Search method, system and storage medium for vehicle-mounted system

Abstract

An embodiment of the present application provides a search method, system, and storage medium for an in-vehicle system, wherein the search method includes the following steps: responding to a search instruction, obtaining environmental features and user features, the search instruction including: initial text features; the environmental features are used to characterize the real-time status of the vehicle; the user features are used to characterize the user's historical behavior; performing semantic generalization processing on the initial text features to generate a text feature collection; fusing the text feature collection, the environmental features, and the user features into a fusion vector; performing a first recall based on the text feature collection to generate a first candidate set; performing a second recall based on the fusion vector to generate a second candidate set; and fusing and sorting the first candidate set and the second candidate set according to a preset fusion weight strategy to generate a search collection.

Description

Technical Field

[0001] The present application relates to the technical field of vehicle-mounted systems, and in particular to a search method, system, and storage medium for vehicle-mounted systems. Background Art

[0002] With the development of intelligent connected vehicles, in-vehicle systems are becoming increasingly important interactive portals for users to access information such as navigation, media, and services. In-vehicle search, as one of its core functions, directly impacts the efficiency and experience of information acquisition while driving. Existing in-vehicle systems often utilize traditional information retrieval architectures based on keyword matching. Typical examples include the Elasticsearch (ES) system, which uses an inverted index and combines classic statistical models such as BM25 and TF-IDF for search ranking.

[0003] However, users have limited attention and a short interaction window while driving, and traditional information retrieval architectures applied to in-vehicle systems have problems such as low recall rate and poor result relevance. Summary of the Invention

[0004] Embodiments of the present application provide a search method, system, and storage medium for an in-vehicle system, which can improve the recall rate and result relevance of the search system for the in-vehicle system.

[0005] In a first aspect, this embodiment provides a search method for an in-vehicle system, comprising the following steps:

[0006] In response to a search instruction, obtaining environmental features and user features, the search instruction including: initial text features; the environmental features are used to characterize the real-time state of the vehicle; the user features are used to characterize the user's historical behavior;

[0007] Performing semantic generalization processing on the initial text features to generate a text feature collection;

[0008] fusing the text feature collection, the environment feature, and the user feature into a fusion vector;

[0009] Perform a first recall based on the text feature collection to generate a first candidate set;

[0010] Perform a second recall based on the fusion vector to generate a second candidate set;

[0011] According to a preset fusion weight strategy, the first candidate set and the second candidate set are fused and sorted to generate a search set.

[0012] In some embodiments, performing semantic generalization processing on the initial text features to generate a text feature collection includes:

[0013] Performing intent recognition on the initial text features through a prompt word model to generate intent information, wherein the intent information includes basic intent and topic intent;

[0014] Performing synonym expansion on the basic intent to generate a first synonym collection;

[0015] Performing synonym expansion on the subject intent to generate a second synonym collection;

[0016] The text feature collection is generated based on the first synonym collection and the second synonym collection.

[0017] In some embodiments, performing synonym expansion on the subject intent to generate a second synonym set includes:

[0018] generating a recall prompt word according to the user characteristics and the subject intention;

[0019] The recall prompt words are input into a preset expansion model to generate a second synonym collection.

[0020] In some embodiments, performing synonym expansion on the subject intent to generate a second synonym set includes:

[0021] Inputting the subject intention into a preset expansion model to generate a collection of common synonyms;

[0022] Calculating the correlation between the synonyms in the general synonym collection and the user features;

[0023] The second synonym collection is selected from the general synonym collection based on the relevance; the relevance of the synonyms in the second synonym collection is greater than the relevance of other synonyms in the general synonym collection.

[0024] In some embodiments, performing a first recall based on the text feature collection to generate a first candidate set includes:

[0025] limiting the recall scope for the first synonym set;

[0026] Keyword matching is performed on the second synonym set within the recall range to generate the first candidate set.

[0027] In some embodiments, fusing the text feature collection, the environment feature, and the user feature into a fusion vector includes:

[0028] Converting the text feature collection into a text feature vector, converting the environment feature into an environment feature vector, and converting the user feature into a user feature vector;

[0029] The text feature vector, the environment feature vector and the user feature vector are weightedly fused to generate a fusion vector.

