A Structured Data Generation Method, Device, Terminal Device, and Storage Medium

Abstract

The present application provides a method, device, terminal device and storage medium for generating structured data, which is applicable to the technical field of data processing. The method includes: obtaining heterogeneous cabin data; obtaining current state data and current behavior data according to the heterogeneous cabin data; generating initial structured data according to the current state data, current behavior data, preset prompt text information and preset initial structured data generation model; and generating target structured data according to the initial structured data and preset target structured data generation model. The present application can realize the conversion processing of structured representation of multi-source heterogeneous data in the intelligent cockpit, effectively improve the data processing ability and knowledge expression level of the intelligent cockpit, can flexibly adapt to the changing requirements of intelligent cockpit data, and provides strong support for the development of the intelligent cockpit.

Description

Technical Field

[0001] This application belongs to the technical field of data processing, and particularly relates to a method, device, terminal device, and storage medium for generating structured data. Background Art

[0002] With the continuous upgrade of vehicle automation and the rapid development of intelligent cockpit technology, the interaction mode in the cockpit has gradually evolved from traditional physical buttons and dashboard operations to more complex multimodal interaction methods, such as voice commands, gesture control, touch screen operations, etc. This upgrade in the interaction mode makes the driver's behavior more diverse, the interaction intention more concealed, and there are complex correlations between different behaviors, further exacerbating the complexity of cockpit data. In addition, intelligent cockpits involve a large amount of multi-source heterogeneous data, including status information collected by sensors, user interaction data, environmental perception data, etc. These data are large in scale, diverse in format, and inconsistent in semantic levels, making it difficult to intuitively identify the correlation relationships between information and increasing the difficulty of extracting valuable knowledge from the original data. Due to the lack of effective data parsing and organization methods, the utilization rate of cockpit data is low, the data redundancy is high, and the information fragmentation is serious, making it difficult to efficiently support in-depth mining and accurate reasoning of downstream tasks such as behavior analysis, intention prediction, and personalized recommendation.

[0003] In the prior art, deep learning models are used to automatically extract structured information from unstructured texts in cockpit data. However, deep learning models usually require a large amount of labeled data for training. In specific fields, especially in fields with strong professionalism or scarce data, it may be very difficult and expensive to obtain sufficient high-quality training data. Especially in the field of intelligent cockpits, due to the highly contextual and private nature of the data, publicly available high-quality cockpit knowledge data is scarce, resulting in the difficulty of directly applying traditional supervised learning methods. In addition, traditional deep learning models often lack reasoning capabilities and are difficult to accurately identify the logical relationships in the highly dynamic and implicitly causal data in the cockpit, thus affecting the integrity and usability of the structured expression. Summary of the Invention

[0004] In view of this, the embodiments of this application provide a method, device, terminal device, and storage medium for generating structured data, aiming to solve the problems of high training costs and inability to adapt to the complex dynamic changes of intelligent cockpits in the technology of automatically extracting structured information from unstructured texts in cockpit data, resulting in low integrity and effectiveness of the extracted and processed structured data.

[0005] The first aspect of the embodiments of this application provides a method for generating structured data, including:

[0006] Obtain heterogeneous cockpit data;

[0007] Based on the cabin state heterogeneous data, obtain the current state data and the current behavior data;

[0008] Generate initial structured data according to the current state data, the current behavior data, the preset prompt text information, and the preset initial structured data generation model;

[0009] Generate target structured data according to the initial structured data and the preset target structured data generation model.

[0010] The second aspect of the embodiments of the present application provides a structured data generation device, including:

[0011] A cabin state heterogeneous data acquisition module, configured to acquire cabin state heterogeneous data;

[0012] A current state behavior data generation module, configured to obtain the current state data and the current behavior data according to the cabin state heterogeneous data;

[0013] An initial structured data generation module, configured to generate initial structured data according to the current state data, the current behavior data, the preset prompt text information, and the preset initial structured data generation model;

[0014] A target structured data generation module, configured to generate target structured data according to the initial structured data and the preset target structured data generation model.

[0015] The third aspect of the embodiments of the present application provides a terminal device, where the terminal device includes a memory and a processor, and a computer program that can run on the processor is stored on the memory. When the processor executes the computer program, the steps of the structured data generation method described in the first aspect above are implemented.

[0016] The fourth aspect of the embodiments of the present application provides a computer-readable storage medium, including: storing a computer program, and when the computer program is executed by a processor, the steps of the structured data generation method described in the first aspect above are implemented.

[0017] The beneficial effects of the embodiments of the present application compared with the prior art are as follows: By processing the cabin state heterogeneous data, the current state and behavior data are obtained, which is convenient for deeply understanding the real-time situation of the cockpit. Using the preset prompt text and the initial structured data generation model to further sort out the data relationship, and then outputting the structured data that meets the requirements through the target structured data generation model, so as to realize the standardized processing of multi-source heterogeneous data, improve the quality and effectiveness of the cabin state data, and provide reliable data support for the downstream tasks of the intelligent cockpit. Description of the Drawings

[0018] To more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0019] Figure 1 It is a schematic flowchart of the implementation of the structured data generation method provided in the first embodiment of the present application;

[0020] Figure 2 It is a schematic flowchart of the implementation of the structured data generation method provided in the second embodiment of the present application;

[0021] Figure 3 It is a schematic flowchart of the implementation of the structured data generation method provided in the third embodiment of the present application;

[0022] Figure 4 It is a schematic flowchart of the implementation of the structured data generation method provided in the fourth embodiment of the present application;

[0023] Figure 5 It is a schematic flowchart of the implementation of the structured data generation method provided in the fifth embodiment of the present application;

[0024] Figure 6 It is a schematic flowchart of the implementation of the structured data generation method provided in the sixth embodiment of the present application;

[0025] Figure 7 It is a schematic flowchart of the implementation of the structured data generation method provided in the seventh embodiment of the present application;

[0026] Figure 8 It is a schematic diagram of the structure of the structured data generation device provided in the embodiments of the present application;

[0027] Figure 9 It is a schematic diagram of the terminal device provided in the embodiments of the present application. Specific embodiments

[0028] In the following description, specific details such as specific system structures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present application. However, those skilled in the art should clearly understand that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present application.

