Film and television resource searching method, display device, storage medium and program product

Abstract

The invention discloses a film and television resource searching method, a display device, a storage medium and a program product, and relates to the technical field of artificial intelligence. Extracting a first keyword related to the film and television resource attribute from the natural language query, and executing first film and television resource retrieval in a film and television resource library based on the first keyword to obtain a first retrieval result; when the first keyword is extracted, natural language query is input into the first large model for semantic analysis, a target retrieval instruction is generated, and the target retrieval instruction is structured data containing film and television resource attribute fields and corresponding values of the film and television resource attribute fields; executing second film and television resource retrieval in a film and television resource library based on the target retrieval instruction to obtain a second retrieval result; and determining a target film and television resource according to the first retrieval result and the second retrieval result. The intelligent film and television resource searching method and device achieve the intelligent film and television resource searching effect considering the response speed and the searching precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and particularly relates to a movie and television resource search method, a display device, a storage medium and a computer program product. BACKGROUND

[0002] In the prior art, it is difficult to balance speed and accuracy in improving the intelligent level of movie and television resource search. Specifically, a conventional technology calls a large language model to analyze a user's question and generate a preliminary search result; thereafter, the preliminary search result is post-processed, such as expanding, comparing and re-scoring the characters and themes in the result, and then combined with user preferences to sort and obtain a final search result. Although the subsequent cumbersome processing process can make the final search result more accurate, since the large language model takes a long time to generate a complete text, and the subsequent cumbersome processing process must wait for the complete text to be completely finished before starting, the overall response speed becomes slower.

[0003] In summary, the prior art route introduces a lengthy subsequent processing process for the results generated by the large language model, which cannot balance speed and accuracy, resulting in a long waiting time for the user and difficulty in achieving fast and accurate intelligent movie and television search. SUMMARY

[0004] The main purpose of the present application is to provide a movie and television resource search method, a display device, a storage medium and a program product, aiming to solve the technical problem that the prior art is difficult to achieve fast and accurate intelligent movie and television search.

[0005] To achieve the above-mentioned purpose, the present application provides a movie and television resource search method, which is applied to a display device and includes the following steps: Obtaining a natural language query; Extracting a first keyword related to a movie and television resource attribute from the natural language query, and performing a first movie and television resource retrieval in a movie and television resource library based on the first keyword to obtain a first retrieval result; While extracting the first keyword, inputting the natural language query into a first large model for semantic analysis to generate a target retrieval instruction, wherein the target retrieval instruction is structured data containing a movie and television resource attribute field and its corresponding value; performing a second movie and television resource retrieval in the movie and television resource library based on the target retrieval instruction to obtain a second retrieval result; Determining a target movie and television resource according to the first retrieval result and the second retrieval result.

[0006] In an embodiment, after the step of obtaining the first retrieval result, the method further includes the following steps: In response to the first confirmation operation on the first search result, it is determined that there is a movie and television resource matching the natural language query in the first search result. According to a preset sorting rule, the movie and television resources in the first search result are sorted. The first movie and television resource after sorting is determined as a target movie and television resource.

[0007] In an embodiment, the method further comprises: From the first search result, a preset number of movie and television resources are selected as reference movie and television resources. Attribute information of the reference movie and television resources is obtained, the attribute information including type, leading actor, director, and release year. Based on the attribute information, a movie and television resource attribute field in the target search instruction is corrected to generate an optimized search instruction. Based on the optimized search instruction, a third movie and television resource search is performed in the movie and television resource library to obtain a third search result.

[0008] In an embodiment, the step of determining a target movie and television resource according to the first search result and the second search result comprises: From the first search result and the second search result, a coincident movie and television resource is determined. The first movie and television attribute information corresponding to the coincident movie and television resource in the first search result is semantically compared with the second movie and television attribute information corresponding to the coincident movie and television resource in the second search result. If the comparison result is a semantic conflict, different weight coefficients are respectively assigned to the first search result and the second search result according to the movie and television attribute of the semantic conflict. According to the weight coefficients, the movie and television resources in the first search result and the second search result are respectively scored, and the movie and television resource with the highest score is determined as the target movie and television resource.

[0009] In an embodiment, after the step of obtaining a natural language query, the method further comprises: The natural language query is input to a second large model to obtain a parsing result, wherein the second large model is configured to parse the semantics of the natural language query, and when the semantics point to a unique movie and television work, output the movie name of the movie and television work as the parsing result. If the parsing result is a movie name, the movie and television resource corresponding to the movie name is taken as a target movie and television resource.

[0010] In an embodiment, after the step of obtaining a parsing result, it comprises: If the analysis result is a preset identifier, a second keyword related to a movie and television resource attribute is extracted from the natural language query. Based on the second keyword, fourth movie and television resource retrieval is performed in a movie and television resource library to obtain fourth retrieval results, and target movie and television resources are determined from the fourth retrieval results.

[0011] In addition, to achieve the above-mentioned purpose, the present application also provides a movie and television resource search device, which comprises: The acquisition module is configured to acquire a natural language query. The first retrieval module is configured to extract a first keyword related to a movie and television resource attribute from the natural language query, and perform first movie and television resource retrieval in a movie and television resource library based on the first keyword to obtain first retrieval results. The second retrieval module is configured to input the natural language query into a first large model for semantic analysis while extracting the first keyword to generate a target retrieval instruction, wherein the target retrieval instruction is structured data containing a movie and television resource attribute field and its corresponding value; and perform second movie and television resource retrieval in the movie and television resource library based on the target retrieval instruction to obtain second retrieval results. The determination module is configured to determine target movie and television resources according to the first retrieval results and the second retrieval results.

[0012] In addition, to achieve the above-mentioned purpose, the present application also provides a display device, which comprises a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the movie and television resource search method as described above.

