Model training feature generation method and music associational word display method

Abstract

The invention relates to a model training feature generation method and a music associational word display method, and relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring a plurality of user real-time interaction data records corresponding to a current interaction period; adding the music keyword corresponding to the current interaction period into a pre-generated real-time data stream to obtain an updated real-time data stream; music keywords with the implementation time meeting the current statistical time period are intercepted from the updated real-time data stream to serve as target keywords, interaction operations associated with the same keywords in the target keywords are subjected to merging statistics, and real-time operation statistical characteristics of the target keywords are obtained; generating model training features according to the real-time operation statistical features of the target keyword; the model training features are used for training a real-time popularity response model; the real-time popularity response model is used for outputting associative words of the music search words. By adopting the method, the display effect of the music associational words can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method for generating model training features, a method and apparatus for displaying music association words, a computer device, a computer-readable storage medium, and a computer program product. Background Art

[0002] With the rapid development of digital music and mobile music, users search for various music queries in the search interface of music-related software programs, so many song search tasks are generated every day. In order to provide users with the convenience of using the music search function and to make the music search function respond more quickly and in real time, new requirements need to be put forward for the music search function and association function to meet users' needs for the speed and efficiency of the search function.

[0003] However, the current music keyword search and association technology has large delays everywhere, whether in terms of data collection, feature update or model update, resulting in a large overall delay in the music keyword search and association process. Therefore, the current music keyword search and association technology cannot respond to user needs in a timely manner, which in turn makes the display effect of music association words poor. Summary of the invention

[0004] Based on this, it is necessary to provide a model training feature generation method and a display method, device, computer equipment, computer-readable storage medium and computer program product for improving the display effect of music association words in response to the above-mentioned technical problems.

[0005] In a first aspect, an embodiment of the present application provides a method for generating model training features. The method comprises:

[0006] Acquire multiple user real-time interaction data records corresponding to the current interaction cycle; each of the user real-time interaction data records includes the interaction operations performed by each user on each music object in the current interaction cycle, and the music keywords associated with each music object;

[0007] Adding the music keywords corresponding to the current interactive cycle to the pre-generated real-time data stream to obtain the updated real-time data stream; the real-time data stream is a data stream obtained by arranging the music keywords in the order of the implementation time of the corresponding interactive operations;

[0008] The music keyword whose implementation time meets the current statistical time period is intercepted from the updated real-time data stream as the target keyword, and the interactive operations associated with the same keyword in the target keyword are combined and counted to obtain the real-time operation statistical features of the target keyword;

[0009] Model training features are generated based on the real-time operation statistical features of the target keyword; the model training features are used to train a real-time heat response model; the real-time heat response model is used to output associative words of the music search term.

[0010] In one embodiment, the step of intercepting the music keyword whose execution time satisfies the current statistical time period from the updated real-time data stream as the target keyword comprises:

[0011] A sliding time window whose time length satisfies the current statistical time period is set for the real-time data stream; the sliding time window is used to represent a time window from the start time of the current statistical time period to the end time of the current statistical time period;

[0012] According to the start time and the end time represented by the sliding time window, the music keyword whose implementation time is between the start time and the end time is intercepted from the real-time data stream as the target keyword.

[0013] In one embodiment, generating model training features according to real-time operation statistical features of the target keyword includes:

[0014] Adding the real-time operation statistical features of the target keyword to the original feature set associated with the real-time heat response model to obtain a supplemented feature set; the original feature set is the supplemented feature set obtained in the previous time period of the current statistical time period;

[0015] The score values ​​of the interactive operations associated with the same target keyword in the supplemented feature set are combined and calculated to obtain a real-time supplement feature of the target keyword;

[0016] The model training feature is generated according to the real-time complement feature of the target keyword and the real-time operation statistical feature of the target keyword.

[0017] In one embodiment, generating the model training feature according to the real-time complement feature of the target keyword and the real-time operation statistical feature of the target keyword includes:

[0018] Input the real-time operation statistical features of the target keyword into the trained heat model, so that the heat model outputs the keyword heat feature corresponding to the target keyword; the keyword heat feature at least includes a first heat feature and a second heat feature; the first heat feature represents the heat value of each of the target keywords at the implementation time, and the second heat feature represents the heat value growth rate of each of the music keywords;

[0019] The keyword heat feature of the target keyword and the real-time complement feature of the target keyword are combined to generate the model training feature.

[0020] In one embodiment, the step of combining the keyword heat feature of the target keyword and the real-time complement feature of the target keyword to generate the model training feature includes:

[0021] Acquire the correlation feature associated with the target keyword and the target scene; wherein the target scene is a scene in which the user performs the user interactive operation on the music object corresponding to the target keyword;

[0022] The keyword heat feature of the target keyword, the real-time complement feature of the target keyword, and the relevance feature of the target keyword are combined to generate the model training feature.

[0023] In a second aspect, the present application provides a method for displaying music-associated words. The method comprises:

[0024] In response to a music search term input by a user, recalling a plurality of associated words corresponding to the music search term;

[0025] The plurality of associated words are sorted by a pre-trained real-time heat response model to obtain a sorting result of the plurality of associated words; the pre-trained real-time heat response model is obtained by training using the model training feature described in any one of the first aspects above;

[0026] According to the sorting results of the multiple associative words, the multiple associative words corresponding to the music search word are displayed.

