A media information identification method, device, equipment and medium

Abstract

This invention discloses a media information identification method, apparatus, device, and medium. The media information identification method includes the following steps: acquiring the original path information of the media resource to be identified, and performing hierarchical semantic parsing on the original path information to obtain a set of candidate media names including multiple candidate media names; determining the target identification pattern of the media resource to be identified based on the original path information; based on the target identification pattern, inputting each candidate media name in the candidate media name set into a preset main data source for initial retrieval to obtain a preliminary identification result of the media resource to be identified; when the preliminary identification result does not meet the preset identification conditions, obtaining secondary retrieval constraints through a preset large model, and inputting the secondary retrieval constraints into the main data source for secondary retrieval to obtain the media entity information of the media resource to be identified. This method significantly improves the accuracy of media information identification through a dual retrieval method that combines hierarchical semantic parsing and large model constraints.

Description

Technical Field

[0001] This invention relates to the field of digital media processing technology, and in particular to a media information identification method, apparatus, device, and medium. Background Technology

[0002] With the widespread use of home media centers, network-attached storage devices, smart media players, and personal media library systems, users typically store video resources such as movies, TV series, animations, and documentaries on local disks, shared directories on local area networks, cloud drive mount directories, or hybrid resource libraries. To enable features such as poster display, plot summaries, actor information, categorized search, continue playback, and favorites recommendations, the media player backend usually needs to automatically map unstructured file paths to standardized media entity information.

[0003] However, in real-world scenarios, media resource path naming often lacks standardized specifications. For example, the same media resource may simultaneously contain Chinese and English names, translations, year, resolution, encoding group, subtitle group, audio track information, compilation identifier, season number, source site marker, and other noisy strings. For complex scenarios such as TV series, animations, anime series, film series, director's cuts, remakes, and compilation directories, relying solely on a single filename or a single external database search often fails to accurately identify the target media entity. The identification results exhibit numerous false matches, missed matches, and ambiguous matches, leading to low accuracy in media information identification. Summary of the Invention

[0004] This invention provides a media information identification method, apparatus, device, and medium to address the problem of low accuracy in existing media information identification methods.

[0005] In a first aspect, this application provides a media information identification method, comprising the steps of: obtaining the original path information of the media resource to be identified, and performing hierarchical semantic parsing on the original path information to obtain a set of candidate media names including multiple candidate media names; determining the target identification pattern of the media resource to be identified based on the original path information; inputting each of the candidate media names in the set of candidate media names into a preset main data source for initial retrieval based on the target identification pattern to obtain a preliminary identification result of the media resource to be identified; when the preliminary identification result does not meet the preset identification conditions, obtaining secondary retrieval constraints through a preset large model, and inputting the secondary retrieval constraints into the main data source for secondary retrieval to obtain media entity information of the media resource to be identified.

[0006] Secondly, this application provides a media information identification device, comprising: a set acquisition module, configured to acquire the original path information of the media resource to be identified, and perform hierarchical semantic parsing on the original path information to obtain a set of candidate media names including multiple candidate media names; a pattern determination module, configured to determine the target identification pattern of the media resource to be identified based on the original path information; an initial retrieval module, configured to input each candidate media name in the set of candidate media names into a preset main data source for initial retrieval based on the target identification pattern, and obtain a preliminary identification result of the media resource to be identified; and a secondary retrieval module, configured to, when the preliminary identification result does not meet the preset identification conditions, obtain secondary retrieval constraints through a preset large model, and input the secondary retrieval constraints into the main data source for secondary retrieval, and obtain the media entity information of the media resource to be identified.

[0007] Thirdly, this application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described media information recognition method.

[0008] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned media information recognition method.

[0009] The technical solutions for the aforementioned media information identification method, apparatus, computer equipment, and storage medium include the following steps: acquiring the original path information of the media resource to be identified, and performing hierarchical semantic parsing on the original path information to obtain a set of candidate media names including multiple candidate media names; determining the target identification pattern of the media resource to be identified based on the original path information; inputting each candidate media name in the candidate media name set into a preset main data source for initial retrieval based on the target identification pattern to obtain the preliminary identification result of the media resource to be identified; when the preliminary identification result does not meet the preset identification conditions, obtaining secondary retrieval constraints through a preset large model, and inputting the secondary retrieval constraints into the main data source for secondary retrieval to obtain the media entity information of the media resource to be identified. This method significantly improves the accuracy and robustness of path-driven media identification by integrating hierarchical semantic parsing and large model structured reasoning as a dual retrieval mechanism. Furthermore, by determining the target identification pattern of the media resource to be identified, it enables adaptive identification of media resources from different sources, in different formats, and with different naming conventions, making it particularly suitable for scenarios with massive media resources that are non-standardized in naming, multi-source heterogeneous, and have chaotic naming. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart of a media information identification method according to an embodiment of the present invention; Figure 2 This is a specific flowchart of step S2 in the media information identification method according to an embodiment of the present invention; Figure 3 This is a specific flowchart of step S3 in the media information identification method of one embodiment of the present invention; Figure 4 This is a specific flowchart of step S33 in the media information identification method of an embodiment of the present invention; Figure 5 This is another specific flowchart of step S33 in the media information identification method in one embodiment of the present invention; Figure 6 This is a flowchart illustrating a specific step S4 of the media information identification method in an embodiment of the present invention, involving a secondary retrieval. Figure 7 This is a schematic diagram of a media information recognition device according to an embodiment of the present invention; Figure 8 This is a schematic diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] In one embodiment, such as Figure 1 As shown, a media information identification method is provided, including the following steps: Step S1: Obtain the original path information of the media resource to be identified, and perform hierarchical semantic parsing on the original path information to obtain a set of candidate media names including multiple candidate media names.

