Resource retrieval method and device, storage medium and computer program product

Abstract

The embodiment of the invention provides a resource retrieval method and device, a storage medium and a computer program product. The method comprises the following steps: acquiring input information of a user; under the condition that the input information comprises text information, splitting the text information to obtain a plurality of pieces of split text information; retrieving in a resource library based on each piece of split text information to obtain a first candidate resource corresponding to each piece of split text information; a comprehensive similarity result of each first candidate resource and target text information is calculated, target resources matched with the input information are determined from all the first candidate resources according to the comprehensive similarity results, and the target text information comprises the text information and the multiple pieces of split text information. According to the technical scheme, the accuracy and effectiveness of the resource retrieval result can be effectively improved.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the technical field of data retrieval. More specifically, embodiments of the present disclosure relate to a resource retrieval method, device, storage medium, and computer program product. Background Art

[0002] This section aims to provide background or context for the embodiments of the present disclosure recited in the claims. The description herein is not admitted to be prior art merely by virtue of being included in this section.

[0003] Currently, in the resource retrieval scenario, a common retrieval practice is to retrieve and recall the user's input information as a whole. However, this approach often easily recalls interference items with low relevance, that is, it may return a large number of results that do not match the user's actual needs, thereby reducing the accuracy of the retrieval results and the user experience. Summary of the Invention

[0004] In a first aspect of the embodiments of the present disclosure, a resource retrieval method is provided. The method includes:

[0005] Obtain the user's input information;

[0006] When the input information includes text information, split the text information to obtain a plurality of split text information;

[0007] Respectively perform retrieval in the resource library based on each split text information to obtain first candidate resources corresponding to each split text information;

[0008] Respectively calculate the comprehensive similarity results between each first candidate resource and the target text information, and determine the target resource that matches the input information from each first candidate resource according to the comprehensive similarity results. The target text information includes the text information and the plurality of split text information.

[0009] Optionally, the splitting the text information to obtain a plurality of split text information includes at least one of the following: when the format of the text information does not conform to a predefined format, convert the format of the text information to the predefined format and split the converted text information; screen out preset special texts from the text information and filter the special texts to obtain the plurality of split text information, where the special texts include delimiter symbols and special words.

[0010] Optionally, the resource library includes an original resource library, an index resource library, and a resource matching relationship library. Among them, the index resource library is constructed by performing an inverted index on the original resources stored in the original resource library, and the resource matching relationship library is used to record the matching relationships between the original resources; the step of retrieving in the resource library based on each split text information to obtain the first candidate resources corresponding to each split text information respectively includes: retrieving in the original resource library based on each split text information respectively to obtain the original resources corresponding to each split text information respectively as the first candidate resources; retrieving in the index resource library based on each split text information respectively to obtain the associated document lists corresponding to each split text information respectively as the first candidate resources; retrieving in the resource matching relationship library based on each split text information respectively to obtain the associated resources of the resources corresponding to each split text information, and using the associated resources as the first candidate resources, where the associated resources match the resources corresponding to the corresponding split text information.

[0011] Optionally, in the song retrieval scenario, the resource matching relationship library is used to record the matching relationships between the audio fingerprint features of the song resources; the step of retrieving in the resource matching relationship library based on each split text information respectively to obtain the associated resources of the resources corresponding to each split text information respectively includes: when the multiple split text information includes song resource IDs, retrieving in the resource matching relationship library based on the song resource IDs to obtain the associated resources of the resources corresponding to the song resource IDs.

[0012] Optionally, the matching relationships between the audio fingerprint features of the song resources are obtained by comparing the audio fingerprints of the song resources, or by comparing the audio slices of the song resources; among them, the audio fingerprints of the song resources include at least one of cover fingerprints, timbre fingerprints, and spectrum fingerprints.

[0013] Optionally, the step of calculating the comprehensive similarity results between each first candidate resource and the target text information respectively includes: classifying the target text information according to the text dimension to which the target text information belongs, where the text dimension includes the resource name dimension, the resource content producer dimension, and the resource attribution set dimension; for each first candidate resource, calculating the similarity between the first candidate resource and the target text information included in each text dimension respectively to obtain the similarity results corresponding to the first candidate resource in each text dimension respectively; determining the comprehensive similarity result corresponding to the first candidate resource according to the dimension weights of each text dimension and the similarity results corresponding to the first candidate resource in each text dimension respectively.

[0014] Optionally, for each first candidate resource, calculating the similarity between the first candidate resource and the target text information included in each text dimension respectively to obtain the similarity result corresponding to the first candidate resource in each text dimension includes: if there are at least two different resource text information for the first candidate resource in any text dimension, calculating the similarity between the at least two resource text information and the target text information included in the any text dimension respectively, and taking the maximum similarity in the calculation results as the similarity result corresponding to the first candidate resource in the any text dimension.

[0015] Optionally, for each first candidate resource, calculating the similarity between the first candidate resource and the target text information included in each text dimension respectively to obtain the similarity result corresponding to the first candidate resource in each text dimension includes: in the case that there are multiple target text information in any text dimension, calculating the similarity between each target text information in the multiple target text information and the first candidate resource respectively, and determining the statistic of the calculation results as the similarity result corresponding to the first candidate resource in the any text dimension.

[0016] Optionally, calculating the similarity between each target text information in the multiple target text information and the first candidate resource respectively includes: calculating the similarity between each target text information in the multiple target text information and the first candidate resource respectively by using a string edit distance algorithm.

[0017] Optionally, the input information further includes audio information, and the method further includes: retrieving in the resource library based on the audio information to obtain second candidate resources; calculating the audio similarity results between each second candidate resource and the audio information respectively; determining the target resource matching the input information from each first candidate resource according to the comprehensive similarity result includes: determining the target resource from each first candidate resource and the second candidate resources according to the audio similarity result and the comprehensive similarity result.

[0018] Optionally, determining the target resource matching the input information from each first candidate resource according to the comprehensive similarity result includes: screening each first candidate resource according to predefined resource screening conditions; for the screened first candidate resources, taking the first candidate resources with the comprehensive similarity result greater than a preset similarity threshold as the target resources, or sorting the screened first candidate resources in descending order according to the comprehensive similarity result, and taking the preset number of first candidate resources ranked at the front as the target resources.

[0019] Optionally, it further includes: assembling the target resource and its associated information, and returning the assembled information to the user, where the associated information of the target resource includes the creation experience of the target resource and the search popularity of the target resource.

[0020] In the second aspect of the embodiments of the present disclosure, a resource retrieval device is provided, and the device includes:

[0021] An acquisition unit, configured to acquire input information of a user;

[0022] A splitting unit, configured to split the text information to obtain a plurality of split text information when the input information includes text information;

[0023] A retrieval unit, configured to respectively perform retrieval in a resource library based on each split text information to obtain first candidate resources corresponding to each split text information;

[0024] A determination unit, configured to respectively calculate a comprehensive similarity result between each first candidate resource and target text information, and determine a target resource that matches the input information from each first candidate resource according to the comprehensive similarity result, where the target text information includes the text information and the plurality of split text information.

[0025] In the third aspect of the embodiments of the present disclosure, an electronic device is provided, including:

[0026] A processor;

[0027] A memory for storing executable instructions of the processor;

[0028] Wherein, the processor realizes the method as described in any one of the first aspect by running the executable instructions.

[0029] In the fourth aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, on which computer instructions are stored, and when the instructions are executed by a processor, the method as described in any one of the first aspect is realized.

[0030] In the fifth aspect of the embodiments of the present disclosure, a computer program product is provided, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the method as described in any one of the first aspect are realized.

[0031] The above embodiments of the present disclosure have at least the following beneficial effects:

[0032] The disclosed solution splits the input information of the user, and then retrieves based on each split text information respectively to obtain multiple first candidate resources. Then, calculate the comprehensive similarity results of each first candidate resource with the split text information and the input text information, and screen out the target resources from the multiple retrieved first candidate resources according to the comprehensive similarity results.

[0033] Since each split text information contains more detailed and refined content respectively, retrieving based on each split text information respectively can effectively retrieve first candidate resources with higher matching degrees for each split text information, thereby avoiding the introduction of interference items with lower matching degrees or even completely irrelevant ones. Furthermore, screening out the target resources from these first candidate resources with higher matching degrees can effectively ensure the accuracy and effectiveness of the target resources and improve the user's resource retrieval experience. Brief Description of the Drawings

[0034] By referring to the following detailed description with reference to the drawings, the above and other objects, features and advantages of the exemplary embodiments of the present disclosure will become easy to understand. In the drawings, several embodiments of the present disclosure are shown in an exemplary and non-limiting manner, wherein:

[0035] Figure 1 is a schematic flowchart of a process for retrieving resources in a related technology provided by an exemplary embodiment;

[0036] Figure 2 is a schematic system architecture diagram provided by an exemplary embodiment;

[0037] Figure 3 is a schematic flowchart of a resource retrieval method provided by an exemplary embodiment;

[0038] Figure 4 is a schematic diagram of constructing an inverted index provided by an exemplary embodiment;

[0039] Figure 5 is a schematic diagram of resource retrieval provided by an exemplary embodiment;

[0040] Figure 6 is a schematic diagram of determining the matching relationship of song resources provided by an exemplary embodiment;

[0041] Figure 7 is a schematic architecture diagram of a resource retrieval platform provided by an exemplary embodiment;

[0042] Figure 8 is a schematic full-flowchart of a resource retrieval method provided by an exemplary embodiment;

[0043] Figure 9It is a schematic diagram of constructing a resource matching relationship library provided by an exemplary embodiment;

[0044] Figure 10 It is a block diagram of a resource retrieval device provided by an exemplary embodiment;

[0045] Figure 11 It is a schematic diagram of a readable storage medium corresponding to the above method provided by an exemplary embodiment;

[0046] Figure 12 It is a schematic diagram of an electronic device capable of implementing the above method provided by an exemplary embodiment.