[0030] In some embodiments, performing a second recall based on the fusion vector to generate a second candidate set includes:

[0031] Calculating the similarity between the fusion vector and each resource vector in a pre-built resource vector library;

[0032] Selecting a target resource vector having a similarity greater than a preset similarity from the resource vector library;

[0033] The second candidate set is generated based on the resources corresponding to the target resource vector.

[0034] In some embodiments, the environmental features include: vehicle state parameters, interaction timing parameters; the first candidate set includes multiple candidate resources, each candidate resource corresponds to a first score, and the second candidate set includes multiple candidate resources, each candidate resource corresponds to a second score;

[0035] The step of fusing and sorting the first candidate set and the second candidate set according to a preset fusion weight strategy to generate a search set includes:

[0036] Normalizing the first score of each candidate resource in the first candidate set and the second score of each candidate resource in the second candidate set;

[0037] Obtaining the vehicle state parameters and interaction timing parameters, and confirming the current interaction reliability;

[0038] When the interaction reliability is lower than a preset threshold, reducing the weight of each candidate resource in the first candidate set, increasing the weight of each candidate resource in the second candidate set, and generating a corresponding fusion score based on the weight of each candidate resource in the first candidate set and the first score after normalization, and generating a corresponding fusion score based on the weight of each candidate resource in the second candidate set and the second score after normalization; or,

[0039] When the interaction reliability is higher than or equal to a preset threshold, the weight of each candidate resource in the first candidate set is increased, the weight of each candidate resource in the second candidate set is decreased, and a corresponding fusion score is generated based on the weight of each candidate resource in the first candidate set and the normalized first score, and a corresponding fusion score is generated based on the weight of each candidate resource in the second candidate set and the normalized second score;

[0040] The candidate resources in the first candidate set and the second candidate set are uniformly sorted based on the fusion score to generate the search set.

[0041] In a second aspect, this embodiment further provides a search system for an in-vehicle system, including:

[0042] A response module is used to respond to a search instruction and obtain environmental features and user features, wherein the search instruction includes: initial text features; the environmental features are used to represent the real-time status of the vehicle; and the user features are used to represent the user's historical behavior;

[0043] A semantic processing module, configured to perform semantic generalization processing on the initial text features to generate a text feature collection;

[0044] A vector conversion module, configured to fuse the text feature collection, the environment feature, and the user feature into a fusion vector;

[0045] A first recall module, configured to perform a first recall based on the text feature collection to generate a first candidate set;

[0046] A second recall module, configured to perform a second recall based on the fusion vector to generate a second candidate set;

[0047] A fusion module is used to fuse and sort the first candidate set and the second candidate set according to a preset fusion weight strategy to generate a search set.

[0048] In a third aspect, this embodiment further provides a computer-readable storage medium on which a computer program is stored. The computer program is loaded by a processor to execute the steps of the search method for a vehicle-mounted system in any one of the above embodiments. Beneficial effects

[0049] The search method for an in-vehicle system disclosed in this application proposes a dual-path recall method. By performing semantic generalization processing on the search instructions and combining the user's historical characteristics with the vehicle's current environmental characteristics, keyword recall and fusion vector recall are performed respectively. A weight control mechanism based on the interactive reliability score is introduced to perform weighted fusion sorting on the recall results, thereby achieving highly relevant, robust and responsive search result output in different driving scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0051] Figure 1 This is a flowchart of a search method for an in-vehicle system provided in some embodiments of the present application. Figure 1 ;

[0052] Figure 2 This is a flowchart of a search method for an in-vehicle system provided in some embodiments of the present application. Figure 2 ;

[0053] Figure 3 This is a flowchart of a search method for an in-vehicle system provided in some embodiments of the present application. Figure 3 . DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0055] In the description of the present application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the said features. In the description of the present application, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0056] “A and / or B” includes the following three combinations: A only, B only, and a combination of A and B.

[0057] The use of "suitable for" or "configured to" in this application is intended to be open and inclusive language, and does not exclude devices that are adapted or configured to perform additional tasks or steps. In addition, the use of "based on" is intended to be open and inclusive, as a process, step, calculation, or other action that is "based on" one or more stated conditions or values ​​may, in practice, be based on additional conditions or values ​​beyond those stated.