[0029] To illustrate the technical solutions described in the present application, the following will be described through specific embodiments.

[0030] Figure 1 The following is a flowchart showing the implementation of the structured data generation method provided in the first embodiment of the present application:

[0031] Step S101: Obtain heterogeneous cabin data.

[0032] In this embodiment, the heterogeneous cabin data can be automatically collected by the intelligent cockpit and can include data generated by multimodal interaction methods, such as voice command data, gesture control data, touch screen operation data, etc.

[0033] Step S102: Obtain the current state data and the current behavior data according to the heterogeneous cabin data.

[0034] In this embodiment, the heterogeneous cabin data can be first converted into a standardized text representation through a semantic parsing method to make the heterogeneous cabin data have a clearer semantic structure and enhance the parsability of the data. Then, attribute definition processing can be performed on the text representation, which can be semantic state attribute definition and semantic behavior attribute definition. That is, discrete data such as gear information data, engine state data, and in-vehicle air conditioner state data, and continuous data such as vehicle speed data and temperature data in the heterogeneous cabin data are discretized for continuous variables, that is, they are divided into finite intervals and specific level labels are assigned to avoid data node explosion and improve calculation efficiency. In addition, to improve the information granularity, the behavior attributes are represented hierarchically, that is, divided into a general attribute layer and a detailed attribute layer. Among them, the in-vehicle air conditioner state data includes air conditioner switch state data and air conditioner mode data, etc. After semantic state attribute definition and semantic behavior attribute definition, the current state data and the current behavior data are obtained. It can be understood that the behavior data can be used to represent the behavior measures taken by the driver, such as turning on the navigation, turning on the music player, turning on the air conditioner, etc., and the state data corresponds to the behavior data and is used to represent the states of various functions in the intelligent cockpit, such as the navigation state is on, the music player state is on, the air conditioner state is on, etc.

[0035] Step S103: Generate initial structured data according to the current state data, the current behavior data, the preset prompt text information, and the preset initial structured data generation model.

[0036] In this embodiment, the preset initial structured data generation model can be set based on the LLM model. The preset prompt text information can be set manually and is used to limit the output content range of the initial structured data generation model to avoid outputting unnecessary redundant information. The current state data, the current behavior data, and the prompt text information can be used as the input data of the initial structured data generation model, and the output result of the initial structured data generation model is used as the initial structured data.

[0037] Step S104: Generate target structured data based on the initial structured data and a preset target structured data generation model.

[0038] In this embodiment, the preset target structured data generation model can be set based on the LLM model. The initial structured data generation model and the target structured data generation model can be trained based on different data. The initial structured data can be used as the input data of the target structured data generation model, and the output result of the target structured data generation model can be used as the target structured data, so as to realize the conversion and output of structured data.

[0039] The structured data generation method provided by the embodiments of the present application processes cabin state heterogeneous data to obtain the current state and behavior data, which is convenient for deeply understanding the real-time situation of the cockpit. The preset prompt text and the initial structured data generation model are used to further sort out the data relationship, and then the target structured data generation model outputs the structured data that meets the requirements, so as to realize the standardized processing of multi-source heterogeneous data, improve the quality and effectiveness of cabin state data, and provide reliable data support for downstream tasks of the intelligent cockpit.

[0040] Figure 2 The flowchart of the structured data generation method provided by the second embodiment of the present application is shown. The difference from the first embodiment above is that step S102 specifically includes:

[0041] Step S201: Perform attribute definition processing on the cabin state heterogeneous data to obtain state attribute data and behavior attribute data.

[0042] In this embodiment, the attribute definition processing can be semantic state attribute definition and semantic behavior attribute definition. That is, first convert the cabin state heterogeneous data into a standardized text representation, so that the cabin state heterogeneous data has a clearer semantic structure and enhances the parsability of the data. Then, perform attribute definition processing on the text representation, which can be semantic state attribute definition and semantic behavior attribute definition. That is, for discrete data such as gear information data, engine state data, and in-vehicle air-conditioning state data, and continuous data such as vehicle speed data and temperature data in the cabin state heterogeneous data, the continuous variables are discretized, divided into finite intervals and specific grade labels are assigned, so as to obtain state attribute data and behavior attribute data.

[0043] Step S202: Obtain mapped behavior data according to the state attribute data and a preset semantic mapping function.

[0044] In this embodiment, the preset semantic mapping function can be set manually and is used to map the state attribute data into behavior data to obtain the mapped behavior data. Thus, for some behavior data that are difficult to input, the mapped behavior data are used for identification and supplementation.

[0045] Step S203: Obtain the backtracking mapped state data according to the behavior attribute data and the preset semantic backtracking function.

[0046] In this embodiment, the preset semantic backtracking function can be set manually and is used to perform backtracking processing on the behavior attribute data to obtain the backtracking mapped state data. Thus, for some state data that are difficult to input, the backtracking mapped state data are used for identification and supplementation. For example, the existing behavior attribute data are insufficient in capturing behaviors such as seat adjustment and door opening / closing, and the state attribute data lack descriptions of the states of the devices in the cockpit. State mapping can obtain relevant behavior data from the state attribute data. For example, the state of the seat rail motor can represent the seat adjustment behavior; backtracking mapping can mine the states of the devices in the cockpit from the behavior attribute data. For example, the behavior of playing music proves that the music state is on.

[0047] Step S204: Perform supplementation processing on the state attribute data and the behavior attribute data according to the mapped behavior data and the backtracking mapped state data to obtain the state data intermediate variable and the behavior data intermediate variable.