[0027] In one embodiment, the recalling of multiple associated words corresponding to the music search word includes:

[0028] Through a pre-trained recall model, multiple associative words corresponding to the music search word are recalled from an associative word database; wherein the associative word database records an index relationship between the search word and the associative word, and the index relationship is created when the number of times the search word is searched is greater than a preset threshold.

[0029] In a third aspect, the present application also provides a model training feature generation device. The device comprises:

[0030] A user real-time interaction data record acquisition module is used to acquire multiple user real-time interaction data records corresponding to the current interaction cycle; each of the user real-time interaction data records includes the interaction operations performed by each user on each music object in the current interaction cycle, and the music keywords associated with each music object;

[0031] A real-time data stream generation module, used for adding the music keywords corresponding to the current interactive cycle to the pre-generated real-time data stream to obtain the updated real-time data stream; the real-time data stream is a data stream obtained by arranging the music keywords in the order of the implementation time of the corresponding interactive operations;

[0032] A real-time operation statistical feature generation module is used to intercept the music keyword whose implementation time meets the current statistical time period from the updated real-time data stream as the target keyword, and to combine and count the interactive operations associated with the same keyword in the target keyword to obtain the real-time operation statistical feature of the target keyword;

[0033] The model training feature generation module is used to generate model training features according to the real-time operation statistical features of the target keyword; the model training features are used to train the real-time heat response model; the real-time heat response model is used to output the associative words of the music search term.

[0034] In a fourth aspect, the present application also provides a device for displaying music association words. The device comprises:

[0035] An associative word recall module, used to respond to a music search word input by a user and recall a plurality of associative words corresponding to the music search word;

[0036] A sorting result generating module, used for sorting the multiple associative words by using a pre-trained real-time heat response model to obtain a sorting result of the multiple associative words; the pre-trained real-time heat response model is trained by using the model training features described in any one of the first aspects above;

[0037] The associative word display module is used to display the multiple associative words corresponding to the music search word according to the sorting results of the multiple associative words.

[0038] In a fifth aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements: the steps of the method described in any one of the first aspects above; or the steps of the method described in any one of the second aspects above.

[0039] In a sixth aspect, the present application further provides a computer-readable storage medium, wherein a computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method described in any one of the first aspects above or the steps of the method described in any one of the second aspects above are implemented.

[0040] In a seventh aspect, the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements: the steps of the method described in any one of the first aspects above; or the steps of the method described in any one of the second aspects above.

[0041] The above-mentioned model training feature generation method, device, computer equipment, storage medium and computer program product first obtain the user's real-time interaction data record in the current interaction cycle, then add the music keywords of the current cycle to the pre-generated real-time data stream, and then intercept the music keywords that match the current statistical time period from the updated real-time data stream as the target keyword, and merge the interactive operations of the same keyword in the target keyword to obtain the real-time operation statistical features of the target keyword to constitute the model training features and use them for the association of search terms; through the generation of real-time data streams and batch association statistics, the operation records generated in real time in the music platform are successfully converted into feature data that can be used by the model, reducing data processing delays in multiple aspects such as data collection, feature update and model update. Under the background of huge music search volume, it still provides users with efficient music entry search and association services, improves the timeliness of responding to user needs, and improves the display effect of music association words. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A schematic diagram of a flow chart of a method for generating model training features in one embodiment;

[0044] Figure 2 A schematic diagram of a flow chart of a method for generating model training features in another embodiment;

[0045] Figure 3 A schematic diagram of a flow chart of a method for displaying music association words in one embodiment;

[0046] Figure 4 A schematic diagram of the structure of a clustering model for searching associative words in the music field in one embodiment;

[0047] Figure 5 is a schematic diagram of the principle of real-time data flow in an embodiment;

[0048] Figure 6 is a schematic diagram of the principle of a sliding window in an embodiment;

[0049] Figure 7 A schematic diagram of the effect of a music search interface in one embodiment;

[0050] Figure 8 It is a structural block diagram of a model training feature generation device in one embodiment;

[0051] Fig. 9 is a structural block diagram of a display device for music association words in one embodiment;

[0052] Fig.10 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0054] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0055] In one embodiment, a method for generating model training features is provided. This embodiment uses the method applied to a terminal as an example. It is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0056] S101, obtaining multiple user real-time interaction data records corresponding to the current interaction cycle.

[0057] The current interaction cycle refers to the time period for collecting real-time user interaction data records. For example, a current interaction cycle of 1 minute means that real-time user interaction data records are collected every 1 minute.

[0058] The real-time interaction data record of each user includes the interaction operations performed by each user on each music object in the current interaction cycle, as well as the music keywords associated with each music object. For example, within 1 minute, user U1 collected music object D1 (the associated music keyword is Word1) once, and user U2 browsed the page where music object D2 (the associated music keyword is Word2) is located twice. Then the real-time interaction data record of users corresponding to this 1 minute can be:

[0059] Data record 1: The number of times Word1 is collected is 1;

[0060] Data record 2: Word2 has been viewed 2 times.

[0061] Among them, the music object can be a song, a singer, an album, the display page of a music work, etc. Naturally, the music keywords associated with the music object can be the text of a song name, the text of a singer's name, or the text of an album name, etc. The interactive operations performed on each music object can be click, favorite, evaluate, forward, browse, play, download, comment, and other operations.

[0062] S102, adding the music keywords corresponding to the current interactive cycle to the pre-generated real-time data stream to obtain an updated real-time data stream.

