default compression strategy according to the task level when the task classification confidence is high; when the confidence is insufficient, it can intelligently select the most suitable compression method with the help of a large model. Therefore, in the first feasible implementation, step S32 may include steps B11 to B21: Step B11: Obtain the preset information threshold; Step B21: If the task classification confidence level is greater than or equal to the preset confidence threshold, the preset compression strategy corresponding to the current task scenario is used as the target context compression strategy corresponding to the current interactive task.

[0051] In this embodiment, the aforementioned preset confidence threshold can be a fixed critical value for judging the reliability of the classification result, with a default value of 0.9. Alternatively, the preset confidence threshold can be a dynamic critical value adjusted according to the actual scenario; for example, it can be lowered to 0.85 when there are many simple tasks and raised to 0.95 when there are many complex tasks. Therefore, this preset confidence threshold serves as the basis for switching the aforementioned dual-path selection mechanism. Correspondingly, the aforementioned preset compression strategy can be a compression strategy pre-set for different task scenarios and verified to be adapted to the characteristics of that scenario. It is the default compression scheme for each task scenario under high-confidence scenarios, i.e., when the task classification confidence is greater than or equal to the preset confidence threshold. Therefore, the controller can pre-store the mapping relationship between each task scenario and the corresponding preset compression strategy, so that when the task scenario classification result is reliable, the corresponding compression strategy can be quickly called as the target context compression strategy, simplifying the compression strategy selection process, improving the information compression response speed, and meeting the real-time requirements of basic operation tasks.

[0052] In a feasible implementation, the current task scenario in this embodiment includes: a basic operation scenario, a contextual service scenario, and / or a cross-domain collaboration scenario. The preset compression strategy corresponding to the basic operation scenario is a key information extraction strategy, which is a compression strategy for extracting preset key information contained in multi-source heterogeneous information. The preset compression strategy corresponding to the scenario service is a visual compression strategy, which is a compression strategy that converts the multi-source heterogeneous information into an image to be analyzed and performs visual language analysis on the image to be analyzed. The preset compression strategy corresponding to the cross-domain collaborative scenario is a summary-style text compression strategy, which is a compression strategy that extracts the core semantic information of the multi-source heterogeneous information.

[0053] It's important to understand that the aforementioned basic operational scenarios can be task scenarios with low complexity (e.g., calling a single atomic capability), high real-time requirements (e.g., response latency < 500 milliseconds), and shallow interaction depth (e.g., single-turn interaction). The core characteristics of this scenario are: single modality and instant response. Typical cockpit tasks may include air conditioning control, volume adjustment, and window opening / closing.

[0054] The aforementioned scenario-based service can be a scenario with medium task complexity (e.g., calling 2-3 atomic capabilities or multi-step conditional judgments in a single domain (e.g., "navigate to A, avoid highways" involving path planning and conditional filtering)), medium real-time requirements (e.g., response latency of 500 milliseconds to 3 seconds), and medium interaction depth (e.g., 2-5 rounds of interaction). The core characteristics of this scenario are: limited scenarios and multi-turn dialogues. Typical cockpit tasks may include navigation queries and music recommendations.

[0055] The aforementioned cross-domain collaborative scenarios can be those with high task complexity (e.g., calling ≥4 atomic capabilities or explicitly involving ≥2 independent scenario collaboration in serial / parallel processing), low real-time requirements (e.g., response latency >3 seconds), and deep interaction (e.g., ≥5 rounds of interaction). The core characteristics of this scenario are: multi-scenario linkage and strong logical dependencies. Typical cockpit tasks may include travel planning + restaurant reservation, navigation + refueling + charging planning, etc.

[0056] It should be noted that, in the pre-set compression strategy, the key information extraction strategy corresponding to the above-mentioned basic operation scenario can refer to the compression strategy that extracts key elements of multi-source heterogeneous information and filters redundant information to form context information in the format of "instruction-parameter pair". Among them, the key elements, that is, the above-mentioned pre-set key information, can be information that plays a decisive role in the execution of the basic operation task, and may include the operation object (air conditioner, car window), control parameters (temperature, volume), execution instructions (open, adjust), etc.

[0057] For example, in basic operation scenarios, this embodiment can pre-construct a key information database that clearly defines the key information types corresponding to different basic operation tasks (such as key information for air conditioning control tasks including the operation object, temperature, and wind speed); then, semantic parsing is performed on multi-source heterogeneous information to identify preset key information that matches the key information database, filtering out unnecessary content such as small talk and repetition; finally, the extracted preset key information is organized into "instruction-parameter pairs". For example, if a user inputs "Please set the air conditioner to 24℃", it can be compressed into "Air conditioning control - temperature 24℃".

