Data processing method and related product

Abstract

The invention discloses a data processing method and a related product. The method comprises the steps of obtaining S pieces of first granularity content data in content data in an original interaction sequence; performing content screening on the original interaction sequence according to a first similarity between the query data and content data in the original interaction sequence and a second similarity between the query data and the S pieces of first granularity content data to obtain a first screening sequence; obtaining R pieces of second granularity content data in the recommendation data in the recommendation sequence; according to a third similarity between recommendation data in the recommendation sequence and content data in the first screening sequence, R pieces of second granularity content data in the recommendation data in the recommendation sequence and granularity content data with the same granularity type as the R pieces of second granularity content data in the first screening sequence, performing content screening on the first screening sequence, obtaining a second screening sequence; and sorting the recommended data according to the second screening sequence to obtain sorted recommended data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of natural language processing, and particularly relates to a data processing method and related products. BACKGROUND

[0002] In a search ranking scenario, after obtaining a recommendation sequence for recommendation to a user, the original interaction sequence of the user can also be used to predict the interest degree of the user to the recommended data in the recommendation sequence, and the recommended data with a high interest degree is preferentially arranged in the front, so as to improve the user experience. However, the content data contained in the original interaction sequence of the user can reach tens of thousands or even hundreds of thousands in length, so it is necessary to preliminarily screen the original interaction sequence of the user, select relevant content data, and perform ranking on the recommendation sequence, so as to reduce the calculation pressure and improve the calculation efficiency.

[0003] The existing sequence screening usually adopts a two-stage QSU-GSU (Query Search Unit & General Search Unit) sequence screening scheme, wherein the QSU stage mainly uses the overall similarity matching of query data-content data for screening, and the GSU stage mainly uses the overall similarity matching of content data-content data for screening. However, when the information difference of each granularity content data contained in the content data is large or the content data contains multi-modal data, the overall feature vector corresponding to the content data is often difficult to reflect the local feature or cross-modal information, which will cause that the truly relevant content data may be missed in the screening process, or the irrelevant content data may be retained, thereby affecting the accuracy of subsequent ranking of the recommended data. SUMMARY

[0004] The present application provides a data processing method and related products, wherein the related products include a data processing device, an electronic device, a computer readable storage medium and a computer program product.

[0005] In a first aspect, a data processing method is provided, comprising:

[0006] performing first granularity content splitting on content data in an original interaction sequence corresponding to an interaction object, to obtain S first granularity content data in the content data in the original interaction sequence; the S is an integer greater than 1;

[0007] obtaining a first similarity between query data and the content data in the original interaction sequence, obtaining a second similarity between the query data and the S first granularity content data respectively, and performing content screening on the original interaction sequence according to the first similarity and the S second similarities, to obtain a first screening sequence;

[0008] perform second granularity content splitting on the recommendation data in the recommendation sequence associated with the query data to obtain R second granularity content data in the recommendation data in the recommendation sequence; the R is an integer greater than 1;

[0009] obtain a third similarity between the recommendation data in the recommendation sequence and the content data in the first screening sequence, and perform content screening on the first screening sequence according to the third similarity, the R second granularity content data in the recommendation sequence in the recommendation data, and granularity content data in the first screening sequence that has the same granularity type as the R second granularity content data, to obtain a second screening sequence;

[0010] sort the recommendation data in the recommendation sequence according to the second screening sequence to obtain sorted recommendation data, and recommend the sorted recommendation data to the interactive object.

[0011] According to any one of the embodiments of the present application, the number of content data in the original interactive sequence is N; the N is an integer greater than 1;

[0012] obtaining a first similarity between the query data and the content data in the original interactive sequence, obtaining a second similarity between the query data and the S first granularity content data respectively, and performing content screening on the original interactive sequence according to the first similarity and the S second similarities to obtain a first screening sequence, comprising:

[0013] traversing the N content data in the original interactive sequence to obtain the kth content data; k is a positive integer less than or equal to N;

[0014] determining the vector cosine similarity between the query data and the kth content data as the first similarity corresponding to the kth content data;

[0015] determining the vector cosine similarity between the query data and the S first granularity content data in the kth content data as the S second similarities corresponding to the kth content data;

[0016] When the N content data in the original interactive sequence is traversed, performing content screening on the original interactive sequence according to the first similarity and the S second similarities corresponding to each content data to obtain a first screening sequence.