[0063] Among them, the real-time data stream is also called real-time flow. The real-time data stream is a data stream obtained by arranging each music keyword in the order of the implementation time of the corresponding interactive operation. Each music keyword is a keyword associated with the music object.

[0064] It should be noted that real-time stream refers to a data set that is continuously generated and updated, usually flowing in chronological order. Real-time stream is used in application scenarios that require quick response and instant feedback. In the field of music keyword search, real-time stream can be a data stream in which multiple keywords are arranged in the chronological order of their appearance. For example, a real-time data stream can be: "Word1, Word2, Word3, Word4, Word5...WordN"; further, music keywords in real-time data streams are allowed to be repeated, that is, when a user performs an operation on the same music object at different times, a real-time data stream can be: "Word1, Word2, Word1, Word4, Word5...WordN" where Word1 is the music keyword associated with the same music object mentioned above.

[0065] Furthermore, the real-time data stream is updated according to the interaction cycle, and each new interaction cycle obtains music keywords and adds them to the real-time data stream to update the real-time data stream. The current interaction cycle in step S102 is the current interaction cycle, and the pre-generated real-time data stream refers to the real-time data stream updated in the previous interaction cycle of the current interaction cycle.

[0066] Exemplarily, the updating process of the real-time data stream may be:

[0067] The real-time data stream obtained after updating in the first minute is "Word1, Word2...WordN". If the music keywords within the current minute include WordN+1, then WordN+1 is added to "Word1, Word2...WordN", and the updated real-time data stream corresponding to the current 1 minute interaction cycle is "Word1, Word2...WordN, WordN+1".

[0068] S103, intercepting music keywords whose implementation time meets the current statistical time period from the updated real-time data stream as target keywords, and merging and counting the interactive operations associated with the same keywords in the target keywords to obtain real-time operation statistical features of the target keywords.

[0069] The current statistical time period refers to the time period for implementing the interception and combined statistics of the target keywords in step S103, and the current statistical time period may be 10 minutes, 20 minutes or 30 minutes.

[0070] The target keyword is at least one music keyword recorded in the real-time data stream within the current statistical time period.

[0071] Among them, the process of combined statistics can be to calculate the index value of the interactive operation in which the music object corresponding to each music keyword participates, that is, the index value corresponding to operations such as click, favorite, evaluate, forward, browse, play, download, comment, etc. For example, the indicator can be the click-through rate (CTR) corresponding to the click operation, or the page views (PV) corresponding to the browse operation, etc.; the music object targeted by the click-through rate can be the entry of the song or singer, and the music object targeted by the page views can be the web page containing the song or singer information.

[0072] The real-time operation statistical features are data features obtained after combined statistics, and are data features obtained by combining statistics for operations corresponding to keywords in the real-time data stream generated in real time.

[0073] S104: Generate model training features based on real-time operation statistical features of the target keyword.

[0074] The model training feature is used to train the real-time heat response model, and the real-time heat response model is used to output the associated words of the music search terms. Specifically, the model training feature is a vector composed of feature values ​​with multiple dimensions. For example, the model training feature can be (a, b, c), where a can be the embedding value of the music keyword, b can be the normalized value of the implementation time, and c can be the indicator value of a certain interactive operation implemented;

[0075] More specifically, for the same music keyword, there are generally multiple types of interactive operations, and naturally, there are also multiple corresponding indicator values. For example, the model training features can be (a, b, c1, c2, c3), where c1 is the indicator value of the collection operation, c2 is the indicator value of the play operation, and c3 is the indicator value of the browse operation.

[0076] Among them, the real-time heat response model is also called a sorting model or a response model. There are many types of models available, such as models based on gradient boosting tree (GBT), gradient boosting decision tree (GBDT), etc. The embodiment of the present application does not limit the specific model type of the real-time heat response model. It only needs to learn the real-time features to realize the sorting function.

[0077] The music search term may be a search term entered by a user in a search page when using a music-related program software, and the associated term is a term associated with the program software based on the search term entered by the user.

[0078] In the above-mentioned model training feature generation method, the real-time user interaction data records within the current interaction cycle are first obtained, and then the music keywords of the current cycle are added to the pre-generated real-time data stream, and then the music keywords that meet the current statistical time period are intercepted from the updated real-time data stream as the target keywords, and the interactive operations of the same keywords in the target keywords are merged and counted to obtain the real-time operation statistical features of the target keywords to constitute the model training features and be used for the association of search terms; through the generation of real-time data streams and batch association statistics, the operation records generated in real time in the music platform are successfully converted into feature data that can be used by the model, reducing data processing delays in multiple aspects such as data collection, feature update and model update. Under the background of huge music search volume, it still provides users with efficient music entry search and association services, improves the timeliness of responding to user needs, and improves the display effect of music association words.

[0079] In one embodiment, music keywords whose implementation time satisfies the current statistical time period are intercepted from the updated real-time data stream as target keywords, including: setting a sliding time window with a time length that satisfies the current statistical time period for the real-time data stream; and intercepting music keywords whose implementation time is between the start time and the end time from the real-time data stream as target keywords according to the start time and the end time represented by the sliding time window.

[0080] Among them, the sliding time window is used to represent the time window from the start time of the current statistical time period to the end time of the current statistical time period. The sliding time window is also called a sliding window, which specifies a window of fixed time length. As time goes by, the window slides on the real-time data stream to process and analyze the data in each window. The fixed time length is reflected by the length of the current statistical time period. For example, if the current statistical time period is 10 minutes, then the sliding time window can represent the time window from the start time T1 to the end time T2 (T2=T1+10 minutes).