[0058] The aforementioned visual compression strategy can be a compression scheme for contextual service scenarios. It can be used to render multi-source heterogeneous information into a visual image, i.e. the image to be analyzed, through preset rules (time order, logical relationship). Then, the image content of the image to be analyzed is parsed through a visual language model to achieve efficient context compression and understanding.

[0059] For example, in a scenario-based service, the controller can collect multi-source heterogeneous information collected in multiple rounds, and then render the multi-source heterogeneous information into an image to be analyzed in a fixed visualization format according to a preset template (such as the top layout of dialogue text, the bottom display of vehicle location icons, vehicle speed values, etc.); finally, it calls a visual language model to parse the image to be analyzed and extract the core information in the image used for task processing (such as navigation destination, road conditions, and requirements), which can achieve 3-4 times data compression.

[0060] Finally, the aforementioned summary-based text compression strategy can be a compression solution for cross-domain collaborative scenarios. In this case, the controller can remove redundant content through a summary generation model and extract the core semantics of multi-source heterogeneous information to generate target context information corresponding to the multi-source heterogeneous information. This core semantics can include key information such as task objectives, step logic, and causal relationships. For example, if a user inputs "First, navigate me to the company, then check if there are any gas stations along the way, and also book a Sichuan restaurant lunch," the compressed version can be "Navigate to the company, check for gas stations along the way, and book a Sichuan restaurant."

[0061] In a second feasible implementation, step S32 may further include steps B12 to B32: Step B12: Obtain the preset information threshold; Step B22: When the confidence level of the task classification is less than the preset confidence threshold, the degree of dimension matching is obtained according to the matching between the preset compression strategy and the multi-dimensional scoring indicators. The multi-dimensional scoring indicators include at least one of compression efficiency, information fidelity and resource adaptability. Step B32: Determine the comprehensive compression score corresponding to the preset compression strategy based on the degree of dimensional matching, and take the preset compression strategy with the highest comprehensive compression score as the target context compression strategy corresponding to the current interaction task.

[0062] Understandably, in low-to-medium confidence scenarios, i.e., when the confidence level of the task is less than the preset confidence threshold, the controller needs to adaptively select the compression scheme for each task scenario based on the actual situation. The aforementioned multi-dimensional scoring indicators involved in the selection process can be three core dimensions for evaluating the adaptability of each preset compression strategy to the current interactive task. In this embodiment, the multi-dimensional scoring indicators may include compression efficiency (i.e., data compression ratio, processing speed), information fidelity (the integrity of retaining key information), and resource adaptability (the degree of matching with the current system resource status).

[0063] Accordingly, the aforementioned dimensional matching degree can characterize the performance score of each preset compression strategy on the three scoring indicators, reflecting the adaptability of each preset compression strategy to the current interactive task and the system state in which the current interactive task is located. At this time, the aforementioned comprehensive compression score can be a comprehensive strategy evaluation score obtained by weighting calculation based on the dimensional matching degree, and then the controller can select the target context compression strategy from each preset compression strategy based on the comprehensive compression score.

[0064] For example, the controller can analyze the task characteristics, context, and resource status of the current interactive task, and select the most suitable compression strategy from the compression strategy library. First, during the compression efficiency evaluation process, the controller can estimate the compression ratio and processing speed of each strategy based on the length and structure of the context to be compressed corresponding to the current interactive task, and generate a compression efficiency score that matches the compression efficiency (the score is positively correlated with the compression ratio and negatively correlated with the processing speed). During the information fidelity assessment process, the controller can combine the current task classification level requirements and the semantic analysis of the current dialogue content to determine which types of information (such as precise parameters, intent flow, and causal chains) need to be retained to meet the current task level. Then, it can determine whether the information integrity corresponding to each compression strategy meets the task level's requirements for contextual coherence (such as cross-domain collaborative tasks requiring high fidelity) and generate an information fidelity score that matches the information fidelity. During the resource adaptability assessment process, the controller can collect real-time resource status such as CPU utilization, available memory, and GPU load through the system monitoring module, and determine whether each compression strategy is suitable for the system resource bottleneck corresponding to the real-time resource status (such as selecting key information extraction with low memory usage when memory is insufficient) or whether it will cause resource overload corresponding to the real-time resource status, and generate a resource adaptability score that matches the resource adaptability.