[0058] In this application, the word "exemplary" is used to mean "serving as an example, illustration, or illustration." Any embodiment described in this application as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments. The following description is given to enable any person skilled in the art to implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that one of ordinary skill in the art can recognize that the present application can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid obscuring the description of the present application with unnecessary details. Therefore, the present application is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in this application.

[0059] As described in the background, traditional in-vehicle search architectures face technical bottlenecks. Specifically, due to the limited attention span of users while driving, the resulting search commands are often inaccurate and contain limited information. When users express themselves in non-standard, colloquial language, or when search terms lack a clear match with the target content, traditional in-vehicle search architectures are prone to issues such as incomplete recall and irrelevant results. These shortcomings not only severely diminish the search experience but also distract the driver, increasing driving risks.

[0060] Reference Figure 1 , an embodiment of the present application provides a search method for an in-vehicle system, comprising:

[0061] S100, responding to a search instruction, obtaining environmental features and user features, the search instruction including: initial text features; environmental features used to characterize the real-time state of the vehicle; user features used to characterize the user's historical behavior;

[0062] S200, performing semantic generalization processing on the initial text features to generate a text feature collection;

[0063] S300, fusing the text feature collection, the environment feature, and the user feature into a fusion vector;

[0064] S400: Perform a first recall based on the text feature collection to generate a first candidate set;

[0065] S500, performing a second recall based on the fusion vector to generate a second candidate set;

[0066] S600: According to a preset fusion weight strategy, the first candidate set and the second candidate set are fused and sorted to generate a search set.

[0067] In step S100, in response to a search instruction, environmental features and user features are acquired, where the search instruction includes: initial text features; environmental features are used to characterize the real-time status of the vehicle; and user features are used to characterize the user's historical behavior.

[0068] In an embodiment of the present application, the search instruction can be input through text input or voice input. The search system of the vehicle system can uniformly parse and process different forms of input to obtain the initial text features included in the search instruction.

[0069] After receiving and responding to search commands, the in-vehicle search system can obtain environmental and user characteristics. Environmental characteristics include, but are not limited to, vehicle status parameters, interaction timing parameters, and time parameters. Vehicle status parameters include, but are not limited to, the vehicle's current location, speed, and acceleration. Time parameters include the current time. Interaction timing parameters include peak interaction periods and low interaction periods. Peak interaction periods indicate periods when the user's attention is focused and the user is idle and available for interaction. Low interaction periods indicate periods when attention is limited and driving tasks are prioritized. For example, peak interaction periods can be during the initial vehicle entry phase (the first 0-30 seconds after the vehicle is unlocked and started) and when the vehicle is stationary (for 10 seconds or longer). Low interaction periods can be during driving. User characteristics include, but are not limited to, historical user behavior records, user interest tags, user profile information, and historical geographic behavior characteristics.

[0070] In step S200, semantic generalization processing is performed on the initial text features to generate a text feature collection.

[0071] Reference Figure 2 In the embodiment of the present application, a text feature collection is generated by the following steps, specifically:

[0072] S201, performing intent recognition on initial text features through a prompt word model to generate intent information, which includes basic intent and topic intent;

[0073] S202, performing synonym expansion on the basic intent to generate a first synonym collection;

[0074] S203, performing synonym expansion on the subject intent to generate a second synonym collection;

[0075] S204: Generate a text feature collection based on the first synonym collection and the second synonym collection.

[0076] Specifically, the base intent indicates the type of action the user intends to perform, typically expressed as a verb. For example, if a user inputs "I want to listen to Zhou's quietness," the base intent is "listen." After synonym expansion for "listen," the first synonym set includes, but is not limited to, ["play," "listen," "come," "play a bit," "play a song"]. The topic intent indicates the subject matter of the content the user is interested in, typically expressed as a noun or phrase. For example, if a user inputs "I want to listen to Zhou's quietness," the topic intent is "Zhou's quietness." The second synonym set includes, but is not limited to, ["Zhou's quietness," "quiet," "lyrical," "quiet songs," "Zhou," "popular songs"]. The generated text feature set can be a feature expression that cross-combines the first and second synonym sets to generate a phrase structure, such as ["play quietness," "play Zhou," "play lyrical songs," "come to some quiet," "listen to Zhou," "come to some pop songs"]. Alternatively, the structures of the first and second synonym sets can be retained separately, or both forms of text feature sets can coexist.