[0047] In the drawings, the same or corresponding reference numerals denote the same or corresponding parts. Detailed implementation manners

[0048] Hereinafter, the principles and spirit of the present disclosure will be described with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and then implement the present disclosure, rather than limiting the scope of the present disclosure in any way. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to be able to convey the scope of the present disclosure to those skilled in the art completely.

[0049] Those skilled in the art know that the embodiments of the present disclosure can be implemented as a system, a device, an apparatus, a method, or a computer-readable storage medium. Therefore, the present disclosure can be specifically implemented in the following forms: completely hardware, completely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0050] According to an embodiment of the present disclosure, a resource retrieval method, device, storage medium, and computer program product are provided.

[0051] In this document, it should be understood that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning. And the data involved in the present disclosure can be data authorized by the user or fully authorized by all parties.

[0052] Hereinafter, with reference to several representative embodiments of the present disclosure, the principles and spirit of the present disclosure will be explained in detail.

[0053] Overview of application scenarios

[0054] Currently, in the resource retrieval scenario, the cosine similarity algorithm + inverted index method is usually used for retrieval. Please refer to Figure 1 , Figure 1 It is a schematic flowchart of retrieving resources in the related art provided by an exemplary embodiment. As Figure 1As shown, taking song resources as an example, first, the existing song resources in the song resource library will be preprocessed. The key fields of the existing song resources (such as song name, artist name, album name) will be extracted and an inverted index will be established, thus constructing a corresponding resource index library. In the retrieval stage 101, according to the user's input information (such as song name, artist name or album name), the inverted index will be used to quickly match relevant resources and return documents of multiple eligible alternative song resources. Figure 1 Among them, the alternative song resources include song resource A (Song A, Artist A, Album A), song resource B (Song B, Artist B, Album B), song resource C (Song C, Artist C, Album C), song resource D (Song D, Artist D, Album D), and so on. Then, the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm is used to calculate the relevance of the documents of each alternative song resource. The calculation principle is shown in the following formulas (1) - (2):

[0055]

[0056] idf(t) = 1 + log(docCount / (docFreq + 1)) (2)

[0057] In formula (1), Term represents a certain word, and tf(tind) represents the term frequency of the word Term in document i. The higher the term frequency, the higher the importance of the word Term in document i and the higher the relevance of document i.

[0058] In formula (2), docCount represents the total number of documents in the resource index library, docFreq represents the number of documents containing the word t, and idf(t) represents the inverse document frequency of the word t. The higher the inverse document frequency, the rarer the word t and the higher its importance.

[0059] Through the above process, some alternative song resources with very low relevance can be filtered according to the relevance calculation results.

[0060] Next, enter the data processing stage 102 to process the song text information (including song name, artist name, album name, etc.) of each reserved alternative song resource to obtain the processed song text information. Among them, data processing includes filtering special characters such as stop words that do not contain key information and converting traditional Chinese characters to simplified Chinese characters. Then, in the matching stage 103, calculate the similarity between the user input information and each processed song text information through the cosine similarity algorithm. It should be noted that if the user input information contains a song name, it is necessary to calculate the similarity between the song name in the user input information and the song names contained in each processed song text information respectively to obtain the song name similarity corresponding to each alternative song resource. If the user input information does not contain a song name, there is no need to calculate the song name similarity. By analogy, calculate the artist name similarity and album name similarity corresponding to each alternative song resource. In the screening stage 104, the screening process is as follows:

[0061] S1041: Determine whether there is an alternative song resource with both the corresponding song name similarity and artist name similarity being 100%.

[0062] If there is, it is determined that the alternative song resource matches the user input information successfully. If not, jump to S1042 for further judgment.

[0063] S1042: Determine whether there is an alternative song resource with both the corresponding song name similarity and artist name similarity being greater than 80% and the album name similarity being greater than 60%.

[0064] If there is, jump to S1043 to calculate the overall similarity of the alternative song resource. If not, it is determined that there is no song resource that matches the user input information in this search.

[0065] S1043: Calculate the overall similarity of the alternative song resource.

[0066] Overall similarity = song name similarity + artist name similarity + 0.5 * album name similarity.

[0067] S1044: Sort the alternative song resources in descending order according to their overall similarity, and take the alternative song resource with the largest overall similarity as the song resource that matches the user input information.

[0068] In summary, during the retrieval process of related technologies, treating the user's input information as a whole for retrieval often leads to the recall of many interfering items with low relevance, and it is necessary to use the TF-IDF algorithm to filter the retrieval results. However, relying solely on the filtering mechanism of the TF-IDF algorithm is sometimes still not sufficient to completely eliminate all retrieval results with low relevance. Some content that is only partially related to the user's query or contains too many common words may still be retained, which not only reduces the overall accuracy of the retrieval results but may also affect the user's retrieval experience.

[0069] Thus, in the scenario of resource retrieval, there may be a problem that many interfering items are retrieved, resulting in inaccurate retrieval results.

[0070] It should be noted that the above application scenarios are only shown for the convenience of understanding the spirit and principle of the present disclosure, and the embodiments of the present disclosure are not limited in this regard. On the contrary, the embodiments of the present disclosure can be applied to any applicable scenario. Summary of the Invention

[0072] In view of this, this specification provides a technical solution for resource retrieval. By splitting the user's input information and performing retrieval based on each piece of text information after splitting, and then calculating the comprehensive similarity results corresponding to multiple first candidate resources obtained by the retrieval, this method can avoid recalling interfering items with low or even completely irrelevant matching degrees, effectively ensure the accuracy and effectiveness of the target resources, and improve the user's resource retrieval experience.

[0073] Exemplary Method

[0074] The following will (in conjunction with the system architecture schematic diagram as shown in Figure 2 ) describe the technical concept of this specification in detail through specific embodiments.

[0075]

System Architecture

[0076] In the embodiments of the present disclosure, the terminal may be an electronic device used by the user. Specifically, the terminal may include, but is not limited to, electronic devices with certain computing capabilities such as smart phones, desktop computers, tablet computers, laptop computers, e-book readers, smart watches, and smart bracelets, and the electronic device can run software or websites of categories such as instant messaging, social networking, gaming, and education, and can be applied to scenarios of multi-person audio and video calls.

[0077] The server may include a single server, a server cluster formed by multiple servers, or a cloud server. Specifically, the server may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It may also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0078] The terminal and the server may be directly or indirectly connected through wired communication or wireless communication, and no special limitation is imposed in this disclosure.

[0079] Based on Figure 2 As shown in the system architecture diagram, the resource retrieval system may include a server and a terminal. The terminal may obtain the input information of the user and upload the input information to the server. The server splits the received input information and retrieves in the resource library based on each split text information to obtain multiple first candidate resources. Then, it calculates the comprehensive similarity result between each first candidate resource and the target text information, and determines the target resource from each first candidate resource according to the comprehensive similarity result. The server returns the target resource to the terminal that submitted the input information, so that the terminal can display the target resource to the user.

[0080] In an embodiment of the present disclosure, the resource retrieval system may also include a terminal. The terminal obtains the input information of the user and splits the input information. Further, it retrieves in the resource library based on each split text information to obtain multiple first candidate resources. Then, it calculates the comprehensive similarity result between each first candidate resource and the target text information, and determines the target resource from each first candidate resource according to the comprehensive similarity result. Finally, the terminal displays the target resource to the user.

[0081] Next, an example will be given by taking the terminal to execute the resource retrieval method provided in the embodiment of the present disclosure. Please refer to Figure 3 , Figure 3 which is a flowchart of a resource retrieval method provided by an exemplary embodiment. The method may include the following steps:

[0082] Step 301: Obtain the input information of the user.

[0083] First of all, it should be noted that the present disclosure does not make any special limitations on the source, type, content, input method, etc. of the input information. In some embodiments, the input information may be text information, audio information, link information, video information, image information, and so on. In other embodiments, the input information may be information created or designed by a user (individual or organization) itself, or information intercepted or copied from publicly available information on the network. The present disclosure does not make specific limitations on this. The user can manually input on the corresponding retrieval page of the client, or input through voice (such as playing audio information, oral output, etc.), or other input methods.

[0084] Step 302: When the input information includes text information, split the text information to obtain multiple split text information.

[0085] When the user's input information includes text information, split the text information. The present disclosure does not make specific limitations on the splitting method, the content of the split text information, the text length, etc. Exemplarily, it can be split according to the semantics of the text information, and the text information with semantic association is divided into the same split text information, so as to ensure that each split text information is more compact and coherent semantically. For example, if the input text information is "Nocturne of Xiaohei", splitting according to semantics, we get three split text information: "Xiaohei", "Nocturne", and "of". Or, split according to a preset text length or according to paragraphs, such as splitting according to a text length of 2 characters.