[0013] In addition, to achieve the above-mentioned purpose, the present application also provides a storage medium, which is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the movie and television resource search method as described above.

[0014] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of the movie and television resource search method as described above.

[0015] The one or more technical solutions provided by the present application have at least the following technical effects: Since the prior art solution pursues accuracy, it must sequentially perform large language model analysis and subsequent cumbersome processing procedures, resulting in a significant increase in overall response delay, which cannot balance the technical problems of search efficiency and result accuracy.

[0016] After obtaining the natural language query, the application synchronously starts two parallel processing procedures: one is to extract the first keyword related to the attribute of the video resource from the natural language query, and perform first video resource retrieval in the video resource library based on the first keyword to obtain the first retrieval result; wherein the retrieval based on the first keyword does not depend on the generation delay of the large model, can quickly return the preliminary result (i.e. the first retrieval result), provides instant feedback, and effectively eliminates the response delay caused by waiting for the large model response.

[0017] The second is to change the role of the large model from a time-consuming content (i.e. search result) generator to an efficient structured instruction generator; while extracting the first keyword, input the same natural language query into the first large model for semantic analysis to generate a structured target retrieval instruction, and perform second video resource retrieval based on the target retrieval instruction to obtain the second retrieval result; wherein the target retrieval instruction output by the first large model can be directly and efficiently executed, eliminating the complex post-processing procedures such as secondary semantic analysis, expansion, scoring, etc. of the large model generation result, thereby greatly compressing the time consumption of the large model from understanding to retrieval.

[0018] Finally, the target video resource is determined according to the first retrieval result and the second retrieval result, so that the final result can not only reduce the initial response delay by using the rapidity of the first video resource retrieval, but also can guarantee the result quality by fusing the accuracy of the second video resource retrieval, thereby realizing the intelligent video resource search effect of considering both response speed and search accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0019] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present application and, together with the specification, serve to explain the principles of the present application.

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0021] Figure 1 A flowchart provided by an embodiment of the video resource search method of the present application; Figure 2 A flowchart provided by another embodiment of the video resource search method of the present application; Figure 3 A flowchart provided by still another embodiment of the video resource search method of the present application; Figure 4 A module structure diagram of the video resource search device of the present application; Figure 5 A device structure schematic diagram of a hardware running environment involved in a video resource search method of the present application.

[0022] The object implementation, functional features and advantages of the present application will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0023] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and do not limit the present application.

[0024] In order to better understand the technical solutions of the present application, the following will be described in detail in conjunction with the drawings and specific embodiments of the specification.

[0025] It should be noted that the execution subject of the present embodiment can be a display device, such as a smart TV, a smart screen, a smart projector, a vehicle-mounted display screen with an operating system, a smart conference tablet, etc. It can also be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc. Or it can be an electronic device or a processor that can realize the above functions. The following will take the display device as an example to describe the present embodiment and the following embodiments.

[0026] Based on this, the present embodiment provides a video resource search method, which is described in detail with reference to Figure 1 , Figure 1 A flowchart of an embodiment of the video resource search method of the present application.

[0027] In the present embodiment, the video resource search method is applied to a display device, which includes the following steps: Step S1, obtaining a natural language query; In a feasible embodiment, the display device obtains a query request expressed by a user in natural language.

[0028] Optionally, the natural language query is search content input by the user in spoken or written form, such as "I want to watch a comedy movie", and the display device receives voice or text input through its input interface to provide raw data for subsequent processing.

[0029] It can be understood that obtaining a natural language query aims to capture the subjective intention and vague demand expressed by the user, ensuring that all subsequent processing is based on the real and direct demand of the user, avoiding the limitations and deviations caused by pre-set options.

[0030] Step S2, extracting a first keyword related to the attribute of the video resource from the natural language query, and performing a first video resource retrieval in the video resource library based on the first keyword to obtain a first retrieval result; In an implementable embodiment, the display device extracts a first keyword related to the attribute of the film and television resource from the natural language query. Then, the first film and television resource retrieval is performed in the film and television resource library based on the first keyword, so as to obtain the first retrieval result.

[0031] Optionally, the attribute of the film and television resource includes a film title, an actor, a director or a type, etc.

[0032] Optionally, the first keyword refers to a word related to the attribute of the film and television resource recognized from the text of the natural language query, such as a film name or an actor name.

[0033] Optionally, the film and television resource library is a database storing information of a plurality of film and television works.

[0034] Optionally, the first film and television resource retrieval is a search operation based on keyword matching, and the first retrieval result, i.e., a preliminary filtered set of film and television resources, can be obtained.

[0035] It can be understood that the reason for the keyword extraction operation is that many natural language queries themselves contain explicit key information, and through efficient string matching, a preliminary relevant result set (i.e., the first retrieval result) can be quickly narrowed down from the massive film and television resource library.

[0036] It can be understood that this step is a fast matching and filtering based on the text features of the natural language query, which realizes the instantaneity and basic coverage rate of the retrieval, can effectively deal with the scene where the user demand is clear and the expression is direct, and provides a comparable and integrable baseline result for the subsequent more complex semantic retrieval.

[0037] Step S3, while extracting the first keyword, inputting the natural language query into the first large model for semantic analysis to generate a target retrieval instruction, wherein the target retrieval instruction is structured data containing the attribute field of the film and television resource and the corresponding value; performing the second film and television resource retrieval in the film and television resource library based on the target retrieval instruction to obtain the second retrieval result. In an implementable embodiment, the display device inputs the same natural language query obtained into the first large model for deep analysis and conversion while extracting the first keyword in parallel, to generate a structured target retrieval instruction; then the display device performs a new retrieval (i.e., the second film and television resource retrieval) in the film and television resource library based on the target retrieval instruction, so as to obtain the second retrieval result.