[0014] It should be noted that the original path information refers to the complete hierarchical storage path string of the media resource to be recognized, including local file paths, network paths, mounted paths, or resource directory paths, etc. It is the only initial input data source for this solution, and no additional external auxiliary information such as media covers, content frames, text introductions, etc. needs to be collected. Hierarchical semantic parsing is different from single file name parsing, which refers to simultaneously performing linked semantic analysis on file names, parent directory names, intermediate directory names, alias segments, Chinese titles, English titles, episode identifiers, year information, etc. in the original path information, taking into account the semantics of the upper-level classification directory, the semantics of the middle-level classification directory, and the semantics of the bottom-level file name, and realizing the joint extraction of the full-path text. The candidate media name refers to the pure media main body name text obtained by stripping the invalid interference fields from the original path information. The candidate media name only retains the core appellation of the media and does not append subsidiary information such as release year, season / episode number, picture quality parameters, etc. One original path information can correspond to generate one or more synonymous or approximate candidate media names.

[0015] Specifically, the system first reads the complete storage path of the media resource to be recognized stored locally or transmitted over the network, and uses this complete storage path as the original path information; further, it performs hierarchical decomposition and semantic extraction on the original path information, differentiates the directory text and file name string at different levels within the path, and剔除resolution, encoding format, release date watermark, special delimiters and other invalid interference texts, and finally generates multiple candidate media names, and all the candidate media names together constitute the candidate media name set.

[0016] Among them, different from only using the file name, this application extracts semantic information from multiple hierarchical positions of the path, including but not limited to: the current file name or directory name; the parent directory name; the main title in the upper-level directory; Chinese name, English name, translated name, alias; the purified name after removing clarity, suppression group, punctuation, special symbols; the non-season / episode interference title after processing the episode naming format. Through multi-source fusion, deduplication, and priority control, a candidate name set with a limited scale but more comprehensive information coverage is generated, so as to improve the recall rate without significantly increasing the retrieval cost.

[0017] As an example, taking the original path information of the media resource to be recognized as "Film / Movie / The Wandering Earth 2 (2023) 4K Ultra HD.mp4" as an example, first perform hierarchical decomposition on this original path information, splitting it into the first-level directory "Film", the second-level directory "Movie", and the file name "The Wandering Earth 2 (2023) 4K Ultra HD.mp4"; then perform hierarchical semantic parsing,剔除year identifier (2023), picture quality identifier (4K Ultra HD), file suffix (.mp4) and other invalid interference contents, and at the same time supplement synonymous appellations in combination with the directory semantics. Finally, the candidate media names generated include: The Wandering Earth 2, The Wandering Earth Two, and the above two candidate media names together form the candidate media name set corresponding to this recognition.

[0018] This application employs full-path hierarchical semantic parsing, establishing a linkage extraction channel between multi-level directories and filenames. It pre-determines media categories based on upper-level directories while removing various invalid and interfering fields within the path. Furthermore, it supplements synonymous candidate names based on path semantics. On one hand, it fully leverages all valid semantic information inherent in the original path, avoiding information loss caused by single-dimensional parsing. On the other hand, by using multiple candidate names as a fallback, it adapts to complex scenarios such as film and television resource aliases, abbreviations, and non-standard naming, providing sufficient and accurate search keywords for subsequent multi-dimensional data source retrieval. This reduces matching errors in subsequent retrieval and sorting stages from the source, thereby improving the overall robustness of the media identification process.

[0019] Step S2: Determine the target recognition pattern of the media resource to be identified based on the original path information.

[0020] It should be noted that the target recognition mode is a unique retrieval and filtering rule obtained based on path structure feature matching. It is used to constrain the retrieval order of candidate media names in subsequent retrieval processes, as well as the filtering and reordering priority of candidate result sets, so as to realize the differentiated identification of media resources such as movie resources and TV series resources.