[0081] For example, a real-time data stream is "Word1, Word2, Word3, Word4, Word5, Word4, Word6". If the time difference from Word3 to Word6 is 10 minutes, and at a certain moment, the start time of the sliding time window is exactly the same as the implementation time of Word3, naturally, the end time of the sliding time window is exactly the same as the implementation time of Word6. At this time, the sliding time window will intercept "Word4, Word5, Word4, Word6" from the real-time data stream "Word1, Word2, Word3, Word4, Word5, Word4, Word6" as the target keywords.

[0082] In this embodiment, a sliding time window is first set for the real-time data stream, and its time length meets the requirements of the current statistical time period. Then, according to the start time and end time represented by the sliding time window, music keywords with implementation time between the start time and the end time are intercepted from the real-time data stream as target keywords, so as to facilitate the combined statistics of interactive operations using the target keywords. Through the application of sliding window technology, this embodiment improves the ability to capture real-time user interactive data, accurately extracts target keywords to generate statistical features reflecting real-time user operations, and is used for model training and application, thereby improving the accuracy and timeliness of target keyword interception.

[0083] In one embodiment, model training features are generated based on real-time operation statistical features of target keywords, including: adding the real-time operation statistical features of the target keywords to the original feature set associated with the real-time heat response model to obtain a supplemented feature set; merging and calculating the score values ​​of interactive operations associated with the same target keywords in the supplemented feature set to obtain real-time supplementary features of the target keywords; and generating model training features based on the real-time supplementary features of the target keywords and the real-time operation statistical features of the target keywords.

[0084] The original feature set is a supplemented feature set obtained in the previous time period of the current statistical time period.

[0085] Among them, the meaning of complementation is to add new features (real-time operation statistical features) to the original feature set, and the merge calculation refers to recalculating the score value of the interactive operation of each keyword after adding the new features (real-time operation statistical features), so as to obtain the real-time complementary features of the target keyword.

[0086] For example, the original feature set includes 100 music keywords and the score values ​​of various interactive operations corresponding to the keywords. One of the music keywords is "singer A", and its corresponding click rate is 10% (the entry was placed 100 times and clicked 10 times);

[0087] The real-time operation statistical features in the current statistical time period include 10 music keywords and the score values ​​of various interactive operations corresponding to the keywords, which also involve the music keyword "singer A" and other music keywords are not repeated with the keywords in the original feature set. The click rate of the music keyword "singer A" in the real-time operation statistical features is 20% (the entry was released 10 times and clicked 2 times). At this time, after the supplementation, the score value (click rate) of the click operation of the music keyword "singer A" needs to be recalculated.

[0088] The process of combined calculation includes: recalculating the total number of deliveries to be: 100+10=110, the total number of clicks to be 10+2=12, then calculating that in the completed feature set, the score value (click rate) of the click operation corresponding to the music keyword "singer A" is 12 / 110 (about 10.9%), then in the combined feature set, the score value (click rate) of the click operation corresponding to the music keyword "singer A" is 10.9%.

[0089] In this embodiment, first, according to the real-time operation statistical features of the target keyword, it is added to the original feature set associated with the real-time heat response model to obtain a supplemented feature set, and then the score values ​​of the interactive operations associated with the same target keyword in the supplemented feature set are merged and calculated to form a combined feature set, and finally, according to the real-time supplementary features of the target keyword in the combined feature set and the real-time operation statistical features of the target keyword, the model training features are generated; this embodiment realizes real-time addition of features by dynamic feature supplementation, and recalculates the supplemented features, thereby ensuring the availability of features, improving the accuracy and effectiveness of model application, and thus improving the association effect of music search terms.

[0090] In one embodiment, the model training features are generated according to the real-time complement features of the target keyword and the real-time operation statistical features of the target keyword in the merged feature set, including:

[0091] Input the real-time operation statistical features of the target keyword into the trained heat model, so that the heat model outputs the keyword heat features corresponding to the target keyword;

[0092] The keyword popularity feature of the target keyword and the real-time complement feature of the target keyword are combined to generate model training features.

[0093] Among them, the heat model is a model used to analyze data change trends in real time. For example, the long short-term memory network is widely used in time series prediction.

[0094] Among them, the keyword heat feature includes at least the first heat feature and the second heat feature. The first heat feature refers to the real-time heat feature of the entry, which characterizes the heat value of each target keyword at the current time. The second heat feature is also called the heat value surge feature, which characterizes the growth rate of the heat value of each music keyword.

[0095] In this embodiment, the real-time operation statistical features of the target keyword are input into the trained popularity model to generate keyword popularity features including real-time popularity features and popularity value surge features. By merging these features, a more comprehensive model training feature is generated. This improves the model's sensitivity to keyword popularity changes and prediction accuracy, helps the system to better recommend music association words, and can reflect network hotspots more quickly and in real time, thereby improving the effect of association word display.

[0096] In one embodiment, the keyword heat feature of the target keyword and the real-time complement feature of the target keyword are combined to generate a model training feature, including:

[0097] Obtain the relevance characteristics of the target keywords and the target scenarios;

[0098] The keyword popularity feature of the target keyword, the real-time complement feature of the target keyword, and the relevance feature of the target keyword are combined to generate model training features.

[0099] Among them, the target scene is a scene in which the user performs user interactive operations on the music object corresponding to the target keyword; and the relevance feature is a feature associated with the target scene.