[0038] Optionally, the first large model refers to an artificial intelligence model deployed locally or in the cloud, which is pre-trained by a large amount of corpus and has strong natural language understanding and generation capabilities.

[0039] Optionally, semantic parsing refers to the process of deconstructing the natural language query by the first large model (e.g., extracting text features, converting to vector form, etc.) to understand the deep user intent, contextual associations, and implicit filtering conditions in the natural language query.

[0040] Optionally, generating the target retrieval instruction refers to the first large model outputting the conclusion of semantic parsing as a command in a regular format that can be directly used by a computer program (e.g., a retrieval engine built-in or connected to the display device).

[0041] Optionally, structured data is a pre-defined format of data organization that contains fields of movie resource attributes (e.g., director, genre, etc.) and their corresponding values, which can be in the form of a JSON object, an XML document, or other standard data exchange formats; for example, for the query "comedy films directed by Stephen Chow", structured data in the form of {“director”: “Stephen Chow”, “genre”: “comedy”} can be generated.

[0042] Optionally, the second movie resource retrieval is a precise and deep search process of the movie resource library using the structured query condition represented by the target retrieval instruction.

[0043] It can be understood that introducing the first large model for semantic parsing and generating structured target retrieval instructions overcomes the limitations of traditional keyword matching, can understand abstract concepts, emotional tendencies, and complex logical relationships, and thus can find resources that better meet the user's real intent from the semantic level and convert them into precise, unambiguous, and machine-readable instructions.

[0044] It can be understood that by generating target retrieval instructions in standard formats such as JSON or XML, the decoupling of "understanding" by the large model and "execution" by the program is achieved. Instead of directly outputting movie resource results, the first large model outputs standard machine instructions, which allows the downstream search engine or database to directly and efficiently execute the target retrieval instruction without the need for secondary parsing and guessing, greatly improving the execution speed and reliability of the entire retrieval process.

[0045] Step S4, determining the target movie resource according to the first retrieval result and the second retrieval result.

[0046] In a feasible embodiment, after obtaining the first retrieval result generated by the first movie resource retrieval and the second retrieval result generated by the second movie resource retrieval, the display device performs comprehensive processing such as comparison or fusion strategy on the two retrieval results to determine the target movie resource to be presented to the user from the two retrieval results.

[0047] Optionally, the first search result refers to a set of video resources obtained by the display device based on a first video resource search operation matched with the first keyword.

[0048] Optionally, the second search result refers to a set of video resources obtained by the display device based on a second video resource search operation based on the target search instruction output by the large model.

[0049] Optionally, the target video resource refers to a preferred video resource list or set generated and presented to the user after the display device comprehensively processes the two sets of video resources.

[0050] It can be understood that the first search result is based on keyword matching, the response is fast and the coverage is wide, but it may not fully reflect the deep intention; the second search result is based on deep semantic understanding, the intention matching is accurate, but the calculation complexity is relatively high. Therefore, by fusing the two, the accurate results of deep semantic matching can be effectively included on the basis of ensuring fast response, thereby comprehensively improving the overall quality of the final search result in terms of coverage and intention relevance.

[0051] The embodiment provides a video resource search method. After obtaining a natural language query, two parallel processing procedures are started synchronously: one is to extract a first keyword related to a video resource attribute from the natural language query, and perform a first video resource search in a video resource library based on the first keyword to obtain a first search result; wherein the search based on the first keyword does not depend on the generation delay of the large model, and can quickly return a preliminary result (i.e., the first search result), thereby providing instant feedback and effectively eliminating the response delay caused by waiting for the response of the large model. The other is to change the role of the large model from a time-consuming content (i.e., search result) generator to an efficient structured instruction generator; while extracting the first keyword, the same natural language query is input into the first large model for semantic analysis to generate a structured target search instruction, and a second video resource search is performed based on the target search instruction to obtain a second search result; wherein the target search instruction output by the first large model can be directly and efficiently executed, thereby eliminating the complex post-processing procedures such as secondary semantic analysis, expansion and scoring of the large model generation result, thereby greatly compressing the time consumption of the large model from understanding to searching. Finally, the target video resource is determined based on the first search result and the second search result, so that the final result can not only reduce the initial response delay by using the quickness of the first video resource search, but also can fuse the accuracy of the second video resource search to ensure the quality of the result, thereby realizing the intelligent video resource search effect with consideration of response speed and search accuracy.

[0052] Exemplarily, please refer to Figure 2 , Figure 2A flowchart of a movie resource search method is provided. Specifically, when a user issues a natural language query to a display device, the display device synchronously executes two processing paths. In the keyword path, a first keyword is extracted from the natural language query, and a fast matching is performed in a movie resource library based on the first keyword to obtain a first search result. In the semantic analysis path, the same natural language query is input into a first large model, the first large model analyzes the deep semantics thereof, and generates a structured target search instruction accordingly. Then, a search is performed based on the target search instruction to obtain a second search result. Finally, the search results returned by the two paths are fused, de-duplicated, sorted, and the like, the breadth of keyword matching and the accuracy of semantic understanding are comprehensively considered, and the final determined target movie resource (such as a preferred list containing the above-mentioned movie) is presented to the user.

[0053] Based on the above-mentioned embodiments of the present application, in another embodiment of the present application, the same or similar contents as the above-mentioned embodiments can be referred to the above introduction, and the subsequent will not be described in detail. On this basis, after the step S2 of obtaining the first search result, the step further comprises: Step S21, in response to the first confirmation operation on the first search result, it is determined that there is a movie resource in the first search result that matches the natural language query; In a feasible embodiment, after the display device obtains the first search result and displays it, the display device continues to monitor the user's interaction behavior. When the display device detects that the user triggers the first confirmation operation for the displayed first search result, it is determined whether there is an item in the movie resource set contained in the first search result that matches the user input natural language query in semantics.