[0065] Then, this embodiment can set the weights of multi-dimensional scoring indicators for different cockpit missions based on their mission characteristics. For example, for missions with high real-time requirements, the weights can be set as follows: compression efficiency 0.4, information fidelity 0.3, and resource adaptability 0.3. For missions with high contextual coherence requirements, the weights can be set as follows: information fidelity 0.4, compression efficiency 0.3, and resource adaptability 0.3. When system resources are scarce, the weight of resource adaptability can be set as 0.4, and the other two items can each be set as 0.3.

[0066] Finally, the controller can perform weighted calculations on the scores and weights of each indicator to obtain a comprehensive compression score, and select the preset compression strategy with the highest comprehensive compression score as the target context compression strategy.

[0067] Therefore, when classification confidence is low, a single-dimensional evaluation cannot fully reflect strategy adaptability, which may lead to a mismatch between the strategy and task requirements or system resources, affecting processing performance or causing resource waste. This embodiment adopts a multi-strategy trade-off mechanism, which comprehensively considers task requirements and system status through multi-dimensional evaluation, improving the accuracy of strategy selection; and combines a dynamic weight adjustment mechanism to adapt to different core task requirements and system status, enhancing the flexibility of context information processing, ensuring the efficiency and accuracy of context processing, and improving system robustness.

[0068] In a third possible implementation, step C1 may be included after step S10: Step C1: When the multi-source heterogeneous information triggers a preset edge scenario, the context compression strategy corresponding to the preset edge scenario is used as the target context compression strategy corresponding to the current interaction task.

[0069] It should be understood that the aforementioned preset edge scenario can be a predefined special scenario. The scene classification compression strategy selection process corresponding to the preset edge scenario has poor adaptability. Therefore, when the controller detects multi-source heterogeneous information that triggers the preset edge scenario, it does not need to perform subsequent task scenario classification. Instead, it can directly process the multi-source heterogeneous information based on the context compression strategy corresponding to the preset edge scenario.

[0070] At this point, the triggering of a preset edge scenario can refer to the controller detecting that multi-source heterogeneous information meets the characteristic conditions of the preset edge scenario. For example, in this embodiment, the preset edge scenario may include scenarios with frequent multimodal interactions, redundant information scenarios, etc. Correspondingly, the scenario characteristics corresponding to a scenario with frequent multimodal interactions can be scenarios where the current interaction task involves two or more modalities (voice, vision, touch), and the interaction frequency is ≥3 times within a preset time window (e.g., 30 seconds). A redundant information scenario can refer to a scenario where key information is repeated ≥3 times, greeting words appear ≥5 times, or redundant information accounts for ≥60% of the total information in the historical dialogue corresponding to the current interaction task.

[0071] At this point, during the process of detecting whether multi-source heterogeneous information triggers an edge scenario, the controller can determine whether a frequent multimodal interaction scenario is triggered by statistically analyzing the number of interaction modal types and the interaction frequency within a preset time window; and determine whether a redundant information scenario is triggered by statistically analyzing the repetition frequency of key information and the occurrence frequency of greeting words, and calculating the proportion of redundant information. If the controller detects that the multi-source heterogeneous information meets the feature conditions of any edge scenario, it can determine that the edge scenario has been triggered. Then, it directly retrieves the compression strategy corresponding to the triggered edge scenario from the preset strategy library as the target context compression strategy, without needing to execute the subsequent conventional scenario matching and confidence judgment process. Specifically, when it is determined to be a frequent multimodal interaction scenario and sufficient system resources are detected, a visual compression strategy can be selected as the target context compression strategy to maintain context richness; when it is determined to be a redundant information scenario, a summary-style text compression strategy can be selected as the target context compression strategy to extract effective core semantics.

[0072] Therefore, this embodiment can cover special cases that cannot be adapted to by preset edge scenarios and corresponding strategies, thereby improving the comprehensiveness of strategy selection; and the edge scenario detection mechanism is simple and efficient, which can quickly identify and trigger the corresponding strategy, thereby improving processing efficiency.