[0048] In this embodiment, the mapped behavior data are supplemented into the behavior attribute data to obtain the behavior data intermediate variable, and the backtracking mapped state data are supplemented into the state attribute data to obtain the state data intermediate variable.

[0049] Step S205: Obtain the current state data and the current behavior data according to the state data intermediate variable, the behavior data intermediate variable, and the preset state matching function.

[0050] In this embodiment, the preset state matching function can be set manually. The state data intermediate variable can be used as the independent variable of the state matching function, and the function value calculated by the state matching function is used as the current state data. Then, the current state data is matched with the behavior data intermediate variable to obtain the current behavior data. It can be understood that the current state data and the current behavior data are in one-to-one correspondence. For example, if the current behavior data is "turn on the navigation", then the current state data is "the current navigation state is on".

[0051] The structured data generation method provided by the embodiments of the present application infers the driver's operation behavior based on the state change or infers the state change from the behavior by mutually mapping the state attribute data and the behavior attribute data, enhances the understanding of the cockpit environment, effectively fills in the blank of attribute definition, so as to have a more comprehensive understanding of the driver's behavior pattern and vehicle usage situation, provides richer and more accurate data support for downstream tasks of the intelligent cockpit, such as behavior analysis, intention prediction, personalized recommendation, etc., enhances the data availability, gives full play to the data value of the intelligent cockpit, and helps optimize the functions of the intelligent cockpit and improve the user experience.

[0052] Figure 3 The implementation flowchart of the structured data generation method provided by Embodiment 3 of the present application is shown. The difference from Embodiment 2 above is that the step S205 specifically includes:

[0053] Step S301: Filter the state data intermediate variable and the behavior data intermediate variable to obtain a first state data intermediate variable and a first behavior data intermediate variable.

[0054] In this embodiment, the filtering process is used to delete duplicate data under the same data point and exclude invalid data or low-value data. Among them, "under the same data point" refers to the data set captured by a specific data collection point or data marking point during the data collection process. The filtering method can be that if the same behavior occurs repeatedly within 5 seconds, only the first behavior is retained. That is, if a certain sensor in the cockpit continuously sends the same behavior information, this data is considered redundant. Specifically, there are behavior attributes with multiple identical contents continuously appearing within the same second in the data, which is common in music-related behaviors. For example, the "start playing music" behavior is repeatedly recorded within the same second or consecutive seconds, which does not conform to the real usage scenario. If it is not filtered, it will lead to a large amount of noise in the process of structured knowledge expression. It can also be that when the user continuously adjusts the temperature or mode of the vehicle air conditioner, only the last setting content is retained. That is, when the user adjusts some cockpit functions, multiple adjustments are usually made to find the optimal value. Specifically, during the data recording process, if all intermediate adjustment steps are retained, a large amount of redundant data will be generated. Therefore, only the user's final setting value can be retained to ensure the simplicity and effectiveness of the data and avoid interference with subsequent analysis and decision-making due to excessive intermediate states. It can also be to exclude behavior attributes with a frequency lower than 3%. That is, during the data analysis process, the occurrence frequency of some behavior attributes is extremely low. The occurrence frequency lower than 3% can be regarded as extremely low. Therefore, they can be classified as occasional events, low-efficiency data, or non-critical behaviors. Specifically, by setting a 3% occurrence frequency threshold, rare behaviors can be excluded to ensure that the most representative and valuable information is retained in the data set and improve the data utilization efficiency. After any of the above methods, the first state data intermediate variable and the first behavior data intermediate variable can be obtained for subsequent sequential logic correction processing.

[0055] Step S302: Perform sequential logic correction processing on the first state data intermediate variable and the first behavior data intermediate variable to obtain a second state data intermediate variable and a second behavior data intermediate variable.

[0056] In this embodiment, the timing logic correction process may refer to that when logical conflicts, violations of facts, time misalignments, etc. occur in the front and rear operations, the operation that most conforms to logic or the actual situation should be selected according to the actual situation to correct the logical contradictions caused by sampling errors. The method of the timing logic correction process can adopt methods such as adjusting or aligning timestamps. For example, when the driver's operation (such as turning on the air conditioner) is wrongly recorded in time as occurring after the next event (adjusting the air volume of the air conditioner), it is necessary to solve the logical contradiction and time misalignment between the front and rear behaviors by adjusting or aligning timestamps, etc., to ensure the correctness of the operation sequence. The first state data intermediate variable and the first behavior data intermediate variable are used as the second state data intermediate variable and the second behavior data intermediate variable after timestamp adjustment or alignment for further processing later.

[0057] Step S303: Perform missing data filling processing on the second state data intermediate variable and the second behavior data intermediate variable to obtain the to-be-matched state data and the current behavior data.

[0058] In this embodiment, the missing data filling processing may refer to, for some missing data, filling in the carry using subsequent values or inferring the missing values based on surrounding data, so as to ensure the continuity and integrity of the data stream. Specifically, the missing data filling processing may be that if the ambient temperature value at a certain moment is invalid, for example, the ambient temperature value exceeds the normal range, then the subsequent valid data can be used for filling; or when the vehicle speed data at a certain moment is missing, the vehicle speed data at several previous and subsequent time points can be used for interpolation filling. The second state data intermediate variable and the second behavior data intermediate variable are used as the to-be-matched state data and the current behavior data after the missing data filling processing for further screening processing later.

[0059] Step S304: Obtain multiple candidate state data according to the current behavior data and a preset state matching function.

[0060] In this embodiment, the preset state matching function can be set artificially. Multiple current behavior data can be used as the independent variables of the state matching function, and the function values calculated are output as multiple candidate state data for further screening processing later.

[0061] Step S305: Screen the multiple candidate state data to obtain the current state data.