[0054] Optionally, the first confirmation operation refers to the user's intention signal of expressing approval or selection from the currently presented first search result to the display device by touching a specific area on the screen, using a remote controller, issuing a specific confirmation voice instruction, or other preset interaction mode.

[0055] Optionally, after the display device receives the first confirmation operation, based on the positive feedback of the user of the first confirmation operation, and in combination with the first search result, it is determined that the first search result at least contains one valid candidate movie resource that can be regarded as corresponding to the user query.

[0056] It can be understood that the introduction of this step is to establish a resource optimization mechanism based on real-time user feedback for converting the feedback behavior of the user into a control signal. When the display device captures the control signal, it can be inferred that the user is satisfied with the quality of the result of the fast keyword search, so as to terminate the time-consuming and resource-consuming deep semantic search process in advance, which enhances the interactivity and response agility with the user.

[0057] Step S22, sorting the movie and television resources in the first search result according to a preset sorting rule; In an implementable embodiment, after determining that there is a matching movie and television resource, the display device reevaluates and calculates all the movie and television resources in the first search result according to a set of preset sorting rules, and generates a new ordered list according to the priority obtained by the evaluation.

[0058] Optionally, the preset sorting rule is an evaluation criterion and calculation logic that is predefined and stored in the display device or an associated server, and is used to quantify the relevance of the movie and television resource to the current natural language query context. The sorting rule can comprehensively consider various factors, such as the accuracy of keyword matching, the popularity index of the movie and television resource, the newness of the release time, and the degree of agreement with the user's historical preferences.

[0059] Optionally, the sorting refers to an operation process in which the display device applies the above-mentioned sorting rule to calculate a comprehensive sorting score for each movie and television resource in the first search result, and reorders them according to the scores.

[0060] It can be understood that even if the user confirms that the first search result contains matching target content, the target content may be at the end of the first search result without processing, or scattered in an inconspicuous middle position. By reordering according to the preset sorting rule, the first search result that has been confirmed to be effective can be internally optimized, and the movie and television resource that best matches the user's comprehensive demand can be placed in the first position. This operation improves the orderliness and presentation efficiency of the new ordered list, ensuring that the user can preferentially access and lock the best option when selecting or the display device automatically determines the target movie and television resource, thereby ensuring high-quality search under a fast search path.

[0061] Step S23, determining the first movie and television resource after sorting as the target movie and television resource.

[0062] In an implementable embodiment, after completing the sorting and obtaining the new ordered list, the display device performs a selection operation, that is, directly extracts the corresponding movie and television resource from the top (i.e., the first position) of the ordered list, and uses it as the target movie and television resource to be output or played in this search task.

[0063] It can be understood that in the scenario where the user explicitly indicates that the first search result contains matching movie and television resources, the minimum delay from user interaction to result output is achieved, the advantages of keyword search in terms of speed are maximized, instantaneous response is ensured, and unnecessary subsequent calculation resource consumption is reduced.

[0064] In the embodiment, a conditional branch is added to the method for searching for a video resource. Specifically, the branch defines a quick processing flow triggered by a first confirmation operation of a user. When the user explicitly expresses approval of the first search result through the first confirmation operation, the display device will skip the subsequent semantic analysis step and multi-result fusion step based on the first large model, and directly enter the sorting and selection logic of the confirmed result, thereby greatly shortening the overall processing delay from query to output of the final target resource. Secondly, by taking the user confirmation operation as the basis for judging whether to perform semantic analysis, the display device can avoid starting the first large model inference process in a large number of simple and explicit query scenarios, saving computing power and memory resources. The conditional branch and steps S1 to S4 form a complementary relationship, and the execution logic of the entire method becomes: after obtaining the first search result, the display device listens to user feedback in real time; if the first confirmation operation is detected, the quick processing flow of steps S21 to S23 is executed; if the operation is not detected, the semantic analysis of step S3 and the result fusion of step S4 are continued in parallel or sequentially. This enables switching between the two processing paths according to the real-time state of user interaction, thereby optimizing the control of processing efficiency and resource consumption while ensuring the completeness and accuracy of the search function.

[0065] Based on the above embodiments of the present application, in another embodiment of the present application, the same or similar content as the above embodiments can be referred to the above introduction, and the following will not be repeated. On this basis, the method for searching for a video resource further comprises the following steps: Step A1, selecting a preset number of video resources from the first search result as reference video resources; In a feasible embodiment, after obtaining the first search result, the display device does not immediately use all of them for subsequent fusion or direct output, but first selects a part of the video resources from the first search result according to a preset number rule (such as the first N items) or selection rule (such as Top N according to the initial score), and marks the selected resources as reference video resources for subsequent analysis.

[0066] Optionally, the preset number is a configurable integer parameter, which is pre-set or calculated according to the total number of the first search result.

[0067] Optionally, the reference video resource refers to a subset of video resources selected from the first search result.

[0068] Understandably, the purpose of this selection step is to provide a data sample set for subsequent search optimization. The first search result originates from the first keyword in the natural language query, and the top-ranking film and television resources in this result have a high degree of relevance to the intent of the natural language query. By selecting a portion of film and television resources from the first search result as reference resources, the display device can obtain a set of film and television resources that have passed the initial screening, which lays a data foundation for subsequent refinement of search instructions.

[0069] Step A2: Obtain the attribute information of the reference film and television resources. The attribute information includes the type, lead actors, director, and release year. In one feasible embodiment, after determining a subset of reference film and television resources, the display device accesses a local or remote film and television resource library to query and obtain detailed attribute information associated with the film and television resources in the film and television resource library.