[0073] In summary, this embodiment integrates multiple compression technologies and dynamically adjusts the context compression strategy based on task characteristics and system resource status, significantly improving information processing efficiency. The controller can dynamically determine the target context compression strategy based on task-level confidence and the current task scenario through a dual-path selection mechanism. The introduction of task-level confidence provides a more comprehensive basis for selecting the target context strategy, allowing for an objective assessment of the reliability of the classification results. The dual-path mechanism balances processing efficiency (direct matching for high confidence, eliminating the need for complex calculations) and accuracy (dynamic optimization for low confidence), effectively adapting to different classification reliability scenarios and improving information processing precision. Furthermore, this embodiment also sets a one-step compression strategy determination method corresponding to preset edge scenarios, further enhancing processing efficiency. Therefore, this embodiment can further improve the adaptability of the compression strategy, avoiding poor processing results due to classification ambiguity and ensuring the accuracy and efficiency of compression task execution.

[0074] This embodiment discloses the following steps: obtaining the task level confidence score corresponding to the current interactive task; obtaining a preset confidence threshold; and, when the task level confidence score is greater than or equal to the preset confidence threshold, using a preset compression strategy corresponding to the current task scenario as the target context compression strategy for the current interactive task. The current task scenario includes: a basic operation scenario, a context service scenario, and / or a cross-domain collaboration scenario. The preset compression strategy for the basic operation scenario is a key information extraction strategy, which is a compression strategy that extracts preset key information contained in multi-source heterogeneous information. The preset compression strategy for the context service scenario is a visual compression strategy, which is a compression strategy that converts multi-source heterogeneous information into an image to be analyzed and performs visual language analysis on the image to be analyzed. The preset compression strategy for the cross-domain collaboration scenario is a summary-style text compression strategy, which is a compression strategy that extracts the core semantic information of multi-source heterogeneous information. This embodiment also discloses that when the task classification confidence level is less than a preset confidence threshold, the dimensional matching degree is obtained based on the matching between the preset compression strategy and multi-dimensional scoring indicators. The multi-dimensional scoring indicators include compression efficiency, information fidelity, and resource adaptability. A comprehensive compression score corresponding to the preset compression strategy is determined based on the dimensional matching degree, and the preset compression strategy with the highest comprehensive compression score is used as the target context compression strategy for the current interactive task. This embodiment also discloses that when multi-source heterogeneous information triggers a preset edge scenario, the context compression strategy corresponding to the preset edge scenario is used as the target context compression strategy for the current interactive task.

[0075] Therefore, this embodiment can directly adopt the preset compression strategy corresponding to the current task scenario as the target context compression strategy in high-confidence paths. In low-confidence paths, it can dynamically select the optimal solution as the target context compression strategy by analyzing the compression efficiency, information fidelity and resource adaptability of the preset compression strategy. When the preset edge scenario is triggered, the corresponding compression strategy can be quickly identified and triggered. This improves information processing efficiency while ensuring the adaptability between the target context compression strategy and the current interactive task, thereby improving information processing accuracy.

[0076] For example, to help understand the technical concept or principle of the information processing method of the interaction model after combining this embodiment with the above-described Embodiments 1 and 2, please refer to Figure 3 , Figure 3 The following is a simplified schematic diagram of the information processing method of the interaction model in this application: The core objective of this application is to provide an information processing solution for an intelligent cockpit interaction model. By collecting multi-source heterogeneous information, matching task scenarios, and determining an appropriate compression strategy, it solves the problems of information redundancy and low efficiency caused by existing unified processing mechanisms, and achieves accurate and efficient processing of contextual information.

[0077] like Figure 3 As shown, the information processing procedure of the interaction model in this application can be as follows: Step 1: Perform scene-level matching based on multi-source heterogeneous information to obtain the current task scene: The system acquires multi-source heterogeneous information corresponding to the current interactive task. Then, based on a pre-built task classification system, it analyzes the multi-source heterogeneous information from three dimensions: task complexity, real-time requirements, and interaction depth to obtain corresponding quantitative indicators, namely task feature vectors. Finally, it compares the task feature vectors with the preset scenario classification standards to determine the current task scenario, which may correspond to basic operation scenario, scenario service scenario, or cross-domain collaboration scenario.

[0078] Or, such as Figure 3 As shown, preset edge scenarios are triggered directly based on multi-source heterogeneous information.

[0079] Step 2: Determine the target context compression strategy based on the current task scenario: Based on the determined current task scenario and task level confidence, the target context compression strategy is adaptively selected from the preset compression strategies.