[0086] In some possible implementation manners, splitting the text information may include a combination of one or more of the following shown splitting methods:

[0087] Judge whether the format of the text information conforms to a predefined format. If not, convert the format of the text information to the predefined format and split the converted text information. Among them, the predefined format may include character encoding, font, font size, language (such as Chinese, English, etc.), simplified / traditional Chinese, bold, etc., and the present disclosure does not limit this. By unifying the format of the text information into the predefined format, on the one hand, it can reduce the workload of the system in parsing the text information and speed up the retrieval efficiency. On the other hand, it can eliminate the format differences of the text information, reduce the impact of different formats on the retrieval results, and improve the effect of cross-language retrieval.

[0088] Filter out the preset special texts from the text information. The preset special texts may include delimiter symbols and special words. Among them, delimiter symbols refer to symbols that have the function of separating texts, such as slashes, commas, full stops, semicolons, backslashes, parentheses, and so on. Special words can be understood as words that do not contain key information, such as stop words. It should be noted that for resource retrieval scenarios in different fields, the corresponding special words are different. For example, in the song resource retrieval scenario, "accompaniment", "remix", and "feat" do not contain the key information of song resources and play no role in the process of retrieving songs. Therefore, these three words belong to special words. In the video resource retrieval scenario, "trailer", "behind-the-scenes footage", and "interview" do not carry the key information about the main content of the video. Therefore, these three words belong to special words. In the actual implementation process of the present disclosure solution, special words can be set according to the business field or business requirements, and the present disclosure does not limit this. After filtering out the preset special texts, use the special texts as delimiter identifiers to separate the text information, and then filter out the special texts to obtain multiple split text information. For example, assume the text information is "Nocturne (Xiaohui)". Then the special text is the parentheses. Use the parentheses as the delimiter identifier to separate the text information, obtaining "Nocturne", "(", "Xiaohui", ")". Then filter out the parentheses to obtain the two split text information "Nocturne" and "Xiaohui". Another example, the text information is "Nocturne remix". Filter out the special text as remix. Use remix as the delimiter identifier to separate the text information, obtaining "Nocturne" and "remix". Then filter out remix to obtain the split text information "Nocturne". This method splits the text information according to the special texts, which can remove the useless interference information in the text information, so that the split text information contains as much useful key information as possible, so that when retrieving later, it can better focus on the key information for retrieval and improve the accuracy of retrieval. For example, "Keywords" and "Keywords (feat. Singer A)" are essentially mutually matching song resources. When the user input information is "Keywords (Singer A)", according to the related technology, retrieving based on the entire text information, because the overall similarity of the song names of these two songs is low, it is easy to cause the song resource "Keywords" not to be recalled, resulting in a poor retrieval experience for the user. However, in this embodiment, after splitting the text information according to the special texts, the obtained split text information is "Keywords" and "Singer A". Then retrieving based on these two split text information, it is easy to recall the song resource "Keywords", thus improving the accuracy of retrieval.

[0089] Step 303: Retrieve in the resource library respectively based on each split text information to obtain the first candidate resources corresponding to each split text information.

[0090] There are many retrieval methods, such as keyword retrieval, semantic retrieval, inverted index retrieval, etc. The present disclosure does not specifically limit the retrieval algorithm and retrieval engine. The number of first candidate resources corresponding to each split text information may be one or more.

[0091] In some possible implementation manners, the resource library may include an original resource library, an index resource library, and a resource matching relationship library. Among them, the original resource library is used to store multiple original resources and the resource text information of the original resources (such as resource ID, resource name, resource creator, resource introduction, etc.).

[0092] The index resource library can be constructed by performing an inverted index on the original resources stored in the original resource library. Exemplarily, the resource text information of the original resources is segmented into Chinese words, and then an inverted index is formed according to the segmented data. Figure 4 It is a schematic diagram of constructing an inverted index provided by an exemplary embodiment. As Figure 4 shown, the original resources include four song resources, namely "Those Distant Dreams", "Nocturne", "Dream", and "Distant Road". The resource names of these four song resources are segmented. The segmentation results of "Those Distant Dreams" include "those", "very", "distant", "of", "dream", "far", "away". The segmentation results of "Nocturne" include "nocturne", "night", "piece". The segmentation results of "Dream" include "dream". The segmentation results of "Distant Road" include "distant", "road". Then, an inverted index is constructed based on the foregoing segmentation results to obtain an associated document list corresponding to each segmentation result. Among them, the associated document list corresponding to "night" includes "Those Distant Dreams", "Nocturne", and "Dream". The associated document list corresponding to "distant" includes "Those Distant Dreams" and "Distant Road". The associated document list corresponding to "dream" includes "Those Distant Dreams" and "Dream". The associated document list corresponding to "road" includes "Distant Road". Therefore, what is stored in the index resource library is the associated document list corresponding to the original resources.

[0093] The resource matching relationship library is used to record the matching relationships between original resources. These matching relationships can be established based on multiple criteria, such as content similarity, source relevance, user behavior patterns, functional complementarity, time series association, semantic relevance, etc. Among them, matching based on content similarity means that when two or more original resources have a high degree of similarity in their content, these original resources can be considered to match each other. Matching based on source relevance means that if multiple original resources come from the same source or author, they can be considered to match each other. Matching based on user behavior patterns means that based on the user's browsing history, purchase records, or other interaction data, it is determined that there are matching relationships between certain original resources, such as song resources that are often played together and documents that are often downloaded simultaneously. Matching based on functional complementarity means that some original resources may not be relevant in content, but they are complementary in function, then these original resources can be considered to match each other. Matching based on time series association means that for original resources published in chronological order, such as news reports, blog articles, etc., the subsequently published original resources may update or supplement the previously published original resources, forming a matching relationship in time. Matching based on semantic relevance means that based on the semantics or context of the original resources, it is determined that the original resources match each other. It should be noted that the above are only examples, and the present disclosure does not limit the specific matching criteria / rules.

[0094] The above three databases can be reused and can be applied to multiple business scenarios, thus greatly improving the reuse rate of data assets.

[0095] When performing a search, it is possible to search in at least one of the original resource library, the indexed resource library, and the resource matching relationship library according to each piece of text information after splitting. When searching in the original resource library, the original resources corresponding to each piece of text information after splitting can be retrieved, and these original resources are used as the aforementioned first candidate resources. When searching in the indexed resource library, a list of associated documents corresponding to each piece of text information after splitting can be retrieved, and these lists of associated documents are used as the aforementioned first candidate resources. When searching in the resource matching relationship library, first determine the resources corresponding to each piece of text information after splitting, and then determine the associated resources of the resources corresponding to each piece of text information after splitting according to the matching relationships stored in the resource matching relationship library. Each associated resource matches the resource corresponding to the corresponding piece of text information after splitting. Furthermore, the determined associated resources are used as the aforementioned first candidate resources.

[0096] Please refer to Figure 5 , Figure 5 which is a schematic diagram of a resource retrieval provided by an exemplary embodiment. As Figure 5As shown, the user's input information includes: "Song Name: Suzume (feat. Jumei)", "Singer Name: RADWIMPS (Raduwimpsu) / Jumei", "Album Name: Suzume feat. Jumei". After splitting these input information, the split text information obtained includes: "Suzume", "Jumei", "RADWIMPS", "Raduwimpsu". Then, based on these split text information, retrievals and recalls are respectively performed in the original resource library, the index resource library, and the resource matching relationship library to obtain multiple first candidate resources.

[0097] In this embodiment, by setting multiple resource libraries of different types, retrievals are then performed in these multiple resource libraries. This multi-way recall mechanism can simultaneously obtain first candidate resources from multiple different databases or data sources, realizing the comprehensive utilization of the characteristics and advantages of different data sources, thereby covering a wider data range, enhancing the comprehensiveness and accuracy of retrieval results, helping to reduce the biases or omissions that may be brought about by single data source retrievals, and realizing providing more accurate and diverse retrieval results for users.

[0098] In some possible implementation manners, in the song retrieval scenario, the retrieved resources are song resources. In this scenario, the resource matching relationship library is used to record the matching relationships of song resources. In this embodiment, the matching relationship of song resources specifically refers to the matching relationship between the audio fingerprint features of song resources. The audio fingerprint feature refers to a feature that can identify audio information and has uniqueness. The present disclosure does not make special limitations on the extraction and comparison of audio fingerprint features. When the similarity of the audio fingerprint features of at least two songs reaches a preset similarity threshold, it can be determined that these songs match each other.

[0099] In the song retrieval scenario, when the split text information contains a song resource ID, the retrieval can be performed in the resource matching relationship library according to the song resource ID. The song resource ID refers to a unique identifier assigned to each song resource for distinguishing different song resources. The song resource ID can be composed of numbers, letters, or other characters. Performing a retrieval in the resource matching relationship library can obtain the associated resources of the resources corresponding to the song resource ID, and use this associated resource as the aforementioned first candidate resource.

[0100] In this embodiment, in the song retrieval scenario, the song resource ID is used for retrieval in the resource matching relationship library. On the one hand, since the song resource ID is the unique identifier of each song, it ensures that a specific song can be accurately identified and located, avoiding confusion between songs with similar or the same names, thereby improving the accuracy of the recall in the resource matching relationship library. On the other hand, considering song copyright, using the song resource ID as the retrieval basis can ensure the accuracy and legality of the retrieval operation.