[0070] Optionally, attribute information refers to structured data fields recorded in the film and television resource database that describe the various characteristics of the film and television resources. Acquisition refers to the process by which the display device reads the pre-stored attribute information field values ​​through a database query interface.

[0071] Optionally, genre, lead actors, director, and release year are common examples of fields in attribute information, and may also include film length, etc.

[0072] Understandably, this step transforms the reference film and television resources from abstract identifiers into a set of features, and then, by obtaining the attribute information of the reference film and television resources, the display device can modify the search instructions based on the attribute information.

[0073] Step A3: Based on the attribute information, modify the film and television resource attribute fields in the target search instruction to generate an optimized search instruction; In one feasible embodiment, the display device performs aggregate analysis on the attribute information of all acquired reference film and television resources (such as statistically analyzing frequency of occurrence and identifying common features). Then, the display device compares the analyzed feature patterns with the target retrieval command generated by the first model, and adds, deletes, or adjusts the weights of the corresponding film and television resource attribute fields and their values ​​in the target retrieval command, thereby generating a new, modified retrieval command, i.e., an optimized retrieval command.

[0074] Optionally, "correction" refers to the display device adjusting the content of the target search instruction based on the attribute distribution of the reference resources, making it closer to the potential user preferences or common characteristics of film and television resources observed from the first search results. For example, if the "genre" attribute of most reference resources is "comedy," the display device may strengthen the value of "comedy" or increase its weight in the optimized search instruction.

[0075] Optionally, the optimized retrieval instruction is an updated version of the target retrieval instruction, which is also in the form of structured data, but the query condition contained therein is changed by incorporating the feedback information from the first retrieval result.

[0076] It can be understood that the modification step is to reflect the direct correspondence between the keyword matching in the user query and the explicit feature distribution of the video resource. Based on this, the target retrieval instruction generated by semantic analysis is modified, realizing feedback and iterative optimization from the preliminary retrieval result to the deep semantic retrieval instruction. This makes the final retrieval instruction not only based on single semantic understanding of the large model, but also incorporates the keyword statistical features of the first retrieval result, which can overcome the possible deviation of single analysis, making the final retrieval result more comprehensive and accurate.

[0077] Step A4, performing a third video resource retrieval in the video resource library based on the optimized retrieval instruction to obtain a third retrieval result.

[0078] In a feasible embodiment, after generating the optimized retrieval instruction, the display device initiates and performs a retrieval operation in the video resource library by taking it as a new query condition. This retrieval is independent of the previous second video resource retrieval and is referred to as a third video resource retrieval, and the result of the third video resource retrieval is a third retrieval result.

[0079] Optionally, the third video resource retrieval is a query process initiated by the display device based on the optimized retrieval instruction.

[0080] Optionally, the third retrieval result is a set of video resources returned by the third video resource retrieval.

[0081] It can be understood that after the display device completes the modification of the target retrieval instruction and generates the optimized retrieval instruction, the third video resource retrieval is performed to verify the effectiveness of the instruction optimization and obtain a more accurate set of video resources (third retrieval result). This ensures that the semantic retrieval condition enhanced by the feature feedback of the first retrieval result has improved retrieval accuracy.

[0082] In this embodiment, the instruction optimization and re-retrieval scheme based on the feedback of the first retrieval result is added. The highly relevant result obtained by the fast keyword retrieval is used to modify the target retrieval instruction generated by the deep semantic analysis, thereby improving the accuracy of the semantic retrieval condition.

[0083] In an embodiment, the output of the third video resource retrieval (third retrieval result) can be provided as a new and better input to the result determination process of step S4, and participate in fusion together with the first retrieval result, thereby further incorporating the distribution features of the first retrieval result on the basis of the original deep semantic understanding, and improving the comprehensiveness and accuracy of the final target video resource from multiple dimensions.

[0084] Based on the above embodiments of the application, in another embodiment of the application, the same or similar content as the above embodiments can be referred to the above introduction, and the following will not be described in detail. On this basis, step S5 comprises: Step S51, from the first search result and the second search result, determine the coincident video resources; In an available embodiment, after obtaining the first search result and the second search result, the display device performs a comparison operation, specifically, traverses each video resource identifier in the first search result, and finds whether there is a same video resource identifier in the second search result, and marks all video resources found in both search results as coincident video resources.

[0085] Optionally, the coincident video resources refer to the video resources appearing in both the first search result and the second search result.

[0086] It can be understood that the coincident video resources meet the conditions of keyword matching and deep semantic understanding at the same time, have high relevance and confidence, and provide a clear comparison sample for subsequent analysis of the difference between the two search paths.

[0087] Step S52, perform semantic comparison on the first video attribute information corresponding to the coincident video resources in the first search result and the second video attribute information corresponding to the coincident video resources in the second search result; In an available embodiment, for each coincident video resource, the display device respectively obtains its attribute description in the first search result (first video attribute information) and its attribute description in the second search result (second video attribute information). Then, the display device compares the two sets of attribute information describing the same video resource in content level, and checks whether there is inconsistency or contradiction in expression.

[0088] Optionally, the first video attribute information and the second video attribute information refer to the text description or characteristic label set of the same video resource used for searching in the respective search result list. For example, in the first search result, the attribute information of a certain movie contains the keywords "science fiction, 2023" input by the user, while in the second search result, the attribute information of the movie is the deep semantic labels such as "type: science fiction, theme: space exploration, era: 2020s" generated by the large model after decomposing the user query "the latest space exploration movie".

[0089] Optionally, the semantic comparison refers to the comparison of the display device on the natural language understanding level of the two sets of attribute information.