[0080] At this point, when the confidence level of the task classification is greater than or equal to the preset confidence threshold, the basic operation scenario can select the key information extraction strategy, the scenario service scenario can select the visual compression strategy, and the cross-domain collaboration scenario can select the summary text compression strategy. When the confidence level of the task classification is less than the preset confidence threshold, the target context compression strategy is adaptively selected from the preset compression strategies based on the matching situation between the preset compression strategy and the multi-dimensional scoring indicators corresponding to the current interactive task.

[0081] When a preset edge scenario is triggered, the context compression strategy corresponding to the preset edge scenario is directly used as the target context compression strategy for the current interaction task.

[0082] Step 3: Guide the interaction model to compress contextual information using the target context compression strategy: The controller guides the interaction model to adaptively process multi-source heterogeneous information based on the selected target context compression strategy.

[0083] If the target context compression strategy is a key information extraction strategy, then the key elements of multi-source heterogeneous information (operation object, parameters, time, etc.) are identified, unnecessary content such as small talk and repetition is filtered out, and the information is compressed into the form of "instruction-parameter pair" (such as "air conditioning control-temperature 24℃"). If a visual compression strategy is adopted, the information contained in the multi-source heterogeneous information, such as multi-turn dialogues and vehicle status, will be rendered into a visual image (e.g., dialogue text arranged in chronological order, with vehicle location and status icons superimposed), and analyzed and processed through a visual language model to achieve 3-4 times data compression. If a summary-based text compression strategy is adopted, the core semantics of multi-source heterogeneous information are extracted through a summary generation model, key information such as task objectives and step logic are retained, and redundant data is removed.

[0084] For example, in a specific implementation, suppose the complex query input by the user corresponding to the current interaction task is: "First, help me navigate to the company, then check if there are any gas stations along the way, and also order lunch."

[0085] At this point, by first analyzing the task characteristics of the current interaction task through comprehensive semantic analysis, we can see that the corresponding scenario information can be travel, energy management and service booking. The operation boundary requires the linkage of three capabilities: "navigation", "POI (Point of Interest) search" and "voice interaction" (corresponding to high task complexity). The travel scenario needs to meet the user's need to get an answer before the next turn (corresponding to medium real-time performance). The service booking scenario may require multiple rounds of dialogue to confirm the gas station brand and lunch flavor (corresponding to medium interaction depth). Therefore, the corresponding generated task feature vector can be (task complexity: high, real-time performance: medium, interaction depth: medium).

[0086] Therefore, the controller can determine that the current task scenario is a cross-domain collaborative scenario based on the preset scenario classification standard and task feature vector. The controller can output the task classification result as: cross-domain collaborative scenario, task classification confidence level 0.85.

[0087] Then, since the task's classification confidence level is lower than the preset confidence threshold of 0.9, adaptive selection is initiated. At this point, during the strategy selection process, while the visual compression strategy can handle multi-turn dialogues, high GPU load may lead to latency; the summary-based text compression strategy preserves causal chains (such as the logic of "navigation-gas station-lunch") and has sufficient CPU resources; the key information extraction strategy may lose details and is unsuitable for complex tasks. Therefore, the summary-based text compression strategy ultimately achieves a higher overall compression score and can be selected as the target context compression strategy to ensure contextual coherence while avoiding resource conflicts.

[0088] Alternatively, in a specific implementation, suppose the complex query input by the user for the current interactive task is: "Navigate to the Capital Airport, the traffic is a bit congested, play some upbeat music for me."

[0089] At this point, by first analyzing the task characteristics of the current interaction task through comprehensive semantic analysis, we can see that the corresponding scenario information is a single travel scenario, and the operation boundary can be two atomic capabilities: navigation and music playback (corresponding to medium task complexity); the travel scenario needs to meet the user's requirement that music should start playing shortly after navigation is initiated (corresponding to medium real-time performance); the travel scenario may require a round of information confirmation, such as "What type of refreshing music would you like to listen to?" (corresponding to medium interaction depth). Therefore, the corresponding generated task feature vector can be (task complexity: medium, real-time performance: medium, interaction depth: medium).

[0090] Therefore, the controller can determine the current task scenario as a context service scenario based on the preset scenario classification standard and task feature vector. The controller can output the task classification result as: context service scenario, task classification confidence level 0.88.