[0021] In this embodiment, this step relies on the original path information obtained in step S1 to first extract the path structure features, and then perform identification pattern matching and selection based on the obtained path structure features to complete the preliminary determination of the media identification pattern, providing a rule basis for subsequent differentiated retrieval and candidate result screening and rearrangement.

[0022] Specifically, such as Figure 2 As shown, step S2 includes the following sub-steps: Step S21: Extract features from the original path information to obtain the path structure features of the media resource to be identified.

[0023] It should be noted that path structure features specifically refer to the objective structure and identification features extracted directly from the original path information, without including the semantic content of the path text. They mainly include the number of path directory levels, whether there is a season-specific identifier (S01E01, Season X, Episode X, EP number), whether there is a film and television category directory, whether there is a drama series directory, and other structured identification information. Furthermore, path structure features only represent the external structural rules of the path and are independent of the candidate media names extracted in step S1.

[0024] In this embodiment, structured feature extraction is performed on the complete original path information. It does not rely on other information such as candidate media names or external media data. It extracts fixed structured features only from the directory hierarchy, directory naming rules, and file name identifier fields of the path itself, and summarizes them to obtain the path structure features.

[0025] Step S22: Based on the path structure features, match the target recognition pattern that is compatible with the path structure features from the preset recognition pattern library.

[0026] It should be noted that the recognition pattern library is a pre-configured rule database stored in the system. The library pre-stores two recognition patterns and their corresponding judgment thresholds, retrieval priorities, and candidate pool filtering logic, namely a movie recognition pattern adapted to single film and television resources and a TV series recognition pattern adapted to serialized episode resources.

[0027] In this embodiment, a recognition pattern library containing movie recognition patterns and TV series recognition patterns is pre-built. The two recognition patterns are respectively bound to different search priorities and candidate result filtering rules. The extracted path structure features are compared one by one with the judgment rules corresponding to each pattern in the recognition pattern library to obtain a unique target recognition pattern that is suitable for the current media resource.

[0028] Before the formal search, the media type priority mode is determined based on at least one of the following characteristics: whether it contains a season / episode expression; whether it is a series / episode directory; whether a series naming pattern exists; whether it is a movie resource directory; and whether there is a hierarchical relationship in the directory structure that is more suitable for the TV series mode. When it is determined to be a series / episode identification mode, the season information, first broadcast year, and series title related to the series / episode resource are used first for subsequent search and filtering. This predictive mechanism reduces cross-type misidentification and significantly improves the accuracy and efficiency of subsequent search and filtering stages.

[0029] As an example, the original path information is [Film / Movie / The Wandering Earth 2 (2023) 4K Ultra HD.mp4]. Step S21 extracts the path structure features: the path is a two-level directory structure, without any season episode identifier field, and the top-level directory is labeled with the movie category. Step S22 compares the set of structure features with the recognition pattern library and matches the target recognition pattern as the movie recognition pattern.

[0030] As another example, the original path information is [TV / TV series / Crazy / Season 1 / Episode 06.mp4]. Step S21 extracts the path structure features: the path is a three-level directory structure, the path contains season and episode exclusive identifiers, and the top-level directory is marked with the TV series category; Step S22 matches the target recognition pattern to obtain the TV series recognition pattern.

[0031] In this application, step S2 relies on the objective path structure features inherent in the original path information to complete unsupervised pattern prediction. It does not require calling external data sources or consuming large model computing power, and completes the preliminary differentiation of media categories at low cost. At the same time, by binding differentiated retrieval and filtering rules through an independent recognition pattern library, the subsequent retrieval and candidate result hierarchical rearrangement can be adapted to the inherent characteristics of different media resources. This avoids cross-type misidentification problems from the retrieval rule level, further matches the subsequent multi-candidate pool filtering logic, and effectively improves the overall accuracy of media recognition.

[0032] Step S3: Based on the target recognition pattern, input each candidate media name in the candidate media name set into the preset main data source for initial retrieval to obtain the preliminary recognition results of the media resources to be identified.

[0033] It should be noted that the main data source is a pre-built authoritative media information database within the system, containing structured film and television metadata (such as title, year, season / episode number, cast and crew, director, screenwriter, genre tags, etc.), supporting high-precision matching and structured retrieval. The main data source employs a multi-dimensional joint matching strategy, supporting the synergistic effect of multiple strategies such as fuzzy title matching, year error-tolerant verification, genre tag weighting, and cast and crew association verification. This ensures stable retrieval of highly relevant candidate results even in common noise scenarios such as abbreviations, aliases, translation differences, or inconsistent year labeling. The initial identification result is the optimal retrieval result obtained after the initial search and pattern-based filtering and rearrangement. This result includes a specific score for the corresponding identification confidence level, used to determine whether to trigger the large-scale model structured parsing and secondary search process.