[0101] In some possible embodiments, the matching relationship of song resources can be determined by comparing the audio fingerprints of the song resources. The audio fingerprints of the song resources may include cover fingerprints, timbre fingerprints or spectral fingerprints. The cover fingerprint can be created by analyzing the structural information such as melody, rhythm, and harmony in the song resource. The timbre fingerprint can be determined by the features related to timbre in the audio signal (such as spectral envelope, harmonic structure, etc.). The spectral fingerprint is created by analyzing the energy distribution of the audio signal at different frequencies. By extracting or creating the audio fingerprints of the song resources and comparing these audio fingerprints, it is possible to confirm whether there is a matching relationship between the audio fingerprint features of the song resources according to the comparison results.

[0102] In addition, the matching relationship of song resources can also be determined by comparing the audio slices of the song resources. Exemplarily, for song resource A and song resource B, these two song resources are sliced respectively to obtain a plurality of audio slices. The length of each audio slice can be the same or different. Then, the similarity between the audio slices of song resource A and the audio slices of song resource B is calculated, and the audio slices with a similarity greater than the preset similarity threshold are determined as target audio slices. If the number of target audio slices is greater than the number threshold, it can be determined that there is a matching relationship between song resource A and song resource B. For example, if song resource A has 10 audio slices and song resource B has 11 audio slices, when there are 8 target audio slices, it can be determined that there is a matching relationship between song resource A and song resource B.

[0103] Figure 6 is a schematic diagram of a method for determining the matching relationship of song resources provided by an exemplary embodiment. As Figure 6 shown, for song resource 1 and song resource 2, the cover fingerprints, timbre fingerprints and spectral fingerprints of these two song resources can be obtained respectively. Furthermore, the cover fingerprints of song resources 1 and 2 are compared to determine whether song resources 1 and 2 are different renditions of the same song. Also, the timbre fingerprints of song resources 1 and 2 are compared to evaluate the similarity and difference in timbre between the two. The spectral fingerprints of song resources 1 and 2 are compared to examine the similarities and differences in their frequency distribution characteristics. Then, the comparison results corresponding to the cover fingerprints, the comparison results corresponding to the spectral fingerprints, and the comparison results corresponding to the timbre fingerprints are comprehensively considered to determine whether the fingerprint comparison is passed. In addition, the audio slices of song resources 1 and 2 can be compared, and it is determined whether the slice comparison is passed according to the comparison results. When the fingerprint comparison is passed or the slice comparison is passed, it is determined that song resources 1 and 2 match each other.

[0104] In this embodiment, since the audio fingerprint is used to represent the unique features of the song resource and is not easily tampered with, determining the matching relationship based on the audio fingerprint comparison result can significantly improve the reliability and accuracy. In addition, compared with comparing the similarity of the whole song, determining the matching relationship of the song resource by comparing audio slices with a smaller granularity can not only speed up the comparison, but also help to compare whether the segments of the song resources are similar at the detailed level, rather than just judging whether the two songs are similar as a whole, thus improving the matching accuracy.

[0105] Step 304: Calculate the comprehensive similarity results between each first candidate resource and the target text information respectively, and determine the target resource that matches the input information from each first candidate resource according to the comprehensive similarity results, where the target text information includes the text information and the multiple split text information.

[0106] The target text information includes the text information input by the user and each split text information. For each first candidate resource, it is necessary to calculate the similarity between the first candidate resource and each split text information respectively to obtain multiple first similarities; and calculate the similarity between the first candidate resource and the text information input by the user to obtain a second similarity. Then, according to the second similarity and the multiple first similarities, the comprehensive similarity result corresponding to the first candidate resource is determined. For example, the comprehensive similarity result can be the weighted average, average, median, standard deviation, etc. of the second similarity and the multiple first similarities. The present disclosure does not particularly limit the specific calculation principle of the comprehensive similarity result. Furthermore, the target resource that matches the user input information is determined according to the comprehensive similarity results corresponding to each first candidate resource respectively. The present disclosure does not limit the specific determination method of the target resource. For example, the first candidate resource with the largest comprehensive similarity result can be selected as the target resource, or the first candidate resource with a comprehensive similarity result greater than the similarity threshold can be selected as the target resource.

[0107] In the above embodiment, since each split text information contains more detailed and refined content respectively, retrieving based on each split text information respectively can effectively retrieve first candidate resources with a higher matching degree for each split text information, thereby avoiding introducing interference items with a lower matching degree or even completely irrelevant. Furthermore, screening the target resource from these first candidate resources with a higher matching degree can effectively ensure the accuracy and effectiveness of the target resource and improve the user's resource retrieval experience.

[0108] In some possible embodiments, when calculating the comprehensive similarity result between each first candidate resource and the target text information, the target text information may be classified according to the text dimension to which the target text information belongs. Among them, the text dimension can be set by the user according to actual needs, and the present disclosure does not limit this. Exemplarily, the text dimension may include, but is not limited to, the resource name dimension, the resource content producer dimension, and the resource attribution set dimension. The resource content producer can be understood as the relevant user of the resource content, including but not limited to the production party, the adaptation party, the upload party, and the copyright party of the resource. The attribution set of each resource can be set by the relevant staff according to business requirements.

[0109] After classification, a set of target text information under each text dimension can be obtained. Each set of target text information contains at least one piece of target text information. Furthermore, for each first candidate resource, the similarity between the first candidate resource and the set of target text information under each text dimension is calculated respectively to obtain the similarity result corresponding to the first candidate resource under each text dimension. Combining Figure 5 the illustrated embodiment, in the song retrieval scenario, the text dimension may include the song name dimension, the artist name dimension, and the album name dimension. The target text information included in the sets of target text information under these three text dimensions is shown in Table 1 below:

[0110] Table 1

[0111]

[0112] Taking the calculation of the similarity result corresponding to the first candidate resource in the song name dimension as an example, assuming that a total of 100 first candidate resources are recalled in this search, then the similarity between these 100 first candidate resources and the set of target text information in the song name dimension needs to be calculated respectively to obtain the similarity results corresponding to these 100 first candidate resources in the song name dimension. Similarly, the similarity results corresponding to these 100 first candidate resources in the artist name dimension / album name dimension can be obtained.

[0113] Then, according to the dimension weights of each text dimension and the similarity results corresponding to each first candidate resource under each text dimension, the comprehensive similarity result corresponding to each first candidate resource is determined. Among them, the dimension weights of each text dimension can be set personalized by the resource retrieval platform party or the user with retrieval needs according to actual needs, or can be analyzed and mined based on the historical retrieval records of a large number of users and the historical evaluation of the retrieval results to determine appropriate dimension weights. The calculation principle of the comprehensive similarity result can refer to the following formula (3):

[0114]

[0115]

[0116] In formula (3), S 综合i Represents the comprehensive similarity result corresponding to the i-th first candidate resource, S 资源名称i Represents the similarity result of the i-th first candidate resource in the resource name dimension, W 资源名称维度 The dimension weight that represents the resource name dimension. 资源内容制作者i Represents the similarity result of the i-th first candidate resource in the resource content producer dimension, W 资源内容制作者维度 The dimension weight that characterizes the resource content producer dimension. 资源归属集合i Represents the similarity result corresponding to the i-th first candidate resource in the resource belonging set dimension, W 资源归属集合维度 The dimension weight that represents the dimension of the resource collection. n represents that there are n text dimensions in total.

[0117] In this embodiment, the target text information is classified according to the text dimension, and the similarity results corresponding to each first candidate resource in each text dimension are calculated, and then the comprehensive similarity results corresponding to each first candidate resource are determined in combination with the dimension weights of each text dimension. This method can flexibly adjust the importance of the target text information under different text dimensions to the comprehensive similarity results corresponding to the first candidate resources through dimension weights, thereby improving the accuracy of subsequent screening of target resources based on the comprehensive similarity results, and ensuring that the screened target resources can meet the user's search needs / preferences. In addition, the dimension weights can be dynamically adjusted according to business / service needs or the user groups they are targeting, thereby achieving the output of more accurate and flexible comprehensive similarity results.

[0118] In some possible implementations, for the recalled multiple first candidate resources, when calculating the similarity results corresponding to each first candidate resource in each text dimension, it is possible that the target text information set under any text dimension contains at least two target text information. Please refer to Table 1, taking the target text information set under the song name dimension as an example, the set contains three target text information "すずめ(铃芽)(feat.十明)", "すずめ", and "铃芽". In this case, it is necessary to calculate the similarity between each of the three target text information and the first candidate resource, and determine the statistics of the calculation results as the similarity results corresponding to the first candidate resource in any text dimension. Among them, the statistic can be the weighted average, mean, standard deviation, median, maximum value, etc. of the calculation results. Those skilled in the art can set the determination algorithm of the statistic according to actual needs, and the present disclosure does not limit this.

[0119] For example, taking the similarity result of the first candidate resource in the song name dimension as an example, the principle of determining the statistic can refer to the following formula (4):

[0120]

[0121] In formula (4), S 歌曲名维度i represents the similarity result corresponding to the i-th first candidate resource in the song name dimension, and S 原始歌名相似度i represents the similarity between the i-th first candidate resource and the text information input by the user in the target text information set in the song name dimension. n represents that there are n split text information in the target text information set in the song name dimension, and n 拆分后歌名相似度i represents the similarity between the i-th first candidate resource and the n-th split text information in the target text information set in the song name dimension. The meaning of formula (4) is as follows: calculate the similarity between each split text information in the target text information set in the song name dimension and the i-th first candidate resource respectively, and screen out the maximum similarity from these calculation results. Calculate the similarity between the text information input by the user in the song name dimension and the i-th first candidate resource, perform an averaging process on this similarity and the aforementioned maximum similarity, and use this average value as the similarity result corresponding to the i-th first candidate resource in the song name dimension.