[0100] Exemplarily, the relevance features include: device features of the target device on which the user performs the interactive operation, features of the geographical location of the target device, or features of the search history of music keywords. Alternatively, the relevance features may include features related to the search term, such as the geographical location, time information, device type, and features of the search history related to the search term when the user enters the search term. For example, when user U1 searches for "popular music" at T1 in the evening, it may be to find music for evening parties or parties. In this case, the time feature "T1 in the evening" can be used to provide more relevant results in combination with the context.

[0101] In this embodiment, by obtaining the relevance features of each target keyword and combining them with the keyword popularity features and the real-time complement features, a comprehensive model training feature is generated to facilitate the subsequent training of the model with comprehensive feature data; the relevance features cover multiple dimensions related to the user's interaction in a specific scenario. By integrating this information, the model can more comprehensively understand user needs and preferences, and improve the sorting effect of the real-time feature response model.

[0102] In another embodiment, Figure 2 As shown, a model training feature generation method is provided, comprising the following steps:

[0103] S201, obtaining multiple user real-time interaction data records corresponding to the current interaction cycle;

[0104] S202, adding the music keywords corresponding to the current interactive cycle to the pre-generated real-time data stream to obtain an updated real-time data stream;

[0105] S203, setting a sliding time window whose time length satisfies the current statistical time period for the real-time data stream;

[0106] S204, according to the start time and the end time represented by the sliding time window, intercepting the music keywords with the implementation time between the start time and the end time from the real-time data stream as the target keywords;

[0107] S205, combining and counting interactive operations associated with the same keyword in the target keyword to obtain real-time operation statistical features of the target keyword;

[0108] S206, adding the real-time operation statistical features of the target keyword to the original feature set associated with the real-time heat response model to obtain a supplemented feature set;

[0109] S207, combining and calculating the score values ​​of the interactive operations associated with the same target keyword in the supplemented feature set to obtain a real-time supplement feature of the target keyword;

[0110] S208, generating model training features according to the real-time complement features of the target keyword and the real-time operation statistical features of the target keyword.

[0111] It should be noted that the specific limitations of the above steps can be found in the specific limitations of a model training feature generation method above, which will not be repeated here.

[0112] In one embodiment, a method for displaying music association words is provided. This embodiment is illustrated by applying the method to a terminal. It is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0113] S301, in response to a music search word input by a user, recalling a plurality of associated words corresponding to the music search word.

[0114] The music search term may be a search term entered by a user in a search page when using a music-related program software, and the associated term is a term associated with the program software based on the search term entered by the user.

[0115] S302, sorting the multiple associative words by using the pre-trained real-time heat response model to obtain sorting results of the multiple associative words.

[0116] Among them, the pre-trained real-time heat response model is obtained by training using the model training features in any of the above embodiments.

[0117] S303, displaying multiple associative words corresponding to the music search term according to the sorting results of the multiple associative words.

[0118] For example, in a user's query, the search term entered by user U1 is Word1. The recall model is first used to obtain the recalled candidate terms: Word11, Word12, Word13; then the trained real-time feature response model is used to calculate the sorting priority of "candidate terms: Word11, Word12, Word13". Finally, the "candidate terms: Word11, Word12, Word13" are sorted according to the priority to obtain the sorting result.

[0119] In the above-mentioned method for displaying music association words, the music search words input by the user are obtained to recall the association words, and the association words are sorted with the help of the pre-trained real-time heat response model. Finally, the related association words are displayed according to the sorting results, so that music recommendations related to the user's search intention can be provided accurately and quickly, improving the user's search experience and helping users to find music content of interest more efficiently. At the same time, the accuracy and relevance of the recommendations are improved through real-time heat response, which enhances user satisfaction and the attractiveness of the platform and improves the display effect of the association words.

[0120] In one embodiment, recalling multiple associative words corresponding to a music search term includes: recalling multiple associative words corresponding to the music search term from an associative word database through a pre-trained recall model; wherein the associative word database records an index relationship between the search term and the associative word, and the index relationship is created when the search term is searched for more than a preset threshold.

[0121] Exemplarily, if the number of active searches for a certain search term by a user reaches a certain threshold, the search term will be indexed into the database and put into the recall model for recall. Active search refers to the search behavior that the user actively types in. Specifically, if the number of active searches for the term "Singer A Concert" exceeds 10 times, "Singer A" will be used as the index of the term "Singer A Concert" and stored in the database corresponding to the recall model. This is convenient for when the user searches with "Singer A" as input, the term "Singer A Concert" can be recalled.

[0122] In this embodiment, the database of the recall model is indexed and stored by statistics of real-time data streams to dynamically update the database to improve the recall capability. The observed popular search terms or query terms are saved in a fixed structure of the database for subsequent retrieval, recommendation and sorting, so as to improve the recall accuracy and efficiency of the subsequent recall model, thereby improving the generation and display effects of associative words.

[0123] In the prior art, with the rapid development of digital music and mobile music, there are various queries for daily search, and many hot spots and popular songs are born every day. In order to provide convenience to users and reflect network hot spots more real-time and quickly, the search of intelligent associative words needs to provide convenience to users and meet user needs more quickly and effectively. Among them, associative words refer to the words that users want to search for based on the partial phrases entered by the user in the search box.