[0077] It is worth mentioning that in the embodiment of the present application, the synonyms contained in the first synonym set and the second synonym set have different granularities. For example, in the second synonym set, the fine-grained expression (closer to the user's specific expression, retaining complete directionality) is "Zhou Moumou's quietness"; the medium-grained expression (abstracting some semantic keywords) is "quiet", "lyrical", "quiet song", "Zhou Moumou"; the coarse-grained expression (generalized to semantic categories or content styles) is "popular songs".

[0078] Furthermore, when generating the second synonym collection, the second synonym collection may also be generated based on both user characteristics and topic intent. In some embodiments, the second synonym collection is generated by the following steps:

[0079] Generate recall prompt words based on user characteristics and topic intent;

[0080] The recall prompt words are input into the preset expansion model to generate a second synonym collection.

[0081] Specifically, if the initial text feature is "I want to listen to light music," the subject intent is "light music," the user features include user interest tags such as jazz, piano, and serenade, the user's historical behavior records include "Bill Evans" and "cafe jazz," and the user profile information is 30–40 years old, then based on the subject intent and the user features, recall prompt words can be generated [User preferences: jazz, piano, serenade; Historical behavior: repeatedly played Bill Evans, cafe jazz; Current search intent: light music; Based on the above information, please expand keywords related to "light music"]. These recall prompt words can then be input into a preset expansion model to generate a second synonym collection. The expansion model can be a lightweight generative model that supports prompt word input.

[0082] In some other embodiments, the second synonym collection is generated by the following steps:

[0083] Input the subject intent into the preset expansion model to generate a collection of common synonyms;

[0084] Calculate the relevance between synonyms in the common synonym collection and user features;

[0085] A second synonym collection is selected from the general synonym collection based on the relevance; the relevance of the synonyms in the second synonym collection is greater than the relevance of other synonyms in the general synonym collection.

[0086] Specifically, a pre-defined expansion model (such as BERT, GPT, or a dictionary system) was used to perform universal synonym expansion on the term "light music," generating a set of universal synonyms: ["relaxing music," "instrumental music," "background music," "piano music," "jazz," "serenade," "lo-fi," "sleep melodies"]. The user profile included the following user interest tags: jazz, piano, and serenade; the user's historical behavior records included "Bill Evans" and "nighttime lo-fi"; and the user profile was 30–40 years old. For each universal synonym, the semantic relevance between the user interest tag and historical behavior records was calculated (using vector cosine similarity or rule-based scoring). The resulting second set of synonyms was ["piano music," "jazz," "serenade," and "lo-fi"].

[0087] In the above two embodiments of generating the second synonym set, user characteristics are introduced into the synonym expansion link of the topic intention, which can improve the accuracy of the first candidate set during the first recall. At the same time, when the user inputs a search instruction, it often has ambiguity and incompleteness (such as "something light"), or there are non-standard spoken expressions. Through the fusion of generalization and personalization, this solution can still infer and cover the real intention even when the user input is not standardized. For example: "Listen to light music" can be automatically associated with "jazz serenade", "piano relaxation music", "Lo-fi background", significantly improving the robustness of the recall and the consistency of the user experience.

[0088] In step S400, a first recall is performed based on the text feature set to generate a first candidate set. In the embodiments of the present application, the first candidate set is specifically generated through the following steps:

[0089] Limit the recall scope for the first synonym set;

[0090] Perform keyword matching on the second synonym set within the recall scope to generate a first candidate set.

[0091] Specifically, when the above-mentioned publicly generated text feature set was mentioned, it was stated that the text feature set can respectively retain the structures of the first synonym set and the second synonym set. In this embodiment, the text feature set that retains the structures of the first synonym set and the second synonym set is used. First, the first synonym set generated by the basic intention recognition is used as a behavior limiting condition to screen the index fields or semantic tags in the resource library to limit the recall scope. Example: When the basic intention is "listen", and the first synonym set is ["play", "listen", "come with", "play for a while"], then when the system recalls resources, it only retrieves media resources (such as songs, albums, radio stations), rather than navigation, question-and-answer, or service resources. Subsequently, within the above-mentioned limited recall scope, keyword matching is performed on the synonyms in the second synonym set, and candidate resources with high relevance are screened out to form a first candidate set. Example: The second synonym set is ["Zhou XX", "quiet", "lyrical", "serenade", "jazz"], and based on the exact match or fuzzy match (Fuzzy) of the ES (Elasticsearch) inverted index, and using BM25, TF-IDF or a customized scoring mechanism to score and rank the candidate resources, finally, the candidate resources with scores higher than the preset threshold will be included in the first candidate set for subsequent fusion sorting.