[0062] In this embodiment, it can be understood that there are more than 30 existing state attributes. If all state attributes need to be comprehensively summarized when describing the vehicle state in which each behavior occurs, the text data will be too cumbersome. The existence of a large amount of state data unrelated to the behavior will increase the understanding difficulty and processing burden of the large model. Therefore, only the state attributes closely related to a specific behavior need to be retained, which can not only reduce the complexity of the data, but also reduce the interference of irrelevant state information on the model performance, so as to more accurately capture the key association between the behavior and the state and ensure the accuracy and practicality of the analysis results. Therefore, it can be to delete the candidate state data unrelated to the specific behavior and only retain the candidate state data related to the specific behavior. Specifically, it can be to calculate the correlation between the candidate state data and the state data corresponding to the specific behavior data, and set a specific correlation threshold. When the correlation between the candidate state data and the state data corresponding to the specific behavior data is greater than the specific correlation threshold, the corresponding candidate state data is retained, so as to complete the screening of the candidate state data and obtain the current state data.

[0063] The structured data generation method provided by the embodiment of the present application filters the state data and behavior data in the multi-source heterogeneous cabin state data to exclude duplicate data and invalid data, and then through time series correction processing, avoids problems such as logical conflicts, violation of facts, and time dislocation in the data used for subsequent calculations. Through missing data filling processing, the missing data is filled to ensure the continuity and integrity of the data stream. Then, through further screening processing, the invalid state data is eliminated, which ensures the integrity, consistency, and validity of the data and reduces the interference for subsequent calculations, effectively improving the quality and reliability of structured data generation.

[0064] Figure 4 The implementation flowchart of the structured data generation method provided by the fourth embodiment of the present application is shown. The difference from the first embodiment above is that the step S103 specifically includes:

[0065] Step S401, obtain historical behavior data.

[0066] In this embodiment, the historical behavior data can be the behavior data of the user during past driving automatically collected by the intelligent cockpit.

[0067] Step S402, obtain behavior sequence data according to the current behavior data, historical behavior data, and a preset behavior tracking function.

[0068] In this embodiment, the preset behavior tracking function can be artificially set and is used to further extract the behavior-related numerical values in the current behavior data and historical behavior data. The current behavior data and historical behavior data can be used as the independent variables of the behavior tracking function, and the calculated function value is output as the behavior sequence data.

[0069] Step S403: Obtain the current state data according to the current state data and the preset state detection function.

[0070] In this embodiment, the preset state detection function can be artificially set and is used to detect the state data corresponding to the current behavior, so as to further extract the state-related numerical values in the current state data. The current state data can be used as the independent variable of the state detection function, and the calculated function value is output as the current state data.

[0071] Step S404: Generate initial structured data according to the current behavior data, current state data, behavior sequence data, preset prompt text information, and preset initial structured data generation model.

[0072] In this embodiment, the preset initial structured data generation model can be set based on the LLM model, or it can be an LLM model trained with data collected from a large number of different intelligent cockpits. The calculation process can be to use the official conversation completion API to submit a series of messages to interact with the LLM model. The preset prompt text information can be artificially set and is used to limit the range of the output content of the initial structured data generation model to avoid excessive redundant text information output by the initial structured data generation model. The current behavior data, current state data, and behavior sequence data combined with the prompt text information can be used as the input data of the initial structured data generation model, and the output data obtained after calculation by the initial structured data generation model is used as the initial structured data. The initial structured data can be output in the form of triples, including text content information and structured knowledge label information. It can be understood that a piece of text content can be extracted into multiple different (entity, relationship, entity) groups through the initial structured data generation model to represent the logical associations between different entities, identify the causal, parallel, or hierarchical relationships existing in the interaction process, and use the text extraction content in the form of (entity, relationship, entity) as a label for the text content. Thus, a piece of text content can have multiple different labels, and multiple text contents and their corresponding multiple different labels are output as the initial structured data.

[0073] The structured data generation method provided by the embodiments of the present application further extracts valid text data from the current behavior data and the current state data through a behavior tracking function and a state detection function, parses the valid text data through an initial structured data generation model, and restricts the range of the output content of the initial structured data generation model through prompt text, avoiding redundant text data in the output content, making the generated initial structured data highly compatible with the cockpit scenario. Whether it is complex driving environment data or diverse user interaction data, it can accurately and effectively reflect the actual operation of the intelligent cockpit, providing a reliable data basis for the subsequent processing of structured data.

[0074] Figure 5 FIG. 4 shows a flowchart of the implementation of the structured data generation method provided by the fifth embodiment of the present application, which is different from the first embodiment above: step S104 specifically includes:

[0075] Step S501: Perform encoding processing on the initial structured data to obtain initial structured data encoding information.

[0076] In this embodiment, the encoding processing can be to process the initial structured data based on the text data encoding function in the StructLlama model. Specifically, the prompt text can be converted into a series of parsable text token sequences through the text tokenizer in the StructLlama model, which is used to split the text into words or subwords and assign a unique ID to each token so that the target structured data generation model can process it. At the same time, the text tokenizer generates position encoding to provide the target structured data generation model with the position information of the tokens in the sequence, enhancing the semantic understanding ability of the initial structured data. The data after encoding processing is output as the initial structured data encoding information.

[0077] Step S502: Generate target structured representation data according to the initial structured data encoding information and a preset target structured data generation model.

[0078] In this embodiment, the LLM backbone part in the StructLlama model can be used as the preset target structured data generation model. The StructLlama model is pre-trained through a dataset. During the training process, the optimization objective is to minimize the loss function of the model in the knowledge prediction task rather than optimizing the parameters in the model. The specific parameter optimization calculation formula can be expressed as:

[0079] ;

[0080] Where represents the optimized model parameters, represents the training dataset in the sample The expected value of Represents the input data of the StructLlama model, consisting of chain instructions and interactive behavior information And related auxiliary information composition, represents structured knowledge tags, Indicates the current parameters Next, the model predicts the next target tag The probability distribution of and the target sequence generated previously . represents the total length of the output sequence, that is, the length of the target knowledge label, t It represents the length of the structured knowledge label. The target structured representation data is the structured knowledge label calculated after the StructLlama model is trained.