[0122] In this embodiment, for the split text information and the text information input by the user in the same text dimension, the similarities are calculated respectively, and the statistical quantities of these similarities are used as the similarity results corresponding to the first candidate resource in this text dimension. This method comprehensively evaluates the combined influence of the split text information and the text information input by the user on the similarity results corresponding to the first candidate resource, ensuring that the similarity results corresponding to the first candidate resource can more comprehensively and accurately reflect the matching degree between the first candidate resource and the user's retrieval requirements, improving the accuracy of the matching, so as to effectively screen out the resource that best matches the user's retrieval requirements from each first candidate resource subsequently.

[0123] In some possible implementation manners, for multiple recalled first candidate resources, when calculating the similarity results corresponding to each first candidate resource in each text dimension respectively, it is mainly to calculate the similarity between the resource text information of each first candidate resource and the target text information set in the corresponding text dimension. For example, referring to Table 1, assume that a certain recalled first candidate resource a is the song resource "Suzume", and its resource text information includes: the song name "Suzume" and "すずめ", the singer "Shimei", and the album name "すずめ". Then, when calculating the similarity results corresponding to this first candidate resource in each text dimension respectively, it is necessary to calculate the similarity between its resource text information and the target text information set in each text dimension.

[0124] Exemplarily, the resource text information of each first candidate resource can be classified according to the text dimension to obtain a resource text information set of each first candidate resource under each text dimension. Each resource text information set contains at least one resource text information. Of course, if a first candidate resource does not have corresponding resource text information under a certain text dimension, the resource text information set of the first candidate resource under the text dimension is deemed not to exist.

[0125] If a first candidate resource has at least two different resource text information in the resource text information set under any text dimension, it is necessary to calculate the similarity between the at least two resource text information and the target text information set under the corresponding text dimension, and take the largest similarity in the calculation result as the similarity result corresponding to the first candidate resource in any text dimension. Taking the first candidate resource a as an example, its resource text information set under each text dimension is shown in Table 2:

[0126] Table 2

[0127]

[0128] According to Table 2, there are two different resource text information in the resource text information set of the first candidate resource a under the song name dimension. Therefore, when calculating the similarity result corresponding to the first candidate resource a in the song name dimension, it is necessary to calculate the similarity between the two resource text information "すずめ" and "铃芽" and the target text information set under the song name dimension (refer to Table 1), and obtain the similarity corresponding to the resource text information "すずめ" and the similarity corresponding to the resource text information "铃芽". Then, the largest similarity of the two similarities is used as the similarity result corresponding to the first candidate resource a in the song name dimension.

[0129] The above method can effectively avoid the situation where the similarity result is low due to the different languages ​​of the resource text information and the target text information. For example, the target text information is "keyword" and the resource text information is "keyword". According to the matching algorithm in the relevant technology, matching "keyword" and "keyword" will result in a conclusion with a low matching degree, which will lead to filtering the candidate resource corresponding to the resource text information "keyword", making it impossible for the user to obtain the candidate resource. According to the scheme of this embodiment, when the recalled song candidate resources have the original name, alias, and translated name (all belonging to the song name dimension), the similarity between the original name, alias, translated name and the target text information set under the song name dimension can be calculated respectively, and the maximum similarity is selected from the calculation results as the similarity result corresponding to the song candidate resource in the song name dimension. The calculation principle can refer to the following formula (5):

[0130] S 歌曲名维度i = max(S 原名相似度i , S 别名相似度i , S 翻译名相似度i ) (5).

[0131] In formula (5), S 歌曲名维度i represents the similarity result corresponding to the i-th first candidate resource in the song name dimension. S 原名相似度i represents the similarity corresponding to the original name of the i-th first candidate resource in the song name dimension. S 别名相似度i represents the similarity corresponding to the alias of the i-th first candidate resource in the song name dimension. S 翻译名相似度i represents the similarity corresponding to the translated name of the i-th first candidate resource in the song name dimension. Obviously, through the above method, it is possible to more accurately obtain the similarity results corresponding to the first candidate resources in each text dimension respectively, and avoid filtering out candidate resources with a relatively high actual matching degree.

[0132] In some possible implementation manners, when calculating the similarity between the target text information and the first candidate resource, the string edit distance algorithm can be used to calculate the similarity between the target text information and the resource text information of the first candidate resource. The string edit distance algorithm refers to the minimum number of edit operations required to convert one string into another between two strings. The allowed edit operations include replacing, inserting, and deleting characters. Generally speaking, the smaller the edit distance, the higher the similarity between the two strings. The principle of the string edit distance algorithm can be referred to the following formula (6):

[0133] S 相似度 = 1 - distance ÷ max(left size - right size ) (6)

[0134] In formula (6), S 相似度 represents the similarity between any target text information and any resource text information, distance represents the edit distance between the aforementioned target text information and the resource text information, left size represents the text length of the aforementioned target text information, and right size represents the text length of the aforementioned resource text information.

[0135] In a retrieval scenario that focuses on or depends on the word order in text information, compared with the cosine similarity algorithm, using the string edit distance algorithm can obtain more accurate similarity results. Because the cosine similarity algorithm essentially calculates the cosine of the angle between two N-dimensional vectors in an N-dimensional space. This method does not consider the arrangement order of words in the text when evaluating similarity, but only focuses on the word segmentation results of the text and their corresponding word frequency distributions. For example, for the two pieces of text information "Are you" and "You are", when calculating the similarity using the cosine similarity algorithm, the similarity is 1. While the string edit distance algorithm is sensitive to the arrangement order of characters. When calculating the similarity of the two pieces of text information "Are you" and "You are" using the string edit distance algorithm, the similarity is 0. Therefore, in a retrieval scenario such as song retrieval that is sensitive to the word order of text information, calculating the similarity between the target text information and the resource text information of the first candidate resource using the string edit distance algorithm can obtain more accurate similarity results, so as to return resources that better meet the user's retrieval needs to the user.

[0136] In some possible implementation manners, the user's input information may include audio information. The audio information may be audio created or output by the user himself, or audio information intercepted, copied or recorded from a public website. The present disclosure does not limit this. When it is detected that the user's input information contains audio information, retrieval can be performed in the candidate audio stored in the resource library based on this audio information to obtain a second candidate resource. It should be noted that the present disclosure does not make a special limitation on the specific retrieval manner of the second candidate resource. Then, calculate the audio similarity results between each second candidate resource and the audio information input by the user respectively. Exemplarily, audio fingerprint feature comparison, audio slice comparison or other audio similarity calculation methods can be used to calculate the audio similarity results. The present disclosure does not limit this.

[0137] Furthermore, the target resource can be determined from the retrieved first candidate resources and second candidate resources according to the comprehensive similarity results respectively corresponding to each first candidate resource and the audio similarity results respectively corresponding to each second candidate resource.

[0138] This embodiment combines two forms of data, namely text information and audio information, as the user's input, and can achieve more accurate resource retrieval. When the user submits input information, the retrieval system will not only analyze the provided text information, but also process the attached audio information to extract key information from these two forms of information respectively. Such a dual retrieval mechanism can help the user find resources that are more matching and accurate to the input information. This method makes full use of information in different formats and improves the comprehensiveness and accuracy of the retrieval results.

[0139] In some possible implementation manners, during the process of screening target resources from various first candidate resources, screening can be performed according to predefined resource screening conditions. The predefined resource screening conditions can be set by the user according to business requirements, or can be personalized according to their own preferences / historical behavior patterns. The present disclosure does not limit this. For example, in a song retrieval scenario, the predefined resource screening conditions can include whether the song resource is online and the recent popularity of the song resource. Further, for the screened first candidate resources, the first candidate resources with a comprehensive similarity result greater than a preset similarity threshold are used as target resources. Or, the screened first candidate resources are sorted in descending order according to the comprehensive similarity result, and the first candidate resources with a preset number at the front of the ranking are used as target resources. It should be noted that for each recalled first candidate resource, it is also possible to first screen out the first candidate resources with a comprehensive similarity result greater than the preset similarity threshold, or screen out the first candidate resources with a preset number at the front of the ranking. Further, secondary screening is performed on the screened first candidate resources according to the predefined resource screening conditions to obtain target resources.

[0140] In the above manner, screening the first candidate resources by using the predefined resource screening conditions and the preset similarity threshold / preset number helps to filter out the first candidate resources that do not meet the predefined resource screening conditions, and at the same time screen out the target resources with a high matching degree with the user input information for return. It realizes meeting the personalized matching needs of users while ensuring the accuracy of the retrieval results.

[0141] In some possible implementation manners, after screening out the target resources, the target resources and their associated information can be assembled. The associated information can include but is not limited to the creation experience and search popularity of the target resources. In some embodiments, when each first candidate resource is recalled, its associated information is recalled together. In this case, the target resources and their associated information can be directly assembled. In other embodiments, in the case of only recalling the first candidate resources, the corresponding associated information can be obtained from the resource library according to the target resources. The present disclosure does not particularly limit the acquisition of the associated information and the specific implementation manner of the assembly. Further, the assembled information is returned to the user. In this way, not only can the target resources matching the user's retrieval requirements be returned to the user, but also additional resource information can be returned to help the user further understand the target resources, which helps to improve the user's satisfaction.