[0091] Since the task's confidence level of 0.88 is still lower than the preset confidence threshold of 0.9, adaptive selection is initiated accordingly. At this point, during the strategy selection process, the key information extraction strategy may lose the contextual association between "traffic jam" and "energizing," leading to inaccurate music recommendations (corresponding to a low to medium information fidelity score). The summary-style text compression strategy can preserve specific contexts, but it is somewhat "heavy" for this moderately complex task (corresponding to a medium compression efficiency score). The visual compression strategy can integrate multi-turn dialogues (possible historical traffic queries and current instructions) and vehicle location information into a "contextual graph," suitable for understanding and associating "navigation-traffic-music preferences," and current resources allow for it (corresponding to a high information fidelity score and a high resource adaptability score). Therefore, the visual compression strategy ultimately achieves a higher overall compression score, and can be selected as the target context compression strategy to best maintain the richness of context between services.

[0092] In summary, to address the issues arising from the current unified context processing mechanism in intelligent cockpits, which fails to consider the resource requirements and processing characteristics of different tasks, leading to information redundancy, mismatched compression methods, and an inability to simultaneously meet the immediate response requirements of simple tasks and the context integrity requirements of complex tasks, this application addresses these problems. It achieves differentiated task identification in different scenarios through comprehensive collection of multi-source heterogeneous information and precise scene matching. Furthermore, based on a scene-matched and adapted compression strategy, it avoids information waste or loss of critical information caused by the unified context compression mechanism, improving information processing efficiency while ensuring processing accuracy. Simultaneously, the adapted compression strategy effectively and accurately reduces the amount of context data, lowering system memory usage and CPU / GPU load, avoiding the unreasonable resource allocation problems caused by existing unified compression schemes, and ensuring the stable operation of the intelligent cockpit system.

[0093] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the information processing method of the interaction model of this application. Any simple transformations based on this technical concept are within the protection scope of this application.

[0094] This application also provides an information processing device for an interactive model; please refer to [reference needed]. Figure 4 , Figure 4 This is a schematic diagram of the module structure of the information processing device for the interaction model in an embodiment of this application. In this embodiment, the information processing device for the interaction model includes: Information acquisition module T1 is used to acquire multi-source heterogeneous information corresponding to the current interactive task; Scene analysis module T2 is used to perform scene level matching on the current interactive task based on the multi-source heterogeneous information to obtain the current task scene; The strategy selection module T3 is used to determine the target context compression strategy corresponding to the current interaction task based on the current task scenario. The information processing control module T4 is used to guide the interaction model to compress the multi-source heterogeneous information through the target context compression strategy, and generate the target context information corresponding to the current interaction task.

[0095] As one possible implementation, in this embodiment, the strategy selection module T3 is further used to obtain the task level confidence level corresponding to the current interaction task; based on the task level confidence level and the current task scenario, the target context compression strategy corresponding to the current interaction task is dynamically determined through a dual-path selection mechanism.

[0096] As one possible implementation, in this embodiment, the strategy selection module T3 is further configured to obtain a preset confidence threshold; when the task classification confidence level is greater than or equal to the preset confidence threshold, the preset compression strategy corresponding to the current task scenario is used as the target context compression strategy corresponding to the current interactive task.

[0097] As one possible implementation, in this embodiment, the current task scenario includes: a basic operation scenario, a contextual service scenario, and / or a cross-domain collaboration scenario; the preset compression strategy corresponding to the basic operation scenario is a key information extraction strategy, which is a compression strategy that extracts preset key information contained in multi-source heterogeneous information; the preset compression strategy corresponding to the contextual service scenario is a visual compression strategy, which is a compression strategy that converts the multi-source heterogeneous information into an image to be analyzed and performs visual language analysis on the image to be analyzed; the preset compression strategy corresponding to the cross-domain collaboration scenario is a summary text compression strategy, which is a compression strategy that extracts the core semantic information of the multi-source heterogeneous information.

[0098] As one possible implementation, in this embodiment, the strategy selection module T3 is further configured to obtain a preset confidence threshold; when the task classification confidence level is less than the preset confidence threshold, the dimension matching degree is obtained based on the matching between the preset compression strategy and the multi-dimensional scoring indicators, wherein the multi-dimensional scoring indicators include at least one of compression efficiency, information fidelity, and resource adaptability; the comprehensive compression score corresponding to the preset compression strategy is determined based on the dimension matching degree, and the preset compression strategy with the highest comprehensive compression score is taken as the target context compression strategy corresponding to the current interactive task.

[0099] As one possible implementation, in this embodiment, the strategy selection module T3 is further configured to, when the multi-source heterogeneous information triggers a preset edge scenario, use the context compression strategy corresponding to the preset edge scenario as the target context compression strategy corresponding to the current interactive task.