[0090] It can be understood that the purpose of semantic comparison is to find the inherent logical differences or potential conflicts of different search paths matching the same resource. Since the first search is based on keywords, its matching attribute is more direct and superficial; the second search is based on large model analysis, and its matching attribute may be closer to the context and potential intention of the query. Through comparison, the possible bias of the two search mechanisms in understanding user intention can be revealed to evaluate the advantages and limitations of each method.

[0091] In step S53, if the comparison result is that there is a semantic conflict, different weight coefficients are respectively assigned to the first search result and the second search result according to the attribute of the semantic conflict. In an available embodiment, when the display device determines through semantic comparison that there is a semantic conflict in the attribute information description of one or more coincident video resources, a weight adjustment mechanism is started. The display device queries the preset weight distribution rule according to the specific video attribute category of the conflict (for example, whether it is a "type" conflict or a "era" conflict), and dynamically assigns different weight coefficients to the first search path (i.e. the path represented by the first search result) and the second search path (i.e. the path represented by the second search result) for generating the coincident resource.

[0092] Optionally, the semantic conflict refers to a state in which the first video attribute information and the second video attribute information are in conflict in semantics for the same video resource.

[0093] Optionally, the weight coefficient is a numerical parameter for quantifying the contribution or credibility of different search paths in the final resource sorting.

[0094] It can be understood that when the determination basis of the same video resource by the two search methods is in conflict, it indicates that there is an essential difference in the determination logic of the relevance of the video resource and the query intention by the two methods. At this time, the reliability of the two methods can be differentiated and weighted according to the conflicting video attribute. For example, if the conflict is in the "director" or "plot theme" deep semantic information, the second search result based on large model semantic analysis can be given a higher weight because the large model is better at handling complex semantic associations and implicit intentions. This differentiated weight distribution makes the fusion strategy more adaptive and accurate.

[0095] In step S54, the video resources in the first search result and the second search result are respectively scored according to the weight coefficients, and the video resource with the highest score is determined as the target video resource.

[0096] In one feasible embodiment, the display device applies the weighting coefficients determined in the previous step to the scoring calculation of the overlapping film and television resource and even all candidate film and television resources. For each film and television resource (including overlapping and non-overlapping resources), the display device calculates a comprehensive score by combining its source (first search result or second search result) and the current weighting coefficient of the source path. Finally, the display device selects the film and television resource with the highest comprehensive score from all candidate resources and identifies it as the target film and television resource.

[0097] Understandably, this allows the final resource recommendation to not only integrate the search results from different paths, but also to finely adjust the influence of each path's results through a dynamic weighting mechanism, thereby improving the accuracy of the recommendation results.

[0098] In this embodiment, a retrieval result fusion method is provided by determining overlapping film and television resources, semantic comparison of attribute information, conflict-driven weight allocation, and weight-based scoring fusion mechanism. This method can intelligently identify and reconcile the differences and contradictions between keyword matching and semantic understanding, and flexibly balance the contributions of all parties using weight coefficients, ultimately achieving more accurate target resource positioning that better meets the complex intentions of users.

[0099] For example, please refer to Figure 3 , Figure 3 This document provides a flowchart illustrating a method for searching film and television resources. Specifically, after a user inputs a natural language query, the display device simultaneously obtains a first search result (based on keyword matching) and a second search result (a target search instruction generated after semantic parsing of the natural language query using a large model). Then, the first and second search results are compared to identify overlapping film and television resources. Next, the first film and television attribute information of the overlapping resources in the first search result and the second film and television attribute information in the second search result are obtained. Subsequently, a semantic comparison is performed on the first and second film and television attribute information to identify semantically conflicting film and television attributes. Based on these semantically conflicting attributes, different weight coefficients are assigned to the first and second search results, respectively. Finally, based on the adjusted weights, the film and television resources in the first and second search results are scored, and the film and television resource with the highest score is identified as the target film and television resource.

[0100] Based on the above embodiments of this application, in another embodiment of this application, the same or similar content as the above embodiments can be referred to the above description, and will not be repeated hereafter. Based on this, after step S1, the following steps are included: Step D1: Input the natural language query into the second large model to obtain the parsing result. The second large model is configured to: parse the semantics of the natural language query, and when the semantics point to a unique film or television work, output the film title of the film or television work as the parsing result. In an embodiment, the display device inputs the natural language query into a pre-configured second large model after obtaining the natural language query. The second large model outputs a resolution result.

[0101] Optionally, the second large model is a large language model independent of the first large model in step S3, and the training target of the second large model is set to be dedicated to analyzing whether the natural language query unambiguously and uniquely refers to a film and television work.

[0102] Optionally, resolving the semantics of the natural language query means that the second large model processes the query text to determine whether the object described by the text has unique certainty.

[0103] Optionally, the resolution result is the content output by the second large model. When the second large model determines that the semantics of the natural language query unambiguously points to a unique film and television work, the resolution result is the film name of the film and television work; otherwise, the resolution result is an indication signal that a unique film cannot be determined.

[0104] It can be understood that according to the output of the second large model, the type of user query intention can be quickly identified. If the output is an explicit film name, it indicates that the user's intention is clear and the query object is unique, and the target film and television resource with unique certainty is directly output.

[0105] Step D2, if the resolution result is a film name, the film and television resource corresponding to the film name is taken as the target film and television resource.

[0106] In an embodiment, after receiving the resolution result of the second large model, the display device determines whether the resolution result is a valid film name. If it is determined to be yes, the display device performs an exact match query in the film and television resource library using the film name as the key information. When the match is successful, the display device directly determines the unique film and television resource found as the target film and television resource.