[0142] Figure 7 is a schematic structural diagram of a resource retrieval platform provided by an exemplary embodiment. As Figure 7 shown, taking song resources as an example, the song resource retrieval platform can be applied to business scenarios such as sound quality improvement, key content supplementary recording, copyright detection, playlist import, cloud disk matching, and artist song supply-demand difference.

[0143] The support domain architecture of the song resource retrieval platform may include the following components or services: off-site library, AIO (All-In-One Optimization), audio algorithm module, and Tianyan. Among them, the off-site library refers to an external database or storage system for storing and managing external song resource data. AIO refers to an optimization solution integrating multiple functions and services. The audio algorithm module is used to integrate algorithms for processing audio data, including audio recognition, audio conversion, audio enhancement, etc. Tianyan is used for real-time monitoring, data analysis, or security protection.

[0144] The infrastructure of the song resource retrieval platform may include the following components or services: DDB (Distributed Database), Hbase, nydus, atomic algorithm module, Pandora, KS, and data warehouse. Among them, DDB can be used to store and manage large-scale data. KS can be used to store and manage keys.

[0145] The song resource retrieval platform of this embodiment can be used in retrieval scenarios such as retrieving in-site songs off-site, retrieving off-site songs in-site (both of which involve the matching between in-site and off-site songs), and pure audio retrieval. Specifically, the song resource retrieval process can be divided into four stages: recall, matching, sorting, and return.

[0146] The recall stage may include a splitting module 701, an audio fingerprint extraction module 702, a song library search module 703, a main site search module 704, a text recall module 705, an audio recall module 706, and a relationship library matching module 707. Among them, (1) for the text information input by the user: the splitting module 701 is used to split the text information input by the user to obtain multiple split text information. The main site search module 704 is used to retrieve in the original song resource library according to each split text information. The relationship library matching module 707 is used to retrieve in the song resource matching relationship library according to the audio ID in the input text information. The song resource matching relationship library stores the matching relationship of the audio fingerprint features of in-site and off-site songs. The text recall module 705 is used to recall the first candidate resources corresponding to each split text information. (2) For the audio information input by the user: the audio fingerprint extraction module 702 is used to extract the audio fingerprint of the audio information input by the user. The song library search module 703 is used to perform audio retrieval according to the extracted audio fingerprint. The audio recall module 706 is used to recall the second candidate resources corresponding to the input audio information.

[0147] The matching stage may include a text matching module 708, a text matching algorithm module 709, a cover song algorithm module 710, a same song algorithm module 711, a lyrics matching algorithm module 712, and a song recognition algorithm module 713. Among them, the text matching algorithm module 709 integrates multiple text matching algorithms. The text matching module 708 is used to call a text matching algorithm from the text matching algorithm module 709 to calculate the comprehensive similarity result between each first candidate resource and the target text information. Exemplarily, the similarity between the first candidate resource and the set of target text information under each text dimension can be calculated first, and then the comprehensive similarity result corresponding to the first candidate resource can be obtained by combining the dimension weights of each text dimension. The cover song algorithm module 710 is used to determine whether the recalled second candidate resource is a cover version of another song resource. The same song algorithm module 711 is used to determine whether the recalled second candidate resource is the same resource as another song resource. The song recognition algorithm module 713 is used to calculate the audio similarity result between the audio slice of the recalled second candidate resource and the audio information slice input by the user. The lyrics matching algorithm module 712 is used to extract the main verse lyrics and chorus lyrics of each second candidate resource respectively, and use the string edit distance algorithm to calculate the similarity corresponding to the main verse lyrics and the similarity corresponding to the chorus lyrics respectively, and then calculate the overall similarity of the second candidate resource according to the weight of the main verse lyrics and the weight of the chorus lyrics.

[0148] The sorting stage may include a text similarity sorting module 714, an audio matching degree sorting module 715, a text and audio sorting module 716, and a result set screening module 717. Among them, the text similarity sorting module 714 is used to sort the first candidate resources according to the comprehensive similarity results corresponding to each first candidate resource respectively. The audio matching degree sorting module 715 is used to sort the second candidate resources according to the audio similarity results corresponding to each second candidate resource respectively. The text and audio sorting module 716 is used to perform a comprehensive sorting of each first candidate resource and the second candidate resource according to the comprehensive similarity result and the audio similarity result. The result set screening module 717 is used to screen each first candidate resource and the second candidate resource according to predefined resource screening conditions, or screen out the first candidate resources / second candidate resources whose comprehensive similarity results / audio comprehensive similarity results are greater than the preset similarity threshold as the target resources.

[0149] The return phase may include a song information supplement module 718, a copyrighted replacement module 719, a same song search module 720, a cover song search module 721, and a similar song search module 722. Among them, the song information supplement module 718 is used to assemble the target resource and its associated information together and return the assembled information to the user. The copyrighted replacement module 719 is used to retrieve a copyrighted resource that matches the target resource within the station when the station does not own the copyright of the target resource, and return the target resource after replacing it with the copyrighted resource. The same song search module 720 is used to retrieve resources within the station that are the same as the target resource (where "the same" means the similarity is greater than or equal to a first preset threshold), and return the same resource and the target resource together. The cover song search module 721 is used to retrieve cover versions of the target resource within the station and return the cover version and the target resource together. The similar song search module 722 is used to retrieve resources within the station that are similar to the target resource (where "similar" means the similarity is greater than a second preset threshold and less than the first preset threshold, and the second preset threshold is less than the first preset threshold), and return the similar resource and the target resource together. In this way, on the basis of returning the target resource, additional resources / information can be returned to the user, which helps to improve the user's retrieval experience.

[0150] Figure 8 is a full - process schematic diagram of a resource retrieval method provided by an exemplary embodiment. Please refer to Figure 8 Taking song resources as an example, first, the existing song resources in the original song resource library are pre - processed. The keyword fields of the existing song resources (such as song name, artist information, album information) are extracted and an inverted index is established, thereby constructing a corresponding index resource library. In this embodiment, the resources stored in the index resource library include song resource A (Song A, Artist A, Album A), song resource B (Song B, Artist B, Album B), song resource C (Song C, Artist C, Album C), song resource D (Song D, Artist D, Album D), and so on.

[0151] Start resource retrieval and enter the recall phase 801. The process of this phase is as follows:

[0152] S8011: Perform simplified - traditional Chinese conversion on the text information in the user - input information.

[0153] In this embodiment, the predefined format is simplified Chinese. Therefore, when the text information input by the user is in traditional Chinese, the text information needs to be converted to simplified Chinese.

[0154] S8012: Filter out the special vocabulary included in the text information.

[0155] In this embodiment, the special vocabulary may include "accompaniment", "remix", "feat".

[0156] S8013: Split the text information based on the parentheses contained in the text information.

[0157] S8014: Split the text information based on delimiter symbols.

[0158] S8015: Obtain multiple pieces of split text information.

[0159] S8016: Multi-channel recall.

[0160] Retrieve in the original resource library, index resource library, and resource matching relationship library respectively according to each piece of split text information. Among them, for the construction process of the resource matching relationship library, refer to Figure 9 the embodiments shown, which will not be elaborated here for the time being.

[0161] S8017: Obtain multiple first candidate resources.

[0162] Then, enter the matching stage 802, which mainly calculates the comprehensive similarity results between each first candidate resource and the target text information respectively.

[0163] First, classify the target text information into the target text information under the song name dimension, the target text information under the artist name dimension, and the target text information under the album name dimension according to the text dimension. Similarly, classify the resource text information of each first candidate resource into the resource text information under the song name dimension, the resource text information under the artist name dimension, and the resource text information under the album name dimension according to the text dimension.

[0164] Then, calculate the similarity scores corresponding to each first candidate resource in the song name dimension, artist name dimension, and album name dimension respectively. Taking the similarity score in the song name dimension as an example, determine whether the resource text information of each first candidate resource under the song name dimension contains the song name, song name alias, or song name translation of the first candidate resource. Calculate the similarity scores between the song name, song name alias, and song name translation and the target text information under the song name dimension respectively to obtain the song name similarity score, song name alias similarity score, and song name translation similarity score. Then, calculate the similarity score corresponding to the first candidate resource in the song name dimension (abbreviated as "total song name similarity score") according to the song name similarity score, song name alias similarity score, and song name translation similarity score. Exemplarily, the total song name similarity score of each first candidate resource can be obtained through the following formula (7):

[0165]

[0166] In formula (7), S 歌曲名维度i represents the total song name similarity score of the i-th first candidate resource. S 原歌名iCharacterize the similarity between the song name of the i-th first candidate resource and the song name in the text information input by the user. n represents that there are n split song names in the set of target text information in the song name dimension. n 拆分后歌名i Characterize the similarity between the song name of the i-th first candidate resource and the n-th split song name in the set of target text information.

[0167] S 原歌名别名i Characterize the similarity between the alias of the song name of the i-th first candidate resource and the song name in the text information input by the user. n 拆分后歌名别名i Characterize the similarity between the alias of the song name of the i-th first candidate resource and the n-th split song name in the set of target text information.