[0034] In this embodiment, this step follows the candidate media name set output in step S1 and the target recognition pattern determined in step S2. Based on the candidate media name set, a search request is initiated to the main data source. The first round of database search is completed using the candidate media name as the search keyword. Then, based on the adapted target recognition pattern, the results are hierarchically constructed to obtain multiple candidate result sets. The multiple candidate result sets are then differentiated and sorted. Finally, the preliminary recognition results of the first round are output, providing a basis for judgment for the secondary search constraints of the subsequent large model.

[0035] Specifically, such as Figure 3 As shown, step S3 includes the following sub-steps: Step S31: Input each candidate media name in the candidate media name set into the main data source for initial retrieval and obtain the candidate retrieval results corresponding to each candidate media name.

[0036] It should be noted that the candidate search results are all the original matching data returned by the main data source after searching for a single candidate media name. They include various results such as exact matching, fuzzy matching, and approximate alias matching. There is a lot of redundant and mismatched data, which cannot be directly used as the recognition result output.

[0037] In this embodiment, each candidate media name in the candidate media name set is used as an independent search keyword and input into a preset main data source in batches to complete the text matching search. The main data source is a dedicated database that stores massive amounts of standardized film and television metadata, containing complete entity information such as the standard name, year, season, cast and crew of compliant movies and TV series across the entire network. For each candidate media name, multiple original search data with different matching degrees are returned, which are the corresponding candidate search results.

[0038] Step S32: Based on each candidate retrieval result, construct multiple candidate result sets.

[0039] It should be noted that the candidate result set is a group of search results categorized according to a unified matching dimension. Each candidate result set only retains data that meets the matching requirements of a single dimension, which is used for subsequent multi-dimensional cross-verification to improve the rigor of result selection.

[0040] In this embodiment, the conventional method of directly selecting the first search result is abandoned. Instead, all candidate search results are uniformly classified and grouped in multiple dimensions. Based on core dimensions such as name consistency, year matching degree, quarterly information matching degree, and name similarity, independent candidate result sets are built, namely, name complete matching set, year matching set, quarterly information matching set, and high similarity name set. All valid search data are fully retained to avoid the direct discarding of valid matching results.

[0041] In other embodiments, a finer-grained matching dimension combination strategy can be adopted to construct a more refined set of candidate results, such as a set of candidate results for the same year, a set of candidate results for the same quarter, a set of candidate results for both the same year and the same quarter, a set of candidate results for exact name matching, a set of candidate results for matching the original English name, and a set of candidate results for similar names.

[0042] Step S33: Based on the target recognition pattern, filter and rearrange the candidate result sets to obtain preliminary recognition results.

[0043] In this embodiment, the target recognition pattern determined in step S2 is invoked, and the corresponding exclusive filtering rules are matched. For different types of media resources, differentiated set intersection judgment logic and result sorting logic are adopted to filter and prioritize multiple sets of candidate results step by step, and finally the single result with the highest confidence is selected as the preliminary recognition result corresponding to this initial search.

[0044] As an example, the media resources to be identified include film resources, and the corresponding target identification pattern includes film identification patterns.

[0045] like Figure 4 As shown, for movie resources that lack season information and rely solely on movie titles and release years for verification, step S33 includes the following sub-steps: Step S331: Based on the movie recognition pattern, combined with the movie title and release year of the movie resources, perform intersection determination on the candidate result sets corresponding to the movie resources to obtain multiple movie candidate results that simultaneously satisfy the movie title and release year.

[0046] In the movie recognition mode, the filtering logic related to season information is masked. Only the intersection of the set of exact name matches and the set of year matches is selected. Search data that simultaneously meet the requirements of matching movie title and release year are retained as movie candidate results. Invalid search results that match the movie title but not the release year, or match the release year but have a different movie title, are removed. The movie title and release year are pre-defined standard fields in the movie metadata, ensuring strict alignment with the main data source structure.

[0047] Step S332: Prioritize each movie candidate result and obtain the movie candidate result with the highest priority as the preliminary identification result.

[0048] The sorting weights are set according to title matching completeness, year deviation value, and data source popularity in that order to obtain the movie sorting score of each movie candidate result. The movie sorting scores are sorted in descending order, and the best candidate result with the highest movie sorting score and the highest confidence (i.e. the movie candidate result with the highest priority) is selected as the preliminary identification result of the movie resource.

[0049] As another example, the media resources to be identified include TV series resources, and the corresponding target identification pattern includes TV series identification patterns.

[0050] like Figure 5 As shown, considering the characteristics of TV series resources that combine three verification dimensions—TV series title, first broadcast year, and season information—step S33 includes the following sub-steps: Step S333: Based on the drama series recognition pattern, and combining the drama series title, first broadcast year and season information of the drama series resources, perform intersection determination on the candidate result sets corresponding to the drama series resources to obtain multiple drama series candidate results that simultaneously meet the drama series title, first broadcast year and season information.