[0092] In step S300, the text feature set, environmental features, and user features are fused into a fusion vector; in step S500, a second recall is performed based on the fusion vector to generate a second candidate set. Specifically, in the embodiments of the present application, the fusion vector is formed through the following steps:

[0093] Convert the text feature set into a text feature vector, convert the environment feature into an environment feature vector, and convert the user feature into a user feature vector;

[0094] The text feature vector, the environment feature vector and the user feature vector are weightedly fused to generate a fusion vector.

[0095] Specifically, a collection of text features can be used to generate a corresponding text feature vector V-text using a preset text encoding model (such as lightweight semantic models such as BERT, SBERT, and SimCSE). Environmental features (including vehicle state parameters, interaction timing parameters, time parameters, etc.) can be encoded as numerical features and input into an embedding model or a multidimensional normalization method can be used to generate an environmental feature vector V-env. User features (including user historical behavior records, user interest tags, user profile information, historical geographic behavior features, etc.) can be converted into a fixed-dimensional vector representation. Conversion methods include but are not limited to methods based on label vector mapping, behavior sequence embedding, and profile feature encoding. The user feature vector is represented by V-user.

[0096] The text feature vector, environment feature vector, and user feature vector are weightedly fused to generate a fusion vector. The dimensions of the three vectors may be different. The text feature vector V-text may be 128-dimensional, the environment feature vector V-env may be 64-dimensional, and the user feature vector V-user may be 64-dimensional. It is necessary to first map the three vectors to the same dimension through linear transformation, and then use weights α, β, and γ to perform weighted summation to obtain the fusion vector V-fusion, V-fusion = α* V-text + β* V-env + γ* V-user, where α, β, and γ can be preset fixed values ​​(such as 0.6, 0.15, and 0.25).

[0097] Traditional in-car search systems rely solely on text vectors, which can easily miss the user's true intent. However, in the secondary recall phase, incorporating contextual features can address ambiguous expressions or non-standard sentences. Furthermore, even if different users enter the same text, the search system can retrieve different resources based on their preferences. For example, when asked "play light music," some people prefer "piano music" while others prefer "lo-fi serenades."

[0098] In this embodiment of the present application, the second candidate set is generated by the following steps:

[0099] Calculate the similarity between the fusion vector and each resource vector in the pre-built resource vector library;

[0100] Filter target resource vectors whose similarity is greater than a preset similarity in the resource vector library;

[0101] A second candidate set is generated based on the resources corresponding to the target resource vector.

[0102] In a pre-built resource vector library, all target resources (such as songs, albums, news, radio programs, etc.) are uniformly encoded, and each resource content (including title, tag, description field, etc.) is converted into a fixed-dimensional resource semantic vector. When calculating the similarity between the fusion vector and the individual resource vectors in the pre-built resource vector library, cosine similarity can be used. Compared to the first recall, the second recall can capture deeper, implicitly associated semantics by expressing the text features, environmental features, and user features after synonym expansion. It can effectively complete the search results when the user input is incomplete, keywords are missing, or the expression is ambiguous.

[0103] Reference Figure 3 In step S600, the first candidate set and the second candidate set are fused and sorted according to a preset fusion weight strategy to generate a search set.

[0104] Specifically, the first candidate set includes multiple candidate resources, each candidate resource corresponds to a first score, and the second candidate set includes multiple candidate resources, each candidate resource corresponds to a second score. The embodiment of the present application generates a search collection based on the following steps:

[0105] S601: Normalize the first score of each candidate resource in the first candidate set and the second score of each candidate resource in the second candidate set;

[0106] S602: Acquire vehicle status parameters and interaction timing parameters, and confirm current interaction reliability;

[0107] S603. When the interaction reliability is lower than a preset threshold, reduce the weight of each candidate resource in the first candidate set, increase the weight of each candidate resource in the second candidate set, and generate a corresponding fusion score based on the weight of each candidate resource in the first candidate set and the normalized first score, and generate a corresponding fusion score based on the weight of each candidate resource in the second candidate set and the normalized second score; or