[0081] Step S503: Decode the target structured representation data to obtain target structured data.

[0082] In this embodiment, the target structured representation data may be data calculated and output by the LLM backbone part in the StructLlama model. The decoding process may be to convert the target structured representation data into triple label data. In order to ensure that the generated triple labels are accurate and consistent, the decoding process may introduce rules or templates to standardize the output format and content. Specifically, the decoding process first extracts relevant entities and relationships from the intermediate representation. This step uses a constraint parsing strategy to ensure that the extracted entities and relationships conform to predefined types, and then combines this information into triple form according to the predefined template. Specifically, the target structured data It can be expressed as:

[0083] ;

[0084] in, Representing Entities The relationship between is the inference function used by the model to match relations; They represent the entity set and relationship set in the intermediate representation respectively.

[0085] The structured data generation method provided by the embodiment of the present application converts the initial structured data into token sequence data that can be processed by the target structured data generation model through text encoding processing, and generates position encoding data, so that the target structured data generation model can accurately understand the semantics and order of the initial structured data, improve the accuracy and integrity of data transfer calculation, deeply process the professional data of the intelligent cockpit through the target structured data generation model, generate target structured representation data, improve the accuracy of knowledge prediction for cabin state data, and be used to provide more reliable knowledge support for various applications of the intelligent cockpit. By decoding the target structured representation data, the generated target structured data can accurately reflect the connotation and logical relationship of the cabin state data, improve the readability and availability of the cabin state data, and facilitate subsequent data storage, reasoning, and application.

[0086] Figure 6 FIG. 4 shows a flowchart of the implementation of the structured data generation method provided by the sixth embodiment of the present application, which is different from the first embodiment above in that: after the step S104, the following steps are further included:

[0087] Step S601: Calculate the semantic similarity of each of the target structured data according to a preset similarity calculation function and a preset similarity calculation weight to obtain the structured data semantic similarity.

[0088] In this embodiment, the preset similarity calculation function can be set manually and can be designed based on trigonometric functions. The preset similarity calculation weight can be set manually and is used to regulate the influence of entities and relationships in the similarity calculation. The semantic similarity of each pair of target structured data can be calculated through the similarity calculation function, and then the weighted sum calculation of the semantic similarity can be performed through the similarity calculation weight to obtain the structured data semantic similarity. Specifically, the target structured data can be a triple, and the semantic similarity of each target structured data can be expressed as:

[0089] ;

[0090] where is used to represent a clustering cluster, , respectively represent two triples, that is, two target structured data;

[0091] Structured data semantic similarity and The calculation formula of can be expressed as:

[0092] ;

[0093] where is a preset similarity calculation weight, which is used to control the influence of entities and relationships in similarity calculation. Indicates calculating the semantic similarity between entities, Indicates calculating the similarity between relationships, Indicates calculating and the similarity between.

[0094] Step S602, determine whether the semantic similarity of the structured data is greater than a preset semantic similarity threshold; if so, go to step S603; if not, do not process the target structured data corresponding to the semantic similarity of the structured data.

[0095] In this embodiment, according to the size relationship between the semantic similarity of the structured data and the preset semantic similarity threshold, clustering calculation is performed on the structured data. The structured data with high semantic similarity is grouped into the same clustering cluster, and the structured data with low semantic similarity is grouped into different clustering clusters. The semantic similarity of the structured data can be the cosine similarity. Assuming that the vector embedding representation of entity is , then the calculation method of the semantic similarity of the structured data can be expressed as:

[0096] ;

[0097] The preset semantic similarity threshold , when or , and are grouped into the same cluster, that is . After clustering, representative triples in each cluster are selected, de-duplicated and normalized to form the final structured data.

[0098] Step S603, generate a structured data clustering cluster according to the target structured data corresponding to the semantic similarity of the structured data.

[0099] In this embodiment, if the semantic similarity of the structured data is greater than the preset semantic similarity threshold, it indicates that the semantic similarities of multiple structured data are high and can be grouped into the same clustering cluster. The structured data can be presented in the form of triples. Taking the clustered triples as an example, the structured data clustering cluster can be expressed as "(Adjust the air-conditioning blowing mode, current state, idle speed)", "(Adjust the air-conditioning blowing mode, in the state of, idle speed)" and "(Idle speed, accompanied by, adjusting the air-conditioning blowing mode)", which are used to represent different expressions of the same semantic text content through the same clustering cluster, but are all used to describe the states between similar relationships or entities.

[0100] Step S604: Standardize the structured data clustering clusters to obtain standard structured data.

[0101] In this embodiment, the standardization process can be deduplication, normalization, and correction based on human experts, which is used to transform the structured data in the structured data clustering clusters into refined, standardized, and accurate standard structured data, facilitating subsequent further analysis, processing, or storage. Among them, the deduplication process can be to delete redundant information with duplicate semantics, the normalization process can be to correct semantic errors, and the correction process based on human experts can be to evaluate the integrity of triples based on prior knowledge, such as identifying missing or incomplete entities, and can also correct the generated structured knowledge, identifying logical errors or inefficient triples.

[0102] The structured data generation method provided by the embodiment of the present application further optimizes the structured data, performs similarity calculation and matching on the structured data, determines semantic equivalence, and is used to cluster high-similarity or duplicate data, so as to identify and delete overlapping information for the repetitive operations of the intelligent cockpit driver, thereby eliminating semantic redundancy, ensuring the consistency of the structured data, making the structured data expression more concise and effective, ensuring the unity and standardization of the structured data, not only eliminating data redundancy, but also promoting the systematic integration and in-depth mining of data, and providing more efficient data support for downstream tasks such as behavior analysis and intention prediction of the intelligent cockpit.