[0051] Specifically, in the TV series recognition mode, the triple-dimensional joint verification logic is enabled. The name exact match set, year match set, and season information match set are retrieved simultaneously for a three-way intersection determination. Only the TV series retrieval results with the same TV series title, premiere year, and season information are retained, avoiding the problem of TV series mis-matching caused by different season information or different premiere years for the same TV series title.

[0052] Step S334: Rearrange the priorities of each TV series candidate result to obtain the TV series candidate result with the highest corresponding priority as the preliminary recognition result.

[0053] Specifically, multi-level sorting weights are set according to the title matching degree, year matching degree, season / episode matching degree, and TV series update completeness. The TV series sorting scores of each TV series candidate result are obtained, and the TV series candidate results are sorted in descending order according to the TV series sorting scores to complete the priority sorting of the TV series candidate results. The optimal candidate result with the highest TV series sorting score and the highest corresponding confidence level (i.e., the TV series candidate result with the highest priority) is selected as the preliminary recognition result corresponding to the TV series resource.

[0054] In some embodiments, a degradation strategy can also be combined. When the confidence level of the optimal candidate result is lower than the confidence threshold or there are ambiguities in the keyword field matching, the backtracking comparison and semantic verification of the sub-optimal candidate set are automatically triggered. For example, perform normalization mapping on abbreviations, aliases, and variant translations in the title. When detecting various season / episode expression forms such as "S01E01", "S1E1", "Season 1 Episode 1", etc., the system automatically normalizes them to the standard structured field "Episode 1 of Season 1"; establish a unified semantic anchor point for TV series with multiple language translations (such as "The Crown", "王冠", "王冠:最终季", etc.), and establish a unified semantic anchor point for multiple language translations, so as to achieve semantic equivalent recognition and result aggregation in cross-language retrieval scenarios.

[0055] In this application, this step relies on the target recognition mode determined in advance to achieve the differential adaptation of the screening logics after the retrieval of two types of media resources, namely movie resources and TV series resources; through grouping multiple candidate result sets, and then performing an intersection determination on multiple candidate result sets, the invalid noise data brought by single-dimensional matching is filtered. At the same time, the year and season / episode associated fields自带 in the path are combined for joint verification, rather than simply relying on name text retrieval; exclusive rearrangement rules are set for different scenarios, fitting the data characteristics of movie resources and TV series resources themselves. It not only solves the problem of messy single-keyword retrieval results and low accuracy, but also联动 with the path structure feature prediction logic in the previous text in a closed loop. The entire first-round recognition is completed only relying on the information obtained from path parsing, without the need to additionally call the computing power of the large model. While ensuring the recognition accuracy, the overall recognition cost is reduced, and it also provides a reliable result and confidence basis for accurately judging whether it is necessary to start the secondary retrieval constraint process of the large model in the future.

[0056] Step S4: If the preliminary identification results do not meet the preset identification conditions, obtain the secondary retrieval constraints through the preset large model, and input the secondary retrieval constraints into the main data source for secondary retrieval to obtain the media entity information of the media resource to be identified.

[0057] It should be noted that the preset identification conditions are the criteria for judging the quality of the initial identification results and are used as the basis for deciding whether to trigger the intervention of the large model. The setting of preset identification conditions needs to balance identification accuracy and system response efficiency, and is usually dynamically calibrated based on statistical analysis of historical mismatch cases. The secondary retrieval constraints are standardized retrieval constraint rules output by the large model after full-domain parsing of the original path information. Compared with unregulated candidate media names, these constraints have stronger retrieval targeting and can adapt to the retrieval format of the main data source without manual intervention. The media entity information is the final, complete, and standardized media identification result output after the initial retrieval, the secondary constraint retrieval by the large model, and cross-confirmation from multiple data sources. This is the final output data of this media identification process.

[0058] In this embodiment, after obtaining the preliminary identification result, it is first determined whether the preliminary identification result meets the preset identification conditions. If the preliminary identification result does not meet the preset identification conditions, the identification failure result is not directly output. Instead, a preset large model is introduced to perform path deep structured analysis, generate standardized secondary search constraints, and carry out accurate secondary search. If a qualified result still cannot be obtained after the secondary search, the auxiliary data source is linked to perform information compensation, and finally mapped back to the main data source to complete unified verification, output compliant and reliable media entity information, and make up for the identification shortcomings of the initial search.