[0124] Therefore, the existing technology has delays in data latency, data collection, feature update, and model update. Therefore, the existing frameworks and methods cannot reflect hot spots in real time and cannot respond to user needs in a timely manner.

[0125] On the other hand, there are many heat response models in the industry, but most of them are for recommendation, which only need to calculate heat information and update and push the hottest data in real time. However, search needs to balance the characteristics of the original data recalled and the characteristics of the new data, and guide the sorting based on relevance.

[0126] It can be seen that the search method of the recommendation field in the existing solution is not well suited to the scenario of the keyword search field of this application, and the hottest ones cannot be pushed to users at will. The search for the Different Opinion Recommendation Scenario (a scenario for classifying and managing users' diverse opinions and preferences) gives users a display list, and recommends documents to users one by one. First of all, the terms here need to ensure high relevance, calculate the term popularity, and unify the original data and the supplemented data into the same data dimension.

[0127] Based on this, the embodiment of the present application provides a model training feature generation method and a music association word display method. In order to facilitate the understanding of those skilled in the art, Figure 4 A structural diagram of a music domain search associative word clustering model is provided. Figure 5 A schematic diagram of the principle of real-time data flow is provided. Figure 6 A schematic diagram of the principle of a sliding window is provided. Figure 7 A schematic diagram of the music search interface is provided. Figures 4 to 7 , a model training feature generation method and a method for displaying music association words are described in detail with a specific embodiment. It is worth noting that the following description is only an exemplary description and not a specific limitation of the application.

[0128] The model training feature generation method and the music association word display method provided in the present application are also called a real-time response method for searching association words in the music field, wherein real-time means responding faster and providing real-time feedback on user behavior. The model training feature generation method and the music association word display method provided in the present application can obtain user behavior and feedback data in real time, add a heat model to respond to changes in the heat of entries, integrate and supplement the original features to update the real-time model, and can consume data in real time through the user's post-search behavior and results, construct recall and sorting methods by calculating the sliding window statistics of the real-time stream, calculate the heat model and complement it with previously outdated features, and update the online data and model in real time through this process.

[0129] like Figure 4As shown, in the structural diagram of the music field search associative word clustering model, the model training feature generation method and the music associative word display method provided by the present application convert the user's behavior operation log data into a real-time stream, consume the real-time stream data (referring to the process in which the data is processed and used immediately after it is generated), and count the data features with a sliding window of a certain size. On the one hand, the heat of the terms actively input by the user is calculated to obtain the number of active searches. When a certain threshold is reached, it is indexed into the database so that it can be recalled through the recall model; on the other hand, the data statistical features are placed in the heat model to calculate the heat of the features; on the third hand, the data of the sliding window is supplemented with the data of the previous model to obtain the supplemented feature data. Finally, the features obtained from the second and third aspects are placed in the heat real-time response model, and the terms recalled by the recall model are input into the real-time heat response model, and the sorting priority of each recalled term is calculated to obtain the final sorting result.

[0130] The calculation process of the model training feature generation method provided in this application may include steps 1 to 6, specifically including:

[0131] Step 1: Collect user behavior data and convert the data into real-time stream. The algorithm expression can be:

[0132] ;

[0133] in, For real-time streaming, For user behavior data, It is the conversion process of real-time stream. Generally, real-time data storage and consumption use flink, oceanous, or other real-time stream methods. Specifically, the real-time stream can be "Word1, Word2, Word3, Word4, Word5...WordN". A real-time data stream such as Figure 5 shown.

[0134] Step 2: Associate and count the real-time stream data with a sliding window of a certain time to obtain data such as the number of clicks and pv (pageview). The algorithm expression can be:

[0135] ;

[0136] in, is the data obtained after sliding window statistics. For statistical process.

[0137] Combined with the real-time data stream in step 1, such as Figure 6As shown in , the sliding window in step 2 specifies a fixed-size window. As time goes by, the window slides on the data stream and processes and analyzes the data in each window. Figure 6 As shown in the figure, there are two keywords, Word3 and Word4, in the sliding window, indicating that the user operations (clicks, PV) of the two keywords "Word3, Word4" are correlated and counted at the current moment to obtain the data characteristics after statistics. For example, the time window is set to the past 10, 20 or 30 minutes, and the clicks and related page views of a song search word in this time period are calculated. The sliding window is continuously updated to obtain real-time statistical data.

[0138] Step 3: If the number of active searches reaches a certain threshold , then the index is stored in the database and put into the recall model for recall. Among them, active search refers to the search behavior that users actively type in. Its algorithm expression can be:

[0139] ;

[0140] in, Active search refers to the search behavior that users actively type in. Term indexing is the process of saving the observed popular search terms or query terms into a fixed structure of the database for subsequent retrieval, recommendation and sorting, so as to improve the recall accuracy and efficiency of the subsequent recall model.

[0141] In addition, the threshold of discrimination can be changed and can be flexibly selected. The recall index can be judged in other ways whether it needs to be indexed and stored. For example, a machine learning algorithm can be used to predict the data to determine whether a certain data is worth indexing. Or, according to the category of the music library, it can be judged which question library should be stored.

[0142] Step 4: Based on the data obtained in step 2, the heat model is used to obtain the real-time heat of the term and the features such as soaring query_hot_fea. The algorithm expression can be:

[0143] ;

[0144] Among them, query_hot_fea refers to the real-time popularity and soaring characteristics of the term. Refers to the calculation process of the heat model. Real-time heat reflects the popularity of the search term within a specific time, such as the number of searches and click-through rates of the term in the past hour. For example, the search volume for "new song release" was 100 times on a certain day. The surge feature refers to the rapid increase in the popularity of the search term within a short period of time, which is usually related to specific events, such as news hotspots, cultural events, etc. For example, on the day when the new song was released, the search volume suddenly reached 2,000 times in a short period of time, indicating a surge in demand.