[0108] S604: When the interaction reliability is higher than or equal to a preset threshold, increase the weight of each candidate resource in the first candidate set, decrease the weight of each candidate resource in the second candidate set, and generate a corresponding fusion score based on the weight of each candidate resource in the first candidate set and the normalized first score, and generate a corresponding fusion score based on the weight of each candidate resource in the second candidate set and the normalized second score;

[0109] S605 : Sort the candidate resources in the first candidate set and the second candidate set uniformly based on the fusion score to generate a search set.

[0110] Specifically, the candidate resources in the first candidate set and the candidate resources in the second candidate set are recalled via different recall paths. By normalizing the first score of each candidate resource in the first candidate set and the second score of each candidate resource in the second candidate set, differences in scoring scales across the different recall paths can be eliminated, ensuring the effectiveness and comparability of subsequent weighted calculations. Examples of such normalization methods include min-max normalization, Z-score normalization, and Softmax transformation.

[0111] Furthermore, vehicle state parameters and interaction timing parameters are obtained as interaction environment evaluation indicators. Vehicle state parameters include but are not limited to the vehicle's current location information, current speed, current acceleration, etc.; interaction timing parameters include: interaction peak period and interaction low-peak period. The interaction peak period is used to indicate the stage when the user's attention is relatively focused and is in an idle or interactive state. The interaction low-peak period is used to indicate the stage when the user's attention is limited and the driving task has a higher priority. For example, the interaction peak period can be: the initial stage of getting in the vehicle (the first 0-30 seconds after the vehicle is unlocked and started), the period when the vehicle is stationary (the vehicle is stationary for ≥10 seconds), etc.; the interaction low-peak period can be: during driving.

[0112] Specifically, the embodiment of the present application uses preset rules to score interaction reliability. Regarding vehicle state parameters, the current vehicle speed is obtained as follows: stationary (0 km / h) corresponds to 1 point, low speed (<30 km / h) corresponds to 0.7 points, and high speed (≥60 km / h) corresponds to 0.2 points. The current acceleration is obtained as follows: close to 0 (indicating stable vehicle operation) corresponds to 1 point, and frequent speed changes (indicating that the vehicle may be accelerating / braking suddenly) corresponds to 0.4. Through linear weighting, the weight of the current vehicle speed is 0.6, and the weight of the current acceleration is 0.4. The vehicle state parameter score R-vehicle used to evaluate interaction reliability can be determined by R-vehicle = 0.6 × f(speed) + 0.4 × f(acceleration), where f(speed) represents the score of the current vehicle speed and f(acceleration) represents the score of the current acceleration. Regarding interaction timing parameters, peak interaction period corresponds to 1 point, and low interaction period corresponds to 0.5 points. The comprehensive score R-interact of the current interaction reliability can be determined by R-interact = 0.5*R-vehicle + 0.5*R-context, where R-context represents the score of the interaction timing parameter. In some embodiments, the preset threshold is 0.6. When the interaction reliability score is lower than the preset threshold, the system believes that the initial text features may be incomplete or noisy, so the dependence on the first candidate set obtained by the first recall should be weakened, and the weight of the second candidate set obtained by the second recall should be increased. At this time, the weight of the first candidate set is lower than the weight of the second candidate set. When the interaction reliability score is higher than or equal to the preset threshold, the system gives priority to trusting the first candidate set obtained based on the first recall, enhances the influence of the first candidate set, and appropriately weakens the reasoning recall of the second candidate set. At this time, the weight of the first candidate set is higher than the weight of the second candidate set.