[0059] It should be noted that the auxiliary data source is a backup film and television metadata database independent of the main data source. It is used to supplement media entries that are missing or outdated in the main data source, and serves only as a fallback data source, not as the benchmark for the final output. Preset identification conditions may include the identification confidence level of the preliminary identification result reaching a preset identification confidence level threshold, or the information matching degree between the key information of the preliminary identification result and the original path information reaching a preset information matching degree threshold, or the key fields of the preliminary identification result having non-negligible semantic ambiguity (such as polysemous abbreviations, platform-specific naming conventions, cross-language translation conflicts, etc.), or the preliminary identification result being null or not identifying any candidate results, etc.

[0060] Furthermore, when the confidence level of the initial identification result is lower than the preset confidence level threshold, or when the information matching degree between the key information and the original path information does not reach the preset information matching degree threshold, the original path information is structured and parsed through a preset large model, and structured fields such as Chinese name, English name, year, media type, quarter information or alias information, and quarter collection number are output as secondary search constraints for secondary search.

[0061] Among them, the gated large model triggering mechanism avoids calling the large model for all paths, reducing inference overhead and processing latency; while improving recognition accuracy, the frequency of calling the large model is controlled within a reasonable threshold, thus ensuring recognition robustness while avoiding excessive calling of the large model, and balancing recognition capability and cost control.

[0062] Furthermore, such as Figure 6 As shown, the secondary search includes the following sub-steps: Step S41: Input the secondary search constraints into the main data source to perform a secondary search and obtain the secondary recognition results of the media resources to be identified.

[0063] It should be noted that the secondary identification result is the core output of the secondary retrieval process, and its quality directly determines the feasibility of subsequent compensation and mapping stages. This result not only includes the set of candidate media entries returned by the main data source, but also includes the confidence score of each entry, field integrity markers, and potential ambiguity hints, providing a quantitative basis for whether to initiate auxiliary data source compensation.

[0064] In this embodiment, unlike step S3 which relies on the fuzzy initial search of candidate media names in plain text, this time the secondary search constraints generated by the large model parsing are used as the search basis to initiate a precise constraint search to the main data source, narrowing the search matching range, eliminating irrelevant and redundant search data, and obtaining a secondary recognition result with higher matching accuracy.

[0065] Step S42: If the secondary recognition result still does not meet the preset recognition conditions, call the preset auxiliary data source for compensation and obtain candidate media information of the media resource to be identified.

[0066] It should be noted that the candidate media information is an intermediate product output by the auxiliary data source during the compensation phase. It includes the original media entries and their basic metadata (such as non-standard Chinese names, platform translations, alias sets, rough year, type tags, etc.) crawled from the third-party metadata database. It has not yet been aligned with the standard field system of the main data source and verified for credibility.

[0067] In this embodiment, if the main data source is subject to objective limitations such as delayed data updates, missing entries, and differences in translations, and still cannot output qualified recognition results after a second search, that is, the second recognition results still do not meet the preset recognition conditions, then the single main data source retrieval logic is broken, and a preset third-party auxiliary data source is called. Relying on the film and television metadata supplemented by the auxiliary data source, candidate media information matching the current media resource to be identified is captured to make up for the data blind spots of the main data source.

[0068] Step S43: Map the candidate media information back to the main data source for secondary confirmation to obtain media entity information.

[0069] It should be noted that secondary verification is a crucial step in this solution to achieve data consistency and enhance result credibility. By aligning and mapping candidate media information with the standard field system of the main data source, the uniqueness of its ID, consistency of its name, and validity of its timestamp are verified. Duplicate entries, expired information, or naming conflicts that may exist in the auxiliary data source are eliminated, ensuring that the output results possess both the authority of the main data source and the completeness of the auxiliary data source. Media entity information is the standardized, consumable, and traceable unique media identifier that this solution ultimately outputs. It includes extended fields such as main data source ID, standardized Chinese and English names, release year, media type (movie / TV series / variety show / documentary, etc.), season / episode structured number (e.g., S01E01), alias mapping table, and data traceability markers (indicating the contribution of the main / auxiliary data source), providing high-confidence data support for downstream business scenarios such as content distribution, copyright management, and intelligent recommendation.

[0070] In this embodiment, to avoid identification confusion caused by inconsistent field specifications and data definitions between the main data source and the auxiliary data source, this solution does not directly use the original results of the auxiliary data source. Instead, it reverse maps the candidate media information obtained from the auxiliary data source to the benchmark main data source, completes data format alignment, secondary verification of entries and unified calibration, and finally outputs standardized and unified media entity information, which can reduce the risk of mismatch caused by differences in field systems of different data sources.

[0071] As an example, taking the original path of the TV series resource to be identified as [TV / TV Series / Joy of Life 2 / Episode 03 Uncut Version.mp4] as an example, after the initial search in step S3, due to the influence of TV series aliases and non-standard naming, the confidence level of the initial identification result is lower than the confidence level threshold, and the information matching degree between the key information of the result and the original path information does not meet the standard, and the preset identification conditions are not met; then the large model is triggered to perform structured parsing on the entire original path and generate corresponding secondary search constraints; the secondary search constraints are input into the main data source to complete the secondary search, but it still cannot match the accurate TV series entry; at this time, the auxiliary data source is called to capture the corresponding TV series candidate media information, and then the information is sent back to the main data source to complete the standardization and secondary confirmation, and finally the complete media entity information containing the standard TV series name, the year of first broadcast, and the season information is output.