[0103] Figure 7 The flowchart showing the implementation of the structured data generation method provided in the seventh embodiment of the present application is different from the sixth embodiment above in that: after the step S604, it further includes:

[0104] Step S701: Obtain logical correction information and truth value annotation information.

[0105] In this embodiment, it can be understood that a triple is marked as correct depending on two factors. One is that the description of the entity is correct, intuitive, and relevant to the input text. The other is that it accurately expresses a real and reasonable relationship, that is, the triple conforms to the logic of the real world. Both the logical correction information and the truth value annotation information can be input by human experts. Take the following triples as examples: (1) Triple: (Adjust the heating level of the co-pilot seat, the environment, the temperature range is (-5, 0]), and the logical correction information input by the human expert is: inaccurate, not meeting the requirement of "the description of the entity is correct, intuitive, and relevant to the input text". The entity description is ambiguous. In the input text data, the temperature information includes ambient temperature, cabin temperature, and air conditioner set temperature. In this triple, the specific attribution of the temperature is not clearly expressed; (2) Triple: (No direct sunlight / very low sunlight intensity, is beneficial to, adjust the air conditioner blowing mode), and the logical correction information input by the human expert is: inaccurate, not meeting the requirement of "accurately expressing a real and reasonable relationship, that is, the triple conforms to the logic of the real world", because the relationship between "no direct sunlight / very low sunlight intensity" and "adjust the air conditioner blowing mode" is not directly relevant or has weak logic. This description fails to clearly show a reasonable causal relationship between the two; (3) Triple: (The driving closed state of the driver's seat, causes, adjust the air conditioner blowing level), and the logical correction information input by the human expert is: inaccurate, not meeting the requirement of "accurately expressing a real and reasonable relationship, that is, the triple conforms to the logic of the real world". There is no causality between "the driving closed state of the driver's seat" and "adjust the air conditioner blowing level", and the expressed relationship lacks correspondence and practical logical support in real life. Based on the logical correction information, human experts further perform truth value annotation, that is, provide a list of "missing" triples for each data point, that is, relevant triples included in the input text but not retrieved by the model. The list of triples provided by the human expert is used as the truth value annotation information.

[0106] Step S702: According to the logical correction information and the truth value annotation information, perform verification and correction processing on the standard structured data to obtain optimized structured data.

[0107] In this embodiment, based on the logical correction information and the truth value annotation information input by human experts, perform comparative verification and modification and calibration processing on the standard structured data, and the corrected data is output as the optimized structured data.

[0108] Step S703: According to the optimized structured data, the preset group structured database, and the preset user intention determination model, obtain user intention information.

[0109] In this embodiment, the preset group structured database can be obtained by automatically collecting and integrating the driving conditions of different drivers by multiple different intelligent cockpits. The preset user intention determination model can be designed based on the LLM model. The optimized structured data and the group structured database can both be used as the input data of the user intention determination model, and the user intention information can be calculated through the user intention determination model.

[0110] In this embodiment, the optimized structured data can be the personal cabin state text data of some drivers, which is presented in the form of triples. By designing customized prompt words and using the context understanding and generation capabilities of the LLM model, the large language model is guided to analyze the data information of specific users and learn the preferences of users, etc., so as to be able to simulate the potential needs and experience feelings of users in different driving scenarios. For example, by analyzing the historical behaviors of drivers, environmental variables (such as the temperature, humidity, and light inside and outside the vehicle), and interaction records, the model can generate demand expressions that conform to the characteristics of specific users, such as emotional expressions like "It's so hot", "The air in the car is not circulating / It's stuffy in the car", "The light is too strong", "So tired", etc., and demand intention expressions like "Can the air conditioner wind not blow on my face", "Want to listen to music for a while", etc. In addition, based on the above perception expressions, the system can further construct hypothetical scenario deductions, and through logical reasoning and knowledge base matching, analyze the influencing factors and functional requirements behind them. For example, for the user feeling of "It's so hot", the possible causes may include too high cabin temperature, the air conditioner not being turned on or the air conditioner temperature being set too high, low air volume, and too high sunlight intensity, etc., and the possible functions involved may include adjusting the air conditioner temperature, changing the wind speed mode, or adjusting the sunshade curtain of the skylight. In addition, key knowledge elements can be automatically extracted, such as environmental perception (high temperature), function control items (air conditioner adjustment, window opening and closing, sun visor closing), and user expected responses (lowering the temperature, increasing the air volume, closing the sun visor, etc.), and virtual entities can be constructed accordingly for storage and invocation in the knowledge base. While constructing the virtual entities, this method uses the structured triple method for knowledge expression. For example, ("It's so hot", expected response, lower the air conditioner temperature), ("The light is too strong", expected response, close the sunshade curtain of the skylight), ("The air is not circulating", expected response, open the window / open the external circulation of the air conditioner), ("So tired", trigger condition, too long driving time / driver fatigue), and ("So tired", expected response, turn on the seat massage / play music), etc., to ensure the parsability and reusability of the structured data and achieve further complementation of the structured data.

[0111] The structured data generation method provided by the embodiments of this application can evaluate and supplement structured data based on the prior knowledge of human experts, correct and complete the data lacking logical association and the missing data, thereby effectively improving the accuracy and integrity of the structured data, ensuring that the structured data conforms to the real logic, determining the driver's personal data through the group structured database and the user intention determination model, generating a demand expression that conforms to the user characteristics, adaptively meeting the diverse needs of users, and providing effective data support for the multi-functional expansion of the intelligent cockpit.

[0112] Corresponding to the method in the above embodiment, Figure 8 The block diagram of the structured data generation device provided by the embodiments of this application is shown. For the convenience of description, only the parts related to the embodiments of this application are shown. Figure 8 The exemplary structured data generation device may be the execution subject of the structured data generation method provided in the foregoing Embodiment 1.