[0100] As one possible implementation, in this embodiment, the scene analysis module T1 is further used to perform comprehensive semantic analysis on the multi-source heterogeneous information to obtain multi-dimensional task information corresponding to the current interactive task. The multi-dimensional task information includes scene information and / or operation boundaries. The multi-dimensional task information is mapped to a preset multi-dimensional hierarchical index to obtain a task feature vector. Based on the degree of matching between the preset scene hierarchical standard and the task feature vector, the current task scene is determined.

[0101] The information processing apparatus for interactive models provided in this application, employing the information processing method for interactive models in the above embodiments, can solve the technical problems of unreasonable resource allocation and low processing efficiency in existing context information processing schemes for interactive models. Compared with the prior art, the beneficial effects of the information processing apparatus for interactive models provided in this application are the same as those of the information processing method for interactive models provided in the above embodiments, and other technical features in the information processing apparatus for interactive models are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0102] This application provides a vehicle, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the information processing method of the interaction model in Embodiment 1 above.

[0103] In this embodiment, the vehicle also integrates an intelligent cockpit system and is equipped with an interactive model of an intelligent connected vehicle (either a fuel vehicle or a new energy vehicle), which can realize multimodal interaction and diverse task execution through the cockpit system. See below for reference. Figure 5 It shows a structural schematic diagram of a vehicle suitable for implementing the embodiments of this application. Figure 5The vehicle shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments described in this application. Figure 5 As shown, the vehicle may include a processing unit 1001 (e.g., a central processing unit, a graphics processor, or the smart cockpit controller mentioned in the above embodiments), which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 1002 or a program loaded from storage device 1003 into random access memory (RAM) 1004. The random access memory 1004 also stores various programs and data required for vehicle operation. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the vehicle to communicate wirelessly or wiredly with other devices to exchange data. Although a vehicle with various systems is shown in the figures, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0104] The user-related data involved in this application, such as the historical interaction records and location context data mentioned above, were all obtained with the user's permission or consent; In other words, when this application is applied to a specific product or technology, user permission is required to acquire and process the relevant data, and the processing of the relevant data must comply with the relevant laws, regulations and regulatory standards of the relevant countries and regions.

[0105] For example, when it is necessary to obtain a user's location context, a prompt to obtain the location context can be displayed on the user's terminal or in-vehicle display screen, and the vehicle can only obtain the user's real-time location context after receiving the user's confirmation operation for the location context.

[0106] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, an embodiment disclosed in this application includes an interactive model information processing program product, which includes an interactive model information processing program carried on a computer-readable medium, the interactive model information processing program containing program code for performing the methods shown in the flowcharts. In such an embodiment, the interactive model information processing program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the interactive model information processing program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0107] The vehicle provided in this application, employing the information processing method of the interaction model in the above embodiments, can solve the technical problems of unreasonable resource allocation and low processing efficiency in the context information processing schemes of existing interaction models. Compared with the prior art, the beneficial effects of the vehicle provided in this application are the same as those of the information processing method of the interaction model provided in the above embodiments, and other technical features of the vehicle are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0108] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0109] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0110] This application provides a storage medium having computer-readable program instructions (i.e., an information processing program for an interaction model) stored thereon, which are used to execute the information processing method for the interaction model in the above embodiments.

[0111] The storage medium provided in this application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0112] The aforementioned storage medium may be included in the vehicle or may exist independently without being installed in the vehicle.

[0113] The aforementioned storage medium carries one or more programs, which, when executed by the vehicle, enable the vehicle to perform contextual information optimization processing of the interaction model.

[0114] Information processing program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0115] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of information processing program products of systems, methods, and interaction models according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0116] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0117] The readable storage medium provided in this application is a storage medium that stores computer-readable program instructions (i.e., the information processing program of the interaction model) for executing the information processing method of the above-described interaction model. This solves the technical problem of unreasonable resource allocation and low processing efficiency in existing interaction model context information processing schemes. Compared with the prior art, the beneficial effects of the storage medium provided in this application are the same as those of the information processing method of the interaction model provided in the above embodiments, and will not be repeated here.

[0118] The above are only some embodiments of this application and do not limit the scope of the solution of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included within the protection scope of this application.