[0113] The first candidate set is recalled through keyword matching, offering fast response and high accuracy, but it requires high clarity of user expression. For example, the first candidate set has a high hit rate for expressions like "listen to Zhou XX," but a low hit rate for expressions like "I want something comfortable." The second candidate set is recalled through fused vectors, adapting to ambiguous expressions and providing strong semantic reasoning, but may suffer from insufficient accuracy or an overly broad recall range. In real-world vehicle scenarios, search quality fluctuates due to factors such as driving conditions, voice interference, and timing. By adjusting the weights of the first and second candidate sets during fusion, this embodiment of the present application enhances the weight of the first candidate set when the user's expression is clear and the system is interacting stably (e.g., when the vehicle is parked), improving the accuracy and intuitive hit rate of the recall results. It also enhances the weight of the second candidate set when the user's voice input is incomplete, ambiguous, or misidentified (e.g., on highways). This allows the system to better "understand" the user's underlying intent, thereby reducing the number of no results or incorrect results. Furthermore, when interaction reliability is high, this embodiment prioritizes lightweight keyword recall, reducing average response latency, enhancing interaction fluency, and improving system response efficiency.

[0114] In some embodiments, in step S605, the candidate resources in the first and second candidate sets are uniformly sorted based on the fusion scores. Before generating the search set, a similarity matching judgment is performed on the candidate resources in the first and second candidate sets. If the same candidate resource is determined to exist, the candidate resource is considered a duplicate resource, only one result is retained, and its scores in the two recall paths are fused. The scores of the fused duplicate candidate resources in the two recall paths can be calculated using a weighted summation method, and a redundant matching weighting term can be introduced to enhance the expression of their relevance.

[0115] For example, the resource "Zhou Moumou·Anjing" is matched in the first candidate set, with a fusion score of 0.81. The resource is also matched in the second candidate set, with a fusion score of 0.74. Based on the weight of the fusion score, the fusion score of the candidate resource can be fused, and the fusion score for the candidate resource can be recalculated. For example, if the weight of the candidate resource in the first candidate set is 0.6 and the weight of the candidate resource in the second candidate set is 0.4, and the redundant matching weighting term is 0.03, the recalculated fusion score of the candidate resource, score_fused, is: score_fused = 0.6×0.81 + 0.4×0.74 + 0.03 ≈ 0.786.

[0116] In summary, the search method for an in-vehicle system disclosed in this application proposes a dual-path recall method, which performs semantic generalization on the search instructions, combines the user's historical characteristics with the vehicle's current environmental characteristics, performs keyword recall and fusion vector recall respectively, and introduces a weight control mechanism based on the interactive reliability score to perform weighted fusion sorting on the recall results, thereby achieving highly relevant, highly robust and highly responsive search result output in different driving scenarios.

[0117] An embodiment of the present application also discloses a search system for an in-vehicle system, comprising: a response module, a semantic processing module, a vector conversion module, a first recall module, a second recall module, and a fusion module.

[0118] The response module is used to respond to search instructions and obtain environmental features and user features. The search instructions include: initial text features; environmental features are used to characterize the real-time status of the vehicle; and user features are used to characterize the user's historical behavior. The semantic processing module is used to perform semantic generalization processing on the initial text features to generate a text feature set. The vector conversion module is used to fuse the text feature collection, environmental features, and user features into a fusion vector. The first recall module is used to perform a first recall based on the text feature collection to generate a first candidate set. The second recall module is used to perform a second recall based on the fusion vector to generate a second candidate set. The fusion module is used to fuse and sort the first candidate set and the second candidate set according to a preset fusion weight strategy to generate a search collection.

[0119] The present application also discloses a computer-readable storage medium storing a computer program, which is loaded by a processor to execute the steps of any of the aforementioned methods for searching an in-vehicle system. In the present application, the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0120] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0121] The above is a detailed introduction to a washing machine and its control method, control system, and computer-readable storage medium provided in the embodiments of the present application. Specific examples are used herein to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and core idea of ​​the present application. At the same time, for those skilled in the art, based on the ideas of the present application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present application.