[0145] Step 5: Based on the data obtained in step 2, the original data features are supplemented to obtain real-time supplementary features. The algorithm expression can be:

[0146] ;

[0147] Among them, query_all_time_fea is the real-time supplement feature. To complement the process.

[0148] The original data features refer to the features in the data set that have been collected and stored before, such as previous search records, user behavior analysis, etc. These data may be in the historical data repository and have been used to optimize the algorithm or model training. Real-time supplementation is to improve the accuracy and real-time performance of recommendations so that the model can reflect the latest needs and behavior trends of users. The supplementation process can be: integrating, updating and calibrating new real-time streaming data with historical data to unify the dimensions and form a more comprehensive feature set.

[0149] Step 6: Based on the features obtained in steps 4 and 5, as well as the correlation features, a real-time feature response model is trained. The algorithm expression can be:

[0150] ;

[0151] in, For the real-time popularity and soaring characteristics of the entry, To complement features in real time, is the correlation feature, Represents the training process of the model.

[0152] On the basis of steps 1 to 6, the calculation process of the method for displaying music association words provided by the present application may include:

[0153] Step 7: Update the model and features to obtain the final sorting results.

[0154] The model training feature generation method and music association word display method provided in this application can respond to hot information in real time, recall new entries, reflect network popularity, provide hot and soaring entries to users, and provide timely feedback on sorting results.

[0155] The model training feature generation method and music association word display method provided in the present application can balance the recalled original data features and the features of new data in the search scenario. For example, the number of newly added data features each time can be dynamically adjusted, and the training effect of the model can be evaluated to determine the optimal ratio of new data features to original features, thereby improving the application effect of the model.

[0156] For example, during a user query, the user enters the search term "Singer A"; based on the search term, the recall model finds all content that may be related to the user query from the database, and recalls "Singer A's Album B" and "Singer A's Recent Concert Songs". Among them, "Singer A's Album B" can be the entry corresponding to the original data feature, and "Singer A's Recent Concert Songs" can be the entry corresponding to the new data feature. By supplementing with feature data, the features of the original data and the features of the new data can be balanced.

[0157] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0158] Based on the same inventive concept, the embodiment of the present application also provides a model training feature generation device for implementing the model training feature generation method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more model training feature generation device embodiments provided below can refer to the limitations of the model training feature generation method above, and will not be repeated here.

[0159] In one embodiment, Figure 8As shown, a model training feature generation device is provided, including: a user real-time interactive data record acquisition module 801, a real-time data stream generation module 802, a real-time operation statistical feature generation module 803, and a model training feature generation module 804, wherein:

[0160] The user real-time interaction data record acquisition module 801 is used to acquire multiple user real-time interaction data records corresponding to the current interaction cycle; each user real-time interaction data record includes the interaction operations performed by each user on each music object in the current interaction cycle, and the music keywords associated with each music object;

[0161] The real-time data stream generating module 802 is used to add the music keywords corresponding to the current interactive cycle to the pre-generated real-time data stream to obtain an updated real-time data stream; the real-time data stream is a data stream obtained by arranging the music keywords in the order of the implementation time of the corresponding interactive operation;

[0162] The real-time operation statistical feature generation module 803 is used to intercept the music keywords whose implementation time meets the current statistical time period from the updated real-time data stream as target keywords, and merge and count the interactive operations associated with the same keywords in the target keywords to obtain the real-time operation statistical features of the target keywords;

[0163] The model training feature generation module 804 is used to generate model training features based on the real-time operation statistical features of the target keyword; the model training features are used to train the real-time heat response model; the real-time heat response model is used to output the associated words of the music search term.

[0164] In one embodiment, the model training feature generation device is used to: set a sliding time window for the real-time data stream whose time length satisfies the current statistical time period; the sliding time window is used to represent the time window from the start time of the current statistical time period to the end time of the current statistical time period; according to the start time and end time represented by the sliding time window, music keywords with an implementation time between the start time and the end time are intercepted from the real-time data stream as target keywords.

[0165] In one embodiment, the model training feature generating device is used to: add the real-time operation statistical features of the target keyword to the original feature set associated with the real-time heat response model to obtain a supplemented feature set; the original feature set is the supplemented feature set obtained in the previous time period of the current statistical time period; the score values ​​of the interactive operations associated with the same target keyword in the supplemented feature set are merged and calculated to obtain the real-time supplementary features of the target keyword; and the model training features are generated based on the real-time supplementary features of the target keyword and the real-time operation statistical features of the target keyword.

[0166] In one embodiment, the model training feature generating device is used to: input the real-time operation statistical features of the target keyword into the trained heat model so that the heat model outputs the keyword heat features corresponding to the target keyword; the keyword heat features include at least a first heat feature and a second heat feature; the first heat feature represents the heat value of each target keyword at the implementation time, and the second heat feature represents the heat value growth rate of each music keyword; the keyword heat feature of the target keyword and the real-time supplement feature of the target keyword are combined to generate the model training feature.