[0072] In this application, by setting preset recognition conditions with dual-indicator linkage, the quality of the initial recognition results is accurately controlled, preventing false qualified results from being released. At the same time, a gating trigger mechanism is adopted, which only calls the large model when both indicators fail to meet the standards, accurately controlling the frequency of large model calls and reducing computing power costs. In addition, a cascading verification logic is adopted, from secondary retrieval from the main data source to backup compensation from the auxiliary data source, and then backfeeding to the main data source for confirmation. The main data source is always used as the benchmark, which not only makes up for the data missing problem of a single data source, but also avoids the recognition errors caused by multi-source data format conflicts and large model illusions. It completely makes up for the recognition shortcomings of the initial retrieval, further improving the overall accuracy and robustness of media recognition under complex non-standard paths, and forming a complete closed loop with the path parsing, pattern discrimination, and hierarchical retrieval process of the aforementioned steps S1-S3.

[0073] In summary, this application discloses a media information identification method, comprising the following steps: obtaining the original path information of the media resource to be identified, and performing hierarchical semantic parsing on the original path information to obtain a set of candidate media names including multiple candidate media names; determining the target identification pattern of the media resource to be identified based on the original path information; inputting each candidate media name in the candidate media name set into a preset main data source for initial retrieval based on the target identification pattern to obtain the preliminary identification result of the media resource to be identified; when the preliminary identification result does not meet the preset identification conditions, obtaining secondary retrieval constraints through a preset large model, and inputting the secondary retrieval constraints into the main data source for secondary retrieval to obtain the media entity information of the media resource to be identified. This method first uses multi-level path semantic parsing and multi-candidate name generation to more accurately cover non-standard naming scenarios; then, through target identification pattern determination, it is easier to match the correct entity information in the scenarios of drama series, collections, and series resources; at the same time, through the collaborative verification mechanism of the main data source and the large model, it ensures that the identification results meet the business usability standards in the three dimensions of semantic consistency, structural integrity, and data authority, thereby improving the identification accuracy and business usability. Furthermore, this method is easily scalable to large-scale media library processing, making it suitable for batch scraping, periodic scanning, and media library synchronization scenarios. It also outputs structured and more reliable media entity information to support accurate association with subsequent poster images, background images, actor information, category recommendations, continue playback, and favorites systems.

[0074] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0075] In one embodiment, a media information identification device is provided, which corresponds one-to-one with the media information identification method in the above embodiments. For example... Figure 7As shown, the media information identification device includes a collection acquisition module 101, a pattern determination module 102, an initial retrieval module 103, and a secondary retrieval module 104. Detailed descriptions of each functional module are as follows: The set acquisition module 101 is used to acquire the original path information of the media resource to be identified, and to perform hierarchical semantic parsing on the original path information to obtain a set of candidate media names including multiple candidate media names.

[0076] The pattern determination module 102 is used to determine the target recognition pattern of the media resource to be identified based on the original path information. The initial retrieval module 103 is used to input each candidate media name in the candidate media name set into a preset main data source for initial retrieval based on the target recognition pattern, and obtain the preliminary recognition results of the media resources to be identified.

[0077] The secondary retrieval module 104 is used to obtain secondary retrieval constraints through a preset large model when the preliminary identification results do not meet the preset identification conditions, and input the secondary retrieval constraints into the main data source for secondary retrieval to obtain the media entity information of the media resource to be identified.

[0078] Specific limitations regarding the media information recognition device can be found in the limitations of the media information recognition method described above, and will not be repeated here. Each module in the aforementioned media information recognition device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0079] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a media information recognition method.

[0080] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the media information recognition method described in the above embodiment, for example... Figure 1As shown in S1-S4, or Figures 2 to 6 As shown, to avoid repetition, it will not be described again here. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in this embodiment of the media information recognition device, for example... Figure 7 The functions of the set acquisition module 101, pattern determination module 102, initial retrieval module 103, and secondary retrieval module 104 shown are not described again here to avoid repetition.

[0081] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When executed by a processor, the computer program implements the media information recognition method described in the above embodiment, for example... Figure 1 As shown in S1-S4, or Figures 2 to 6 As shown, to avoid repetition, it will not be described again here. Alternatively, when the computer program is executed by the processor, it implements the functions of each module / unit in this embodiment of the media information recognition device, for example... Figure 7 The functions of the set acquisition module 101, pattern determination module 102, initial retrieval module 103, and secondary retrieval module 104 shown are not described again here to avoid repetition. The computer-readable storage medium can be non-volatile or volatile.