[0113] Referring to Figure 8 , the structured data generation device includes:

[0114] The cabin state heterogeneous data acquisition module 810 is used to acquire cabin state heterogeneous data;

[0115] The current state behavior data generation module 820 is used to obtain the current state data and the current behavior data according to the cabin state heterogeneous data;

[0116] The initial structured data generation module 830 is used to generate initial structured data according to the current state data, the current behavior data, the preset prompt text information, and the preset initial structured data generation model;

[0117] The target structured data generation module 840 is used to generate target structured data according to the initial structured data and the preset target structured data generation model.

[0118] In the structured data generation device provided by the embodiments of this application, the process of each module realizing its respective functions can be specifically referred to the description of Embodiment 1 shown above, and will not be elaborated here. Figure 1 Shown above

[0119] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of this application.

[0120] It should be understood that, as used in the specification of the present application and the appended claims, the term "comprising" indicates the presence of the described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their combinations.

[0121] It should also be understood that the term "and / or" as used in the specification of the present application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0122] As used in the specification of the present application and the appended claims, the term "if" may be construed, depending on the context, as "when" or "once" or "in response to determining" or "in response to detecting". Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be construed, depending on the context, as meaning "once determined" or "in response to determining" or "once detected [the described condition or event]" or "in response to detecting [the described condition or event]".

[0123] In addition, in the description of the specification of the present application and the appended claims, the terms "first", "second", "third", etc. are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance. It should also be understood that although the terms "first", "second", etc. are used in the text in some embodiments of the present application to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another. For example, the first table may be named the second table, and similarly, the second table may be named the first table, without departing from the scope of the various described embodiments. The first table and the second table are both tables, but they are not the same table.

[0124] Reference to "one embodiment" or "some embodiments" or the like described in the specification of the present application means that a particular feature, structure or characteristic described in connection with that embodiment is included in one or more embodiments of the present application. Thus, statements such as "in one embodiment", "in some embodiments", "in other some embodiments", "in still other embodiments", etc. that appear in different places in this specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in another way. The terms "comprising", "including", "having" and their variants all mean "including but not limited to", unless otherwise specifically emphasized in another way.

[0125] The structured data generation method provided by the embodiments of this application can be applied to terminal devices such as mobile phones, tablet computers, wearable devices, vehicle-mounted devices, augmented reality (AR) / virtual reality (VR) devices, laptop computers, ultra-mobile personal computers (UMPCs), netbooks, personal digital assistants (PDAs), etc. The embodiments of this application do not impose any restrictions on the specific types of terminal devices.

[0126] For example, the terminal device can be a station (STAION, ST) in a WLAN, a cellular phone, a cordless phone, a Session Initiation Protocol (SIP) phone, a Wireless Local Loop (WLL) station, a Personal Digital Assistant (PDA) device, a handheld device with wireless communication function, a computing device or other processing devices connected to a wireless modem, a vehicle-mounted device, a vehicle-to-internet terminal, a computer, a laptop computer, a handheld communication device, a handheld computing device, a satellite wireless device, a wireless modem card, a television set-top box (set top box, STB), a customer premise equipment (CPE), and / or other devices for communicating on a wireless system, as well as next-generation communication systems, such as mobile terminals in a 5G network or mobile terminals in a future evolved Public Land Mobile Network (PLMN) network, etc.

[0127] By way of example and not limitation, when the terminal device is a wearable device, the wearable device can also be a general term for devices developed by applying wearable technology to the intelligent design of daily wear, such as glasses, gloves, watches, clothing, and shoes. A wearable device is a portable device that is either worn directly on the body or integrated into the user's clothes or accessories. A wearable device is not just a hardware device, but also realizes powerful functions through software support, data interaction, and cloud interaction. Broadly speaking, wearable intelligent devices include those with complete functions and large sizes that can achieve complete or partial functions without relying on a smartphone, such as smart watches or smart glasses, as well as those that only focus on a certain type of application function and need to cooperate with other devices such as smartphones, such as various smart bracelets and smart jewelry for monitoring physical signs.

[0128] Figure 9It is a schematic structural diagram of a terminal device provided by an embodiment of the present application. As Figure 9 shown, the terminal device 9 of this embodiment includes: at least one processor 90 ( Figure 9 only one is shown in the figure), a memory 91, and a computer program 92 that can run on the processor 90 is stored in the memory 91. When the processor 90 executes the computer program 92, it implements the steps in the above-mentioned embodiments of various structured data generation methods, such as Figure 1 the steps S101 to S104 shown in the figure. Alternatively, when the processor 90 executes the computer program 92, it implements the functions of each module / unit in the above-mentioned device embodiments, such as Figure 8 the functions of the modules 810 to 840 shown in the figure.

[0129] The terminal device 9 may be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The terminal device may include, but is not limited to, a processor 90 and a memory 91. Those skilled in the art can understand that Figure 9 this is only an example of the terminal device 9 and does not constitute a limitation on the terminal device 9. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, the terminal device may further include an input and sending device, a network access device, a bus, etc.

[0130] The so-called processor 90 may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0131] In some embodiments, the memory 91 may be an internal storage unit of the terminal device 9, such as a hard disk or memory of the terminal device 9. The memory 91 may also be an external storage device of the terminal device 9, such as a plug-in hard disk equipped on the terminal device 9, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the memory 91 may also include both the internal storage unit and the external storage device of the terminal device 9. The memory 91 is used to store an operating system, application programs, a BootLoader, data, and other programs, such as the program code of the computer program. The memory 91 may also be used to temporarily store data that has been sent or will be sent.

[0132] In addition, in each embodiment of the present application, each functional unit may be integrated into one processing unit, may exist separately physically for each unit, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0133] The embodiment of the present application also provides a terminal device, which includes at least one memory, at least one processor, and a computer program stored in the at least one memory and executable on the at least one processor. When the processor executes the computer program, the terminal device implements the steps in any of the above method embodiments.