[0168] And so on, S 原歌名翻译名i Characterize the similarity between the translated name of the song name of the i-th first candidate resource and the song name in the text information input by the user. n 拆分后歌名翻译名i Characterize the similarity between the translated name of the song name of the i-th first candidate resource and the n-th split song name in the set of target text information. Due to the length of the space, this part of the content is not shown in formula (7).

[0169] Similarly, according to the song name similarity score, album name alias similarity score, and album name translated name similarity score, the similarity score corresponding to the first candidate resource in the album name dimension (abbreviation: "total album name similarity score") can be calculated. And, according to the artist name similarity score, artist name alias similarity score, and artist name translated name similarity score, the similarity score corresponding to the first candidate resource in the artist name dimension (abbreviation: "total artist name similarity score") can be calculated.

[0170] Furthermore, a lyrics similarity algorithm can be introduced to calculate the similarity score between the lyrics of each first candidate resource and the target text information (abbreviation: "lyrics similarity score").

[0171] Furthermore, based on the total song name similarity score, total album name similarity score, total artist name similarity score, and lyrics similarity score of the first candidate resource, the comprehensive similarity result corresponding to each first candidate resource is determined.

[0172] In addition, in the case where the information input by the user includes audio information, multiple second candidate resources can be recalled in the recall stage 801. Furthermore, in the matching stage 802, the audio fingerprint of the audio information input by the user can be compared with the audio fingerprint of the second candidate resource to obtain the audio similarity result corresponding to each second candidate resource. The audio fingerprint includes cover fingerprint, timbre fingerprint, and spectrum fingerprint.

[0173] Then, enter the sorting stage 803, and the process of this stage is as follows:

[0174] S8031: Screen each first candidate resource and second candidate resource according to predefined resource screening conditions.

[0175] S8032: Weight the first candidate resources and second candidate resources that meet specific conditions.

[0176] The specific conditions can be set by the user according to business requirements, including but not limited to: resources with recent popularity exceeding a preset popularity threshold. For the first candidate resources and second candidate resources that meet the specific conditions, assign a preset weight to the comprehensive similarity result corresponding to the first candidate resource to obtain a new comprehensive similarity result; and assign a preset weight to the audio similarity result corresponding to the second candidate resource to obtain a new audio similarity result.

[0177] S8033: Sort the first candidate resources and second candidate resources in descending (or ascending) order according to the latest comprehensive similarity results and audio similarity results.

[0178] S8034: Use the candidate resources with comprehensive similarity results or audio similarity results greater than the preset similarity threshold as target resources.

[0179] S8035: Assemble the target resources and their associated information.

[0180] Furthermore, return the assembled information to the user.

[0181] Figure 9 It is a schematic diagram of building a resource matching relationship library provided by an exemplary embodiment. Please refer to Figure 9 , still taking song resources as an example, extract the top 2 million popular songs with heat values from the original resource library within the station. Then extract the cover fingerprints, timbre fingerprints, and spectral fingerprints of these popular songs respectively. For each song stored in the original resource library outside the station, extract the cover fingerprint, timbre fingerprint, and spectral fingerprint of each song. Then, compare the cover fingerprints, timbre fingerprints, and spectral fingerprints of each popular song within the station with those of each song outside the station to obtain the corresponding cover fingerprint scores, timbre fingerprint scores, and spectral fingerprint scores. Furthermore, based on the cover fingerprint scores, timbre fingerprint scores, and spectral fingerprint scores and the preset determination rules, determine whether each popular song within the station matches the songs outside the station. Among them, the preset determination rules may include: when the statistics of the cover fingerprint scores, timbre fingerprint scores, and spectral fingerprint scores are greater than the preset similarity score threshold, determine that the popular song within the station participating in the comparison matches the song outside the station. It should be noted that the matching relationships stored in the resource matching relationship library are not fixed and can be updated regularly or based on changes in song resources.

[0182] Exemplary device

[0183] In an exemplary embodiment of the present disclosure, a resource retrieval device is further provided.

[0184] Please refer to Figure 10 , Figure 10 which is a block diagram of a resource retrieval device provided by an exemplary embodiment.

[0185] As Figure 10 shown, the resource retrieval device 1000 may include: an acquisition unit 1001, a splitting unit 1002, a retrieval unit 1003, and a determination unit 1004. Among them:

[0186] The acquisition unit 1001 is configured to acquire the input information of the user.

[0187] The splitting unit 1002 is configured to split the text information to obtain a plurality of split text information when the input information includes text information.

[0188] The retrieval unit 1003 is configured to respectively perform retrieval in the resource library based on each split text information to obtain first candidate resources corresponding to each split text information.

[0189] The determination unit 1004 is configured to respectively calculate the comprehensive similarity results between each first candidate resource and the target text information, and determine a target resource that matches the input information from each first candidate resource according to the comprehensive similarity results, where the target text information includes the text information and the plurality of split text information.

[0190] In an embodiment, the splitting unit 1002 is specifically configured to: when the format of the text information does not conform to a predefined format, convert the format of the text information to the predefined format and split the converted text information; or, screen out preset special texts from the text information and filter the special texts to obtain the plurality of split text information, where the special texts include delimiter symbols and special words.

[0191] In an embodiment, the resource library includes an original resource library, an index resource library, and a resource matching relationship library. Among them, the index resource library is constructed by performing an inverted index on the original resources stored in the original resource library, and the resource matching relationship library is used to record the matching relationships between the original resources; the retrieval unit 1003 is specifically configured to:

[0192] Retrieve in the original resource library respectively based on each piece of split text information, and obtain the original resources corresponding to each piece of split text information respectively as the first candidate resources; retrieve in the indexed resource library respectively based on each piece of split text information, and obtain the associated document lists corresponding to each piece of split text information respectively as the first candidate resources; retrieve in the resource matching relationship library respectively based on each piece of split text information, and obtain the associated resources of the resources corresponding to each piece of split text information, and use the associated resources as the first candidate resources, where the associated resources match the resources corresponding to the corresponding split text information.

[0193] In one embodiment, in a song retrieval scenario, the resource matching relationship library is used to record the matching relationships between the audio fingerprint features of song resources; the retrieving in the resource matching relationship library respectively based on each piece of split text information to obtain the associated resources of the resources corresponding to each piece of split text information includes: when the multiple pieces of split text information include song resource IDs, retrieving in the resource matching relationship library based on the song resource IDs to obtain the associated resources of the resources corresponding to the song resource IDs.

[0194] In one embodiment, the matching relationships between the audio fingerprint features of the song resources are obtained by comparing the audio fingerprints of the song resources, or by comparing the audio slices of the song resources; wherein, the audio fingerprints of the song resources include at least one of cover fingerprints, timbre fingerprints and spectrum fingerprints.

[0195] In one embodiment, the determining unit 1004 is specifically configured to: classify the target text information according to the text dimension to which the target text information belongs, and the text dimension includes a resource name dimension, a resource content producer dimension, and a resource attribution set dimension; for each first candidate resource, calculate the similarity between the first candidate resource and the target text information included in each text dimension respectively, and obtain the similarity result corresponding to the first candidate resource in each text dimension respectively; determine the comprehensive similarity result corresponding to the first candidate resource according to the dimension weights of each text dimension and the similarity results corresponding to the first candidate resource in each text dimension respectively.

[0196] In one embodiment, for each first candidate resource, calculating the similarity between the first candidate resource and the target text information included in each text dimension respectively, to obtain the similarity result corresponding to the first candidate resource in each text dimension, includes: if there are at least two different resource text information of the first candidate resource in any text dimension, calculating the similarity between the at least two resource text information and the target text information included in the any text dimension respectively, and taking the maximum similarity in the calculation results as the similarity result corresponding to the first candidate resource in the any text dimension.

[0197] In one embodiment, for each first candidate resource, calculating the similarity between the first candidate resource and the target text information included in each text dimension respectively, to obtain the similarity result corresponding to the first candidate resource in each text dimension, includes: in the case that there are multiple target text information in any text dimension, calculating the similarity between each target text information in the multiple target text information and the first candidate resource respectively, and determining the statistic of the calculation results as the similarity result corresponding to the first candidate resource in the any text dimension.

[0198] In one embodiment, calculating the similarity between each target text information in the multiple target text information and the first candidate resource respectively, includes: calculating the similarity between each target text information in the multiple target text information and the first candidate resource respectively by using a string edit distance algorithm.

[0199] In one embodiment, the input information further includes audio information, and the apparatus further includes:

[0200] An audio unit 1005, configured to perform a search in the resource library based on the audio information to obtain a second candidate resource; calculating the audio similarity result between each second candidate resource and the audio information respectively;

[0201] The determining unit 1004 is specifically configured to: determine the target resource from each first candidate resource and the second candidate resource according to the audio similarity result and the comprehensive similarity result.

[0202] In one embodiment, the determining unit 1004 is specifically configured to: screen each first candidate resource according to a predefined resource screening condition; for the screened first candidate resources, taking the first candidate resources with the comprehensive similarity result greater than a preset similarity threshold as the target resources, or sorting the screened first candidate resources in descending order according to the comprehensive similarity result, and taking the first preset number of first candidate resources ranked at the top as the target resources.

[0203] In one embodiment, the apparatus further includes:

[0204] An assembly unit 1006 for assembling the target resource and its associated information and returning the assembled information to the user, where the associated information of the target resource includes the creation experience of the target resource and the search popularity of the target resource.