[0082] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0083] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0084] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0085] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A media information identification method, characterized in that, Including the following steps: Obtain the original path information of the media resource to be identified, and perform hierarchical semantic parsing on the original path information to obtain a set of candidate media names including multiple candidate media names; Based on the original path information, the target identification pattern of the media resource to be identified is determined; Based on the target recognition mode, each candidate media name in the candidate media name set is input into a preset main data source for initial retrieval to obtain the preliminary recognition result of the media resource to be identified; When the initial identification result does not meet the preset identification conditions, secondary retrieval constraints are obtained through a preset large model, and the secondary retrieval constraints are input into the main data source for secondary retrieval to obtain the media entity information of the media resource to be identified.

2. The media information identification method according to claim 1, characterized in that, The step of determining the target identification pattern of the media resource to be identified based on the original path information includes: Feature extraction is performed on the original path information to obtain the path structure features of the media resource to be identified; Based on the path structure features, a target recognition pattern that matches the path structure features is matched from a preset recognition pattern library.

3. The media information identification method according to claim 1, characterized in that, Based on the target recognition pattern, the process involves inputting each candidate media name in the candidate media name set into a preset main data source for initial retrieval to obtain preliminary recognition results for the media resource to be identified, including: Input each of the candidate media names in the candidate media name set into the main data source for initial retrieval, and obtain the candidate retrieval results corresponding to each candidate media name; Based on the aforementioned candidate retrieval results, multiple candidate result sets are constructed; Based on the target recognition pattern, the candidate result sets are filtered and rearranged to obtain the preliminary recognition result.

4. The media information identification method according to claim 3, characterized in that, The media resources to be identified include film resources, and the target identification pattern includes film identification patterns; the preliminary identification result is obtained by filtering and rearranging the candidate result sets based on the target identification pattern, including: Based on the movie recognition pattern, and combined with the movie title and release year of the movie resource, the intersection of each of the candidate result sets corresponding to the movie resource is determined to obtain multiple movie candidate results that simultaneously satisfy the movie title and release year. The movie candidate results are prioritized and rearranged to obtain the movie candidate result with the highest priority as the preliminary identification result.

5. The media information identification method according to claim 3, characterized in that, The media resources to be identified include TV series resources, and the target identification pattern includes TV series identification patterns; the preliminary identification result is obtained by filtering and rearranging the candidate result sets based on the target identification pattern, including: Based on the TV series identification pattern, and combined with the TV series title, first broadcast year and season information of the TV series resources, the intersection of each set of candidate results corresponding to the TV series resources is determined to obtain multiple TV series candidate results that simultaneously satisfy the TV series title, first broadcast year and season information; The candidate drama series results are prioritized and rearranged to obtain the candidate drama series with the highest priority as the preliminary identification result.

6. The media information identification method according to claim 1, characterized in that, The step of inputting the secondary search constraints into the main data source for secondary search to obtain the media entity information of the media resource to be identified includes: The secondary search constraints are input into the main data source to perform a secondary search, thereby obtaining the secondary identification result of the media resource to be identified. If the secondary recognition result still does not meet the preset recognition conditions, a preset auxiliary data source is invoked for compensation to obtain candidate media information of the media resource to be identified. The candidate media information is mapped back to the main data source for secondary confirmation to obtain the media entity information.

7. The media information identification method according to claim 1, characterized in that, The preset identification conditions include the identification confidence level of the preliminary identification result reaching a preset identification confidence level threshold, or the information matching degree between the key information of the preliminary identification result and the original path information reaching a preset information matching degree threshold. When the preliminary identification result does not meet the preset identification conditions, the secondary retrieval constraints are obtained through a preset large model, including: When the recognition confidence is lower than the recognition confidence threshold, or the information matching degree is lower than the information matching degree threshold, the original path information is structured and parsed through the large model to obtain the secondary retrieval constraints.

8. A media information identification device, characterized in that, include: The set acquisition module is used to acquire the original path information of the media resource to be identified, and to perform hierarchical semantic parsing on the original path information to obtain a set of candidate media names including multiple candidate media names; The pattern determination module is used to determine the target identification pattern of the media resource to be identified based on the original path information. The initial retrieval module is used to input each of the candidate media names in the candidate media name set into a preset main data source for initial retrieval based on the target recognition pattern, and obtain the preliminary recognition result of the media resource to be identified. The secondary retrieval module is used to obtain secondary retrieval constraints through a preset large model when the preliminary identification results do not meet the preset identification conditions, and input the secondary retrieval constraints into the main data source for secondary retrieval to obtain the media entity information of the media resource to be identified.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the media information recognition method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the media information recognition method as described in any one of claims 1 to 7.

Images