[0134] The embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it can implement the steps in the above method embodiments.

[0135] The embodiment of the present application provides a computer program product. When the computer program product runs on a terminal device, it enables the terminal device to implement the steps in the above method embodiments when executed.

[0136] When the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such understanding, to implement all or part of the processes in the above-described embodiment methods of the present application, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0137] In the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0138] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction 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 application.

[0139] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0140] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for generating structured data, characterized in that, Including: Obtain cabin state heterogeneous data; Based on the cabin state heterogeneous data, obtain current state data and current behavior data; Based on the current state data, current behavior data, preset prompt text information, and preset initial structured data generation model, generate initial structured data; Based on the initial structured data and preset target structured data generation model, generate target structured data; The step of obtaining current state data and current behavior data according to the cabin state heterogeneous data specifically includes: Perform attribute definition processing on the cabin state heterogeneous data to obtain state attribute data and behavior attribute data; Based on the state attribute data and preset semantic mapping function, obtain mapped behavior data; Based on the behavior attribute data and preset semantic backtracking function, obtain backtracked mapped state data; Based on the mapped behavior data and backtracked mapped state data, perform supplementary processing on the state attribute data and behavior attribute data to obtain state data intermediate variables and behavior data intermediate variables; Based on the state data intermediate variables, behavior data intermediate variables, and preset state matching function, obtain current state data and current behavior data; The step of obtaining current state data and current behavior data according to the state data intermediate variables, behavior data intermediate variables, and preset state matching function specifically includes: Perform filtering processing on the state data intermediate variables and the behavior data intermediate variables to obtain first state data intermediate variables and first behavior data intermediate variables; Perform time-series logic correction processing on the first state data intermediate variables and the first behavior data intermediate variables to obtain second state data intermediate variables and second behavior data intermediate variables; Perform missing data filling processing on the second state data intermediate variables and the second behavior data intermediate variables to obtain state data to be matched and current behavior data; Based on the current behavior data and preset state matching function, obtain multiple candidate state data; Screen the multiple candidate state data to obtain current state data.

2. The structured data generation method according to claim 1, characterized in that The step of generating initial structured data based on the current state data, current behavior data, preset prompt text information, and preset initial structured data generation model specifically includes: Obtain historical behavior data; Based on the current behavior data, historical behavior data, and preset behavior tracking function, obtain behavior sequence data; Based on the current state data and preset state detection function, obtain current state data; Based on the current behavior data, current state data, behavior sequence data, preset prompt text information, and preset initial structured data generation model, generate initial structured data.

3. The structured data generation method according to claim 1, wherein The step of generating target structured data based on the initial structured data and preset target structured data generation model specifically includes: Perform encoding processing on the initial structured data to obtain initial structured data encoding information; Based on the initial structured data encoding information and preset target structured data generation model, generate target structured representation data; Decode the target structured representation data to obtain the target structured data.

4. The structured data generation method according to claim 1, wherein After the step of generating the target structured data according to the initial structured data and the preset target structured data generation model, the following steps are further included: Calculate the semantic similarity of each target structured data according to the preset similarity calculation function and the preset similarity calculation weight to obtain the structured data semantic similarity; When the structured data semantic similarity is greater than the preset semantic similarity threshold, generate a structured data clustering cluster according to the target structured data corresponding to the structured data semantic similarity; Perform normalization processing on the structured data clustering cluster to obtain the standard structured data.

5. The structured data generation method according to claim 4, wherein After the step of performing normalization processing on the structured data clustering cluster to obtain the standard structured data, the following steps are further included: Obtain the logical correction information and the true value annotation information; Perform verification and correction processing on the standard structured data according to the logical correction information and the true value annotation information to obtain the optimized structured data; Obtain the user intention information according to the optimized structured data, the preset group structured database, and the preset user intention determination model.

6. A structured data generation device, characterized in that, Include: A cabin state heterogeneous data acquisition module for acquiring cabin state heterogeneous data; A current state behavior data generation module for obtaining the current state data and the current behavior data according to the cabin state heterogeneous data; An initial structured data generation module for generating initial structured data according to the current state data, the current behavior data, the preset prompt text information, and the preset initial structured data generation model; A target structured data generation module for generating target structured data according to the initial structured data and the preset target structured data generation model; The step of obtaining the current state data and the current behavior data according to the cabin state heterogeneous data specifically includes: Perform attribute definition processing on the cabin state heterogeneous data to obtain state attribute data and behavior attribute data; Obtain the mapped behavior data according to the state attribute data and the preset semantic mapping function; Obtain the backtracked mapped state data according to the behavior attribute data and the preset semantic backtracking function; Perform supplementary processing on the state attribute data and the behavior attribute data according to the mapped behavior data and the backtracked mapped state data to obtain the state data intermediate variable and the behavior data intermediate variable; Obtain the current state data and the current behavior data according to the state data intermediate variable, the behavior data intermediate variable, and the preset state matching function; The step of obtaining the current state data and the current behavior data according to the state data intermediate variable, the behavior data intermediate variable, and the preset state matching function specifically includes: Perform filtering processing on the state data intermediate variable and the behavior data intermediate variable to obtain the first state data intermediate variable and the first behavior data intermediate variable; Perform sequential logic correction processing on the first state data intermediate variable and the first behavior data intermediate variable to obtain a second state data intermediate variable and a second behavior data intermediate variable; Perform missing data completion processing on the second state data intermediate variable and the second behavior data intermediate variable to obtain the to-be-matched state data and the current behavior data; Obtain multiple candidate state data according to the current behavior data and a preset state matching function; Screen the multiple candidate state data to obtain the current state data.

7. A terminal device, characterized in that, The terminal device includes a memory and a processor. A computer program that can run on the processor is stored on the memory. When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images