[0205] The specific details of each module of the above resource retrieval device 1000 have been described in detail in the previously described resource retrieval method process. Therefore, they will not be elaborated here.

[0206] It should be noted that although several modules or units of the resource retrieval device 1000 are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of the two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0207] Exemplary medium

[0208] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is further provided, on which a program product capable of implementing the above method of this specification is stored. In some possible embodiments, various aspects of the present disclosure can also be implemented in the form of a program product, which includes program code. When the program product runs on a terminal device, the program code is used to cause the terminal device to execute the steps according to various exemplary embodiments of the present disclosure described in the above "Exemplary Method" section of this specification.

[0209] Please refer to Figure 11 , Figure 11 which is a schematic diagram of a readable storage medium corresponding to a resource retrieval method provided by an exemplary embodiment.

[0210] Refer to Figure 11 As shown, a readable storage medium 1100 for implementing the above method according to an embodiment of the present disclosure is described. It can use a portable compact disc read-only memory (CD-ROM) and include program code, and can run on a terminal device, such as a personal computer. However, the readable storage medium of the present disclosure is not limited thereto. In the present disclosure, the readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, device, or device.

[0211] The readable storage medium may adopt any combination of one or more readable media. The readable media may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the readable storage medium (a non-exhaustive list) include: an electrical connection having one or more wires, a portable 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.

[0212] The computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which the readable program code is carried. Such a propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable signal medium may also be any readable medium other than the readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device.

[0213] The program code contained on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical fiber cable, RF, etc., or any suitable combination of the above.

[0214] The program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, C++, etc., and also including conventional procedural programming languages such as the "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, executed as a stand-alone software package, partially on the user computing device and partially on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0215] Exemplary electronic device

[0216] In an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above resource retrieval method is further provided.

[0217] Please refer to Figure 12 , Figure 12It is a schematic diagram of an electronic device capable of implementing the above method provided by an exemplary embodiment.

[0218] Reference will now be made to Figure 12 to describe the electronic device 1200 according to this embodiment of the present disclosure. Figure 12 The electronic device 1200 shown is merely an example and should not impose any limitation on the functions and scope of use of the embodiments of the present disclosure.

[0219] As Figure 12 shown, the electronic device 1200 is presented in the form of a general-purpose computing device. The components of the electronic device 1200 may include, but are not limited to: at least one of the above processing units 1201, at least one of the above storage units 1202, and a bus 1203 connecting different system components (including the storage unit 1202 and the processing unit 1201).

[0220] Among them, the storage unit stores program code, and the program code can be executed by the processing unit 1201, so that the processing unit 1201 executes the steps of the various embodiments described above in this specification.

[0221] The storage unit 1202 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 12021 and / or a cache storage unit 12022, and may further include a read-only storage unit (ROM) 12023.

[0222] The storage unit 1202 may also include a program / utility 12024 having a set (at least one) of program modules 12025. Such program modules 12025 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. The reality of a network environment may be included in each or some combination of these examples.

[0223] The bus 1203 may represent one or more of several types of bus structures, including a storage unit bus or a storage unit controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any bus structure in a variety of bus structures.

[0224] The electronic device 1200 can also communicate with one or more external devices 1204 (such as a keyboard, a pointing device, a Bluetooth device, etc.), and can also communicate with one or more devices that enable a user to interact with the electronic device 1200, and / or communicate with any device that enables the electronic device 1200 to communicate with one or more other computing devices (such as a router, a modem, etc.). Such communication can be carried out through the input / output (I / O) interface 1205. Moreover, the electronic device 1200 can also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through the network adapter 1206. As shown in the figure, the network adapter 1206 communicates with other modules of the electronic device 1200 through the bus 1203. It should be understood that although Figure 12 not shown in the figure, other hardware and / or software modules can be used in combination with the electronic device 1200, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0225] Exemplary computer program product

[0226] In an exemplary embodiment of the present disclosure, there is also provided a computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the method described in any of the above embodiments are implemented.

[0227] Through the description of the above embodiments, those skilled in the art can easily understand that the exemplary embodiments described herein can be implemented by software, or can be implemented by the way of software in combination with necessary hardware. Therefore, the technical solution according to the embodiment of the present disclosure can be embodied in the form of a software product, and the software product can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiment of the present disclosure.

[0228] After considering the specification and practicing the disclosure herein, those skilled in the art will easily think of other embodiments of the present disclosure. This application is intended to cover any variations, uses, or adaptations of the present disclosure, and these variations, uses, or adaptations follow the general principles of the present disclosure and include the common general knowledge or conventional technical means in the technical field not disclosed in the present disclosure. The specification and the embodiments are only regarded as exemplary, and the true scope and spirit of the present disclosure are pointed out by the claims.

[0229] It should be noted that although several units / modules or sub-units / modules of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / modules. Conversely, the features and functions of one unit / modules described above can be further divided and embodied by multiple units / modules.

[0230] In addition, although the operations of the method of the present disclosure are described in a specific order in the drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step for execution, and / or one step may be decomposed into multiple steps for execution.

[0231] Although the spirit and principles of the present disclosure have been described with reference to several specific embodiments, it should be understood that the present disclosure is not limited to the specific embodiments disclosed, and the division of each aspect does not mean that the features in these aspects cannot be combined for benefit. This division is only for the convenience of expression. The present disclosure aims to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A resource retrieval method, characterized in that, The method includes: Obtaining the input information of the user; When the input information includes text information, splitting the text information to obtain a plurality of split text information; Respectively retrieving in the resource library based on each split text information to obtain first candidate resources respectively corresponding to the respective split text information; Respectively calculating the comprehensive similarity results between each first candidate resource and the target text information, and determining a target resource that matches the input information from each first candidate resource according to the comprehensive similarity results, where the target text information includes the text information and the plurality of split text information.

2. The method according to claim 1, wherein The splitting the text information to obtain a plurality of split text information includes at least one of the following: When the format of the text information does not conform to a predefined format, converting the format of the text information to the predefined format and splitting the converted text information; Filtering out preset special texts from the text information and filtering the special texts to obtain the plurality of split text information, where the special texts include delimiter symbols and special words.

3. The method according to claim 1, characterized in that, The resource library includes an original resource library, an index resource library, and a resource matching relationship library, where the index resource library is constructed by performing an inverted index on the original resources stored in the original resource library, and the resource matching relationship library is used to record the matching relationships between the original resources; The retrieving in the resource library based on each split text information to obtain first candidate resources respectively corresponding to the respective split text information includes: Respectively retrieving in the original resource library based on each split text information to obtain the original resources respectively corresponding to the respective split text information as the first candidate resources; Respectively retrieving in the index resource library based on each split text information to obtain an associated document list respectively corresponding to the respective split text information as the first candidate resources; Respectively retrieving in the resource matching relationship library based on each split text information to obtain the associated resources of the resources corresponding to the respective split text information, and using the associated resources as the first candidate resources, where the associated resources match the resources corresponding to the respective split text information.

4. The method according to claim 3, characterized in that, In a song retrieval scenario, the resource matching relationship library is used to record the matching relationships between the audio fingerprint features of the song resources; The retrieving in the resource matching relationship library based on each split text information to obtain the associated resources of the resources corresponding to the respective split text information includes: when the plurality of split text information includes song resource IDs, retrieving in the resource matching relationship library based on the song resource IDs to obtain the associated resources of the resources corresponding to the song resource IDs.

5. The method according to claim 4, wherein The matching relationships between the audio fingerprint features of the song resources are obtained by comparing the audio fingerprints of the song resources, or by comparing the audio slices of the song resources; where the audio fingerprints of the song resources include at least one of cover fingerprints, timbre fingerprints, and spectral fingerprints.

6. The method according to claim 1, characterized in that, Calculating the comprehensive similarity results between each first candidate resource and the target text information respectively includes: Classifying the target text information according to the text dimension to which the target text information belongs, where the text dimension includes a resource name dimension, a resource content producer dimension, and a resource attribution set dimension; For each first candidate resource, calculating the similarity between the first candidate resource and the target text information included in each text dimension respectively, to obtain the similarity results corresponding to the first candidate resource in each text dimension respectively; Determining the comprehensive similarity result corresponding to the first candidate resource according to the dimension weights of each text dimension and the similarity results corresponding to the first candidate resource in each text dimension respectively.

7. A resource retrieval device, characterized in that, The apparatus includes: An acquisition unit, configured to acquire the input information of the user; A splitting unit, configured to split the text information to obtain a plurality of split text information when the input information includes text information; A retrieval unit, configured to retrieve in the resource library based on each split text information respectively, to obtain the first candidate resources corresponding to each split text information respectively; A determination unit, configured to calculate the comprehensive similarity results between each first candidate resource and the target text information respectively, and determine a target resource that matches the input information from each first candidate resource according to the comprehensive similarity results, where the target text information includes the text information and the plurality of split text information.

8. An electronic device, characterized in that, Comprising: A processor; A memory for storing executable instructions of the processor; Wherein, the processor realizes the method according to any one of claims 1-6 by running the executable instructions.

9. A computer-readable storage medium having computer instructions stored thereon, characterized in that, When the instruction is executed by the processor, it realizes the method according to any one of claims 1-6.

10. A computer program product, characterized in that, Including a computer program / instructions, when the computer program / instructions are executed by the processor, it realizes the steps of the method according to any one of claims 1-6.

Images