Claims

1. An information processing method for an interactive model, characterized in that, The method includes: Obtain multi-source heterogeneous information corresponding to the current interactive task; Based on the multi-source heterogeneous information, the current interactive task is matched at the scene level to obtain the current task scene; Determine the target context compression strategy corresponding to the current interactive task based on the current task scenario; The target context compression strategy guides the interaction model to compress the multi-source heterogeneous information and generate the target context information corresponding to the current interaction task.

2. The information processing method for the interaction model as described in claim 1, characterized in that, The step of determining the target context compression strategy corresponding to the current interaction task based on the current task scenario includes: Obtain the task classification confidence level corresponding to the current interactive task; Based on the task classification confidence level and the current task scenario, the target context compression strategy corresponding to the current interactive task is dynamically determined through a dual-path selection mechanism.

3. The information processing method for the interaction model as described in claim 2, characterized in that, The step of dynamically determining the target context compression strategy corresponding to the current interaction task based on the task classification confidence level and the current task scenario through a dual-path selection mechanism includes: Obtain the preset threshold; If the confidence level of the task classification is greater than or equal to the preset confidence threshold, the preset compression strategy corresponding to the current task scenario will be used as the target context compression strategy corresponding to the current interactive task.

4. The information processing method for the interaction model as described in claim 3, characterized in that, The current task scenarios include: basic operation scenarios, scenario-based service scenarios, and / or cross-domain collaboration scenarios; The preset compression strategy corresponding to the basic operation scenario is a key information extraction strategy, which is a compression strategy for extracting preset key information contained in multi-source heterogeneous information. The preset compression strategy corresponding to the scenario service is a visual compression strategy, which is a compression strategy that converts the multi-source heterogeneous information into an image to be analyzed and performs visual language analysis on the image to be analyzed. The preset compression strategy corresponding to the cross-domain collaborative scenario is a summary-style text compression strategy, which is a compression strategy that extracts the core semantic information of the multi-source heterogeneous information.

5. The information processing method for the interaction model as described in claim 2, characterized in that, The step of dynamically determining the target context compression strategy corresponding to the current interaction task based on the task classification confidence level and the current task scenario through a dual-path selection mechanism further includes: Obtain the preset threshold; When the confidence level of the task classification is less than the preset confidence threshold, the degree of dimension matching is obtained based on the matching between the preset compression strategy and the multi-dimensional scoring indicators. The multi-dimensional scoring indicators include at least one of compression efficiency, information fidelity and resource adaptability. Based on the degree of dimensional matching, the comprehensive compression score corresponding to the preset compression strategy is determined, and the preset compression strategy with the highest comprehensive compression score is taken as the target context compression strategy corresponding to the current interactive task.

6. The information processing method for the interaction model as described in claim 1, characterized in that, After obtaining the multi-source heterogeneous information corresponding to the current interactive task, the method further includes: When the multi-source heterogeneous information triggers a preset edge scenario, the context compression strategy corresponding to the preset edge scenario is used as the target context compression strategy corresponding to the current interaction task.

7. The information processing method for the interaction model as described in any one of claims 1 to 6, characterized in that, The step of performing scene-level matching on the current interactive task based on the multi-source heterogeneous information to obtain the current task scene includes: A comprehensive semantic analysis is performed on the multi-source heterogeneous information to obtain multi-dimensional task information corresponding to the current interaction task. The multi-dimensional task information includes scene information and / or operation boundaries. The multidimensional task information is mapped to a preset multidimensional hierarchical index to obtain a task feature vector; The current task scenario is determined based on the degree of matching between the preset scenario classification standard and the task feature vector.

8. An information processing device for an interactive model, characterized in that, The information processing device for the interaction model includes: The information acquisition module is used to acquire multi-source heterogeneous information corresponding to the current interactive task; The scene analysis module is used to perform scene-level matching on the current interactive task based on the multi-source heterogeneous information to obtain the current task scene; The strategy selection module is used to determine the target context compression strategy corresponding to the current interaction task based on the current task scenario. The information processing control module is used to guide the interaction model to compress the multi-source heterogeneous information through the target context compression strategy, and generate the target context information corresponding to the current interaction task.

9. A vehicle, characterized in that, The vehicle includes: a memory, a processor, and an information processing program for an interactive model stored in the memory and executable on the processor, the information processing program for the interactive model being configured to implement the steps of the information processing method for the interactive model as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores an information processing program for the interaction model, which, when executed by a processor, implements the steps of the information processing method for the interaction model as described in any one of claims 1 to 7.

Images