[0167] In one embodiment, a model training feature generating device is used to: obtain a relevance feature associated with a target keyword and a target scene; wherein the target scene is a scene in which a user performs a user interactive operation on a music object corresponding to the target keyword; and merge the keyword heat feature of the target keyword, the real-time complement feature of the target keyword, and the relevance feature of the target keyword to generate a model training feature.

[0168] Based on the same inventive concept, the embodiment of the present application also provides a display device for music association words for implementing the display method for music association words involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of one or more display devices for music association words provided below can refer to the limitations of the display method for music association words above, and will not be repeated here.

[0169] In one embodiment, Fig. 9 As shown, a display device for music association words is provided, including: an association word recall module 901, a sorting result generation module 902, and an association word display module 903, wherein:

[0170] The associative word recall module 901 is used to respond to the music search word input by the user and recall multiple associative words corresponding to the music search word;

[0171] The sorting result generating module 902 is used to sort the multiple associative words by using the pre-trained real-time heat response model to obtain the sorting results of the multiple associative words; the pre-trained real-time heat response model is obtained by training using the model training features in any of the above embodiments;

[0172] The associative word display module 903 is used to display multiple associative words corresponding to the music search word according to the sorting results of the multiple associative words.

[0173] In one embodiment, a display device for music association words is used to: recall association words corresponding to music search words from an association word database through a pre-trained recall model; wherein the association word database records an index relationship between the search word and the association word, and the index relationship is created when the search word is searched for more than a preset threshold.

[0174] Each module in the above-mentioned model training feature generation device and music association word display device can be fully or partially implemented by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0175] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a model training feature generation method and a method for displaying music association words are realized. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0176] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0177] In one embodiment, a computer device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps in the above method embodiments are implemented.

[0178] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0179] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0180] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0181] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0182] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for generating model training features, characterized in that: The method comprises: Acquire multiple user real-time interaction data records corresponding to the current interaction cycle; each of the user real-time interaction data records includes the interaction operations performed by each user on each music object in the current interaction cycle, and the music keywords associated with each music object; Adding the music keywords corresponding to the current interactive cycle to the pre-generated real-time data stream to obtain the updated real-time data stream; the real-time data stream is a data stream obtained by arranging the music keywords in the order of the implementation time of the corresponding interactive operations; The music keyword whose implementation time meets the current statistical time period is intercepted from the updated real-time data stream as the target keyword, and the interactive operations associated with the same keyword in the target keyword are combined and counted to obtain the real-time operation statistical features of the target keyword; Model training features are generated based on the real-time operation statistical features of the target keyword; the model training features are used to train a real-time heat response model; the real-time heat response model is used to output associative words of the music search term.

2. The method according to claim 1, characterized in that The step of intercepting the music keyword whose execution time satisfies the current statistical time period from the updated real-time data stream as the target keyword comprises: A sliding time window whose time length satisfies the current statistical time period is set for the real-time data stream; the sliding time window is used to represent a time window from the start time of the current statistical time period to the end time of the current statistical time period; According to the start time and the end time represented by the sliding time window, the music keyword whose implementation time is between the start time and the end time is intercepted from the real-time data stream as the target keyword.

3. The method according to claim 1, characterized in that The generating of model training features according to the real-time operation statistical features of the target keyword includes: Adding the real-time operation statistical features of the target keyword to the original feature set associated with the real-time heat response model to obtain a supplemented feature set; the original feature set is the supplemented feature set obtained in the previous time period of the current statistical time period; The score values ​​of the interactive operations associated with the same target keyword in the supplemented feature set are combined and calculated to obtain a real-time supplement feature of the target keyword; The model training feature is generated according to the real-time complement feature of the target keyword and the real-time operation statistical feature of the target keyword.

4. The method according to claim 3, characterized in that The generating the model training feature according to the real-time complement feature of the target keyword and the real-time operation statistical feature of the target keyword includes: Input the real-time operation statistical features of the target keyword into the trained heat model, so that the heat model outputs the keyword heat feature corresponding to the target keyword; the keyword heat feature at least includes a first heat feature and a second heat feature; the first heat feature represents the heat value of each of the target keywords at the implementation time, and the second heat feature represents the heat value growth rate of each of the music keywords; The keyword heat feature of the target keyword and the real-time complement feature of the target keyword are combined to generate the model training feature.

5. The method according to claim 4, characterized in that The step of combining the keyword heat feature of the target keyword and the real-time complement feature of the target keyword to generate the model training feature includes: Acquire the correlation feature associated with the target keyword and the target scene; wherein the target scene is a scene in which the user performs the user interactive operation on the music object corresponding to the target keyword; The keyword heat feature of the target keyword, the real-time complement feature of the target keyword, and the relevance feature of the target keyword are combined to generate the model training feature.

6. A method for displaying music association words, characterized in that: The method comprises: In response to a music search term input by a user, recalling a plurality of associated words corresponding to the music search term; The plurality of associated words are sorted by a pre-trained real-time heat response model to obtain a sorting result of the plurality of associated words; the pre-trained real-time heat response model is obtained by training using the model training feature according to any one of claims 1 to 5; According to the sorting results of the multiple associative words, the multiple associative words corresponding to the music search word are displayed.

7. The method according to claim 6, characterized in that The recalling of multiple associated words corresponding to the music search word includes: Through a pre-trained recall model, multiple associative words corresponding to the music search word are recalled from an associative word database; wherein the associative word database records an index relationship between the search word and the associative word, and the index relationship is created when the number of times the search word is searched is greater than a preset threshold.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images