Claims

1. A search method for a vehicle-mounted system, characterized in that: include: In response to a search instruction, environmental features and user features are obtained, wherein the search instruction includes: initial text features; the environmental features are used to characterize the real-time state of the vehicle; the environmental features include: vehicle state parameters and interaction timing parameters; and the user features are used to characterize the user's historical behavior; Performing semantic generalization processing on the initial text features to generate a text feature collection; fusing the text feature collection, the environment feature, and the user feature into a fusion vector; Performing a first recall based on the text feature collection to generate a first candidate set, where the first candidate set includes multiple candidate resources, each candidate resource corresponding to a first score; Performing a second recall based on the fusion vector to generate a second candidate set, where the second candidate set includes multiple candidate resources, each candidate resource corresponding to a second score; Normalizing the first score of each candidate resource in the first candidate set and the second score of each candidate resource in the second candidate set; Obtaining the vehicle state parameters and interaction timing parameters, and confirming the current interaction reliability; When the interaction reliability is lower than a preset threshold, reducing the weight of each candidate resource in the first candidate set, increasing the weight of each candidate resource in the second candidate set, and generating a corresponding fusion score based on the weight of each candidate resource in the first candidate set and the first score after normalization, and generating a corresponding fusion score based on the weight of each candidate resource in the second candidate set and the second score after normalization; or, When the interaction reliability is higher than or equal to a preset threshold, the weight of each candidate resource in the first candidate set is increased, the weight of each candidate resource in the second candidate set is decreased, and a corresponding fusion score is generated based on the weight of each candidate resource in the first candidate set and the normalized first score, and a corresponding fusion score is generated based on the weight of each candidate resource in the second candidate set and the normalized second score; The candidate resources in the first candidate set and the second candidate set are uniformly sorted based on the fusion score to generate a search set.

2. The search method for an in-vehicle system according to claim 1, characterized in that: The performing semantic generalization processing on the initial text features to generate a text feature collection includes: Performing intent recognition on the initial text features through a prompt word model to generate intent information, wherein the intent information includes basic intent and topic intent; Performing synonym expansion on the basic intent to generate a first synonym collection; Performing synonym expansion on the subject intent to generate a second synonym collection; The text feature collection is generated based on the first synonym collection and the second synonym collection.

3. The search method for an in-vehicle system according to claim 2, characterized in that: The performing synonym expansion on the subject intent to generate a second synonym collection includes: generating a recall prompt word according to the user characteristics and the subject intention; The recall prompt words are input into a preset expansion model to generate a second synonym collection.

4. The search method for an in-vehicle system according to claim 2, characterized in that: The performing synonym expansion on the subject intent to generate a second synonym collection includes: Inputting the subject intention into a preset expansion model to generate a collection of common synonyms; Calculating the correlation between the synonyms in the general synonym collection and the user features; The second synonym collection is selected from the general synonym collection based on the relevance; the relevance of the synonyms in the second synonym collection is greater than the relevance of other synonyms in the general synonym collection.

5. The search method for an in-vehicle system according to claim 2, characterized in that: The performing a first recall based on the text feature collection to generate a first candidate set includes: limiting the recall scope for the first synonym set; Keyword matching is performed on the second synonym set within the recall range to generate the first candidate set.

6. The search method for an in-vehicle system according to claim 1, characterized in that: The fusing the text feature collection, the environment feature, and the user feature into a fusion vector comprises: Converting the text feature collection into a text feature vector, converting the environment feature into an environment feature vector, and converting the user feature into a user feature vector; The text feature vector, the environment feature vector and the user feature vector are weightedly fused to generate a fusion vector.

7. The search method for an in-vehicle system according to claim 6, characterized in that: The performing a second recall based on the fusion vector to generate a second candidate set includes: Calculating the similarity between the fusion vector and each resource vector in a pre-built resource vector library; Selecting a target resource vector having a similarity greater than a preset similarity from the resource vector library; The second candidate set is generated based on the resources corresponding to the target resource vector.

8. A search system for an in-vehicle system, characterized in that: For implementing the search method for an in-vehicle system according to any one of claims 1 to 7, the search system for an in-vehicle system comprises: A response module is used to respond to a search instruction and obtain environmental features and user features, wherein the search instruction includes: initial text features; the environmental features are used to represent the real-time status of the vehicle; and the user features are used to represent the user's historical behavior; A semantic processing module, configured to perform semantic generalization processing on the initial text features to generate a text feature collection; A vector conversion module, configured to fuse the text feature collection, the environment feature, and the user feature into a fusion vector; A first recall module, configured to perform a first recall based on the text feature collection to generate a first candidate set; A second recall module, configured to perform a second recall based on the fusion vector to generate a second candidate set; A fusion module is used to fuse and sort the first candidate set and the second candidate set according to a preset fusion weight strategy to generate a search set.

9. A computer-readable storage medium, characterized in that A computer program is stored thereon, and the computer program is loaded by a processor to execute the steps of the search method for a vehicle-mounted system according to any one of claims 1 to 7.

Images