[0107] Optionally, the film and television resource corresponding to the film name refers to the film and television data in the film and television resource library whose identification information completely matches the film name provided by the resolution result.

[0108] It can be understood that when the resolution result is a film name, it indicates that the user's demand does not need to be satisfied through multiple rounds of retrieval and result fusion. The display device directly acquires the film and television resource through exact match and terminates all subsequent retrieval steps, which maximally reduces the processing links and realizes the shortest response delay and the lowest calculation cost in a specific scenario.

[0109] In an embodiment, after step D1, the following steps are included: Step D11, if the resolution result is a preset identifier, a second keyword related to the attribute of the film and television resource is extracted from the natural language query; In an implementable embodiment, when the display device obtains the resolution result from the second large model as a preset identifier representing "unknown" or "unable to determine a unique movie", the display device initiates a backup processing flow, i.e., extracting a second keyword related to the attribute of the movie and television resource from the natural language query.

[0110] Optionally, the preset identifier is a specific symbol or string predefined to represent the case that the second large model cannot resolve a unique movie name from the query, for example, "unknown".

[0111] Optionally, the second keyword refers to a word extracted from the natural language query to describe the characteristics of the movie and television resource. Here, it is used as an alternative solution to directly match the movie name in the case of a fuzzy query.

[0112] It can be understood that this step aims to solve the problem of using keyword extraction technology to screen movie and television resources for a fuzzy query when the second large model judges that the user query intention is not clear (i.e., not uniquely referred to).

[0113] Step D12, performing a fourth movie and television resource retrieval in the movie and television resource library based on the second keyword, obtaining a fourth retrieval result, and determining the target movie and television resource from the fourth retrieval result.

[0114] In an implementable embodiment, after successfully extracting the second keyword, the display device performs a new retrieval operation, i.e., the fourth movie and television resource retrieval, in the movie and television resource library using the second keyword as the query condition. After the retrieval is completed, the display device obtains the fourth retrieval result. Subsequently, the display device determines the final target movie and television resource based on the fourth retrieval result through internal preset rules (such as selecting the resource with the highest relevance).

[0115] In this embodiment, by inputting the natural language query into the second large model to determine the intention, and triggering different retrieval paths and processing logic according to different resolution results, simple and clear requests are responded quickly using a low-cost path, and complex requests are processed with more resources to ensure search accuracy, thereby realizing intelligent allocation of retrieval resources and computing overhead, and balancing response efficiency and search quality as a whole. Specifically, when the second large model outputs a clear movie name, the display device can directly output the target resource through one-time accurate database matching, realizing the shortest response path and near-instant response speed. When the second large model outputs a preset identifier, a lighter and faster keyword retrieval path is enabled, which provides a quick solution for processing simple and fuzzy queries and avoids unnecessary complex semantic analysis calculation.

[0116] It should be noted that the above examples are only used for understanding the present application and do not constitute a limitation on the method for searching video resources of the present application, and more forms of simple transformation such as interaction and combination of various embodiments can be carried out based on this technical concept, which are all within the protection scope of the present application.

[0117] The present application also provides a video resource search device, please refer to Figure 4 , the video resource search device comprises: The acquisition module 10 is used for acquiring a natural language query. The first retrieval module 20 is used for extracting a first keyword related to a video resource attribute from the natural language query, and performing first video resource retrieval in a video resource library based on the first keyword to obtain a first retrieval result. The second retrieval module 30 is used for inputting the natural language query into a first large model for semantic analysis while extracting the first keyword to generate a target retrieval instruction, wherein the target retrieval instruction is structured data containing a video resource attribute field and its corresponding value; and performing second video resource retrieval in the video resource library based on the target retrieval instruction to obtain a second retrieval result. The determination module 40 is used for determining a target video resource according to the first retrieval result and the second retrieval result.

[0118] The video resource search device provided by the present application adopts the video resource search method in the above embodiments, which can solve the technical problem of the existing difficulty in realizing fast and accurate intelligent video search. Compared with the prior art, the video resource search device provided by the present application has the same beneficial effects as the video resource search method provided by the above embodiments, and the other technical features in the video resource search device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0119] The present application provides a display device, which comprises at least one processor and a memory connected with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the video resource search method in the first embodiment.

[0120] The following refers to Figure 5The diagram illustrates a structural schematic of a display device suitable for implementing embodiments of this application. The display device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The display device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0121] like Figure 5 As shown, the display device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the display device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. The communication device 1009 allows the display device to exchange data with other devices wirelessly or via wired communication. Although the diagram shows display devices with various systems, it should be understood that it is not required to implement or have all of the systems shown. More or fewer systems may be implemented alternatively.

[0122] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program codes for executing the method shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network through a communication device, or installed from the storage device 1003, or installed from the read-only memory 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiments disclosed in the present application are executed.

[0123] The display device provided by the present application adopts the film and television resource search method in the above-mentioned embodiments, and can solve the technical problem that it is difficult to realize fast and accurate intelligent film and television search. Compared with the prior art, the display device provided by the present application has the same beneficial effects as the film and television resource search method provided by the above-mentioned embodiments, and other technical features in the display device are the same as the features disclosed in the previous embodiment method, which will not be repeated here.

[0124] It should be understood that various parts of the present application can be realized by hardware, software, firmware or a combination thereof. In the description of the above-mentioned embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0125] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0126] The present application provides a computer readable storage medium having stored thereon computer readable program instructions (i.e. computer program) for executing the film and television resource search method in the above-mentioned embodiments.

[0127] The computer readable storage medium provided by the present application may, for example, be a U disk, but is not limited to an electric, magnetic, optical, electromagnetic, infrared, or semiconductor system or device, or any combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electric connection with one or more conductive wires, a portable computer disk, a hard disk, a random access memory (RAM), a read only memory (ROM), an erasable programmable read only memory (EPROM or flash memory), an optical fiber, a portable compact disk read only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present embodiment, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer readable storage medium can be transmitted in any suitable medium, including but not limited to an electric wire, an optical cable, an RF (Radio Frequency), and the like, or any suitable combination of the above.

[0128] The above computer readable storage medium can be contained in a display device, or can exist separately without being assembled into a display device.

[0129] Computer program code for carrying out operations of the present application can be written in one or more programming languages or combinations of languages including object oriented programming languages such as Java, Smalltalk, C++ or conventional procedural programming languages such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0130] The flow and block diagrams in the drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flow and block diagrams can represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may be executed in the reverse order, depending on the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustrations, and combinations thereof, can be implemented by special purpose hardware-based systems that perform the specified functions or operations, or combinations of special purpose hardware and computer instructions.

[0131] The modules involved in the embodiments of the present application can be implemented in software or in hardware. In some cases, the names of the modules do not limit the modules themselves.

[0132] The readable storage medium provided by the present application is a computer readable storage medium, which stores computer readable program instructions (i.e. computer programs) for executing the above-mentioned video resource search method, and can solve the technical problem that the existing intelligent video search is difficult to achieve fast and accurate. Compared with the prior art, the computer readable storage medium provided by the present application has the same beneficial effects as the video resource search method provided by the above-mentioned embodiments, and will not be described here.

[0133] The present application also provides a computer program product, comprising a computer program, which is executed by a processor to implement the steps of the above-mentioned video resource search method.

[0134] The computer program product provided by the present application can solve the technical problem that the existing intelligent video search is difficult to achieve fast and accurate. Compared with the prior art, the computer program product provided by the present application has the same beneficial effects as the video resource search method provided by the above-mentioned embodiments, and will not be described here.

[0135] The above only describes some embodiments of the present application, and does not limit the patent scope of the present application, and any equivalent structural transformation made by using the content of the present application specification and drawings, or direct / indirect application in other related technical fields is included in the patent protection scope of the present application.

Claims

1. A method for searching film and television resources, characterized in that, The film and television resource search method is applied to a display device and includes: Obtain natural language queries; Extract the first keyword related to the attributes of film and television resources from the natural language query, and perform a first film and television resource search in the film and television resource database based on the first keyword to obtain the first search result; While extracting the first keyword, the natural language query is input into the first large model for semantic parsing to generate a target retrieval instruction. The target retrieval instruction is structured data containing film and television resource attribute fields and their corresponding values. Based on the target retrieval instruction, a second film and television resource retrieval is performed in the film and television resource database to obtain a second retrieval result. Based on the first search result and the second search result, the target film and television resources are determined.

2. The film and television resource search method as described in claim 1, characterized in that, After the step of obtaining the first search result, the method further includes: In response to the first confirmation operation of the first search result, it is determined that there are film and television resources in the first search result that match the natural language query; The film and television resources in the first search results are sorted according to the preset sorting rules; The first film and television resource after sorting is selected as the target film and television resource.

3. The film and television resource search method as described in claim 1, characterized in that, The method further includes: From the first search results, a preset number of film and television resources are selected as reference film and television resources; Obtain the attribute information of the reference film and television resources, including the type, lead actor, director, and release year; Based on the attribute information, the film and television resource attribute fields in the target search instruction are modified to generate an optimized search instruction; Based on the optimized search instructions, a third film and television resource search is performed in the film and television resource database to obtain the third search results.

4. The film and television resource search method as described in claim 1, characterized in that, The step of determining the target film and television resource based on the first search result and the second search result includes: Based on the first search result and the second search result, overlapping film and television resources are identified; The first film and television attribute information corresponding to the overlapping film and television resources in the first search result is semantically compared with the second film and television attribute information corresponding to the overlapping film and television resources in the second search result. If the comparison result indicates a semantic conflict, then based on the film and television attributes of the semantic conflict, different weight coefficients are assigned to the first search result and the second search result respectively. Based on the weighting coefficients, the film and television resources in the first search results and the second search results are scored respectively, and the film and television resource with the highest score is determined as the target film and television resource.

5. The method as described in claim 1, characterized in that, After the step of obtaining the natural language query, the method further includes: The natural language query is input into the second large model to obtain the parsing result. The second large model is configured to: parse the semantics of the natural language query, and when the semantics point to a unique film or television work, output the film title of the film or television work as the parsing result. If the parsing result is a movie title, then the film and television resource corresponding to the movie title will be used as the target film and television resource.

6. The method as described in claim 5, characterized in that, Following the step of obtaining the parsed result, the following is included: If the parsing result is a preset identifier, then extract the second keyword related to the attributes of film and television resources from the natural language query; Based on the second keyword, a fourth film and television resource search is performed in the film and television resource database to obtain the fourth search result, and the target film and television resource is determined from the fourth search result.

7. A film and television resource search device, characterized in that, The film and television resource search device includes: The acquisition module is used to acquire natural language queries; The first retrieval module is used to extract first keywords related to film and television resource attributes from the natural language query, and perform a first film and television resource retrieval in the film and television resource database based on the first keywords to obtain a first retrieval result; The second retrieval module is used to extract the first keyword and input the natural language query into the first large model for semantic parsing to generate a target retrieval instruction. The target retrieval instruction is structured data containing film and television resource attribute fields and their corresponding values. Based on the target retrieval instruction, a second film and television resource retrieval is performed in the film and television resource database to obtain a second retrieval result. The determination module is used to determine the target film and television resources based on the first search result and the second search result.

8. A display device, characterized in that, The display device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the film and television resource search method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the film and television resource search method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the film and television resource search method as described in any one of claims 1 to 6.

Images