[0017] According to any one of the embodiments of the present application, the content screening on the original interactive sequence according to the first similarity and the S second similarities corresponding to each content data to obtain a first screening sequence, comprising:

[0018] According to the first similarity corresponding to each content data, the N content data are subjected to first sorting processing, a first sorting sequence is obtained, and the first M content data in the first sorting sequence are determined as a first screening subsequence; the M is a preset screening parameter;

[0019] Each content data respectively corresponds to S second similarities, and the jth second similarity corresponding to each content data is obtained; j is a positive integer less than or equal to S;

[0020] According to the jth second similarity corresponding to each content data, the N content data are subjected to second sorting processing, a second sorting sequence is obtained, and the first M content data in the second sorting sequence are determined as a jth second screening subsequence;

[0021] When the S second similarities corresponding to each content data are traversed, the first screening subsequence and the S second screening subsequences are subjected to merging processing, and a first screening sequence is obtained.

[0022] According to any one of the embodiments of the present application, the first similarity between the query data and the content data in the original interaction sequence is obtained, the second similarity between the query data and the S first granularity content data is obtained, the original interaction sequence is subjected to content screening according to the first similarity and the S second similarities, and a first screening sequence is obtained, which includes:

[0023] The query vector corresponding to the query data is obtained, the content vector corresponding to the content data in the original interaction sequence and the granularity content vector corresponding to the S first granularity content data are obtained;

[0024] The cosine similarity between the query vector and the content vector is calculated, and the first similarity between the query data and the content data in the original interaction sequence is obtained;

[0025] The cosine similarity between the query vector and the S granularity content vectors is calculated, and the second similarity between the query data and the S first granularity content data is obtained;

[0026] The original interaction sequence is subjected to content screening according to the first similarity and the S second similarities, and a first screening sequence is obtained.

[0027] According to any one of the embodiments of the present application, the query vector corresponding to the query data is obtained, the content vector corresponding to the content data in the original interaction sequence and the granularity content vector corresponding to the S first granularity content data are obtained, which includes:

[0028] perform single-granularity encoding processing on the query data based on the multi-modal vector encoding model to obtain a query vector corresponding to the query data;

[0029] obtain global identification data corresponding to the content data in the original interaction sequence, concatenate the S first-granularity content data, the S granularity content delimiters, and the global identification data to obtain multi-granularity content data corresponding to the content data in the original interaction sequence; the end of a first-granularity content data in the multi-granularity content data is connected with a granularity content delimiter, and the global identification data is connected at the end of the last granularity content delimiter;

[0030] perform multi-granularity encoding processing on the multi-granularity content data based on the multi-modal vector encoding model to obtain a global vector corresponding to the global identification data and granularity content vectors respectively corresponding to the S first-granularity content data, and determine the global vector as a content vector corresponding to the content data in the original interaction sequence.

[0031] According to any one of the embodiments of the present application, the number of recommendation data in the recommendation sequence is H, and the H recommendation data include recommendation data P i H is a positive integer, i is a positive integer less than or equal to H; the number of the second screening sequence is H, and one second screening sequence corresponds to one recommendation data in the recommendation sequence; the number of content data in the first screening sequence is L, and the L is an integer greater than 1;

[0032] The third similarity between the recommendation data in the recommendation sequence and the content data in the first screening sequence is obtained, and the first screening sequence is content-screened according to the third similarity, R second-granularity content data in the recommendation data in the recommendation sequence, and granularity content data in the first screening sequence having the same granularity type as the R second-granularity content data, to obtain a second screening sequence, including:

[0033] The L content data in the first screening sequence is traversed to obtain the jth content data; j is a positive integer less than or equal to L;

[0034] The vector cosine similarity between the recommendation data P i and the jth content data is determined as the third similarity corresponding to the jth content data;

[0035] R granularity content data pairs are generated according to the R second-granularity content data in the recommendation data P i and the jth content data; one granularity content data pair includes the recommendation data P ione second granularity content data in the first screening sequence and the jth content data in the first screening sequence have the same granularity type;

[0036] The vector cosine similarity of each of the R pairs of granularity content data is determined as a fourth similarity corresponding to the jth content data;

[0037] When the L content data in the first screening sequence are traversed, the first screening sequence is screened according to the third similarity and the fourth similarity corresponding to the L content data, respectively, to obtain the recommended data P i corresponding second screening sequence.

[0038] According to any one of the embodiments of the present application, the number of the second screening sequences is the same as the number of the recommended data in the recommended sequence; one second screening sequence corresponds to one recommended data in the recommended sequence.

[0039] The sorting of the recommended data in the recommended sequence according to the second screening sequence to obtain the sorted recommended data and the recommendation of the sorted recommended data to the interactive object include:

[0040] Each recommended data in the recommended sequence is subjected to object interaction prediction processing according to each second screening sequence, respectively, to obtain an object interaction interest degree corresponding to each recommended data;

[0041] The recommended data in the recommended sequence is sorted according to the object interaction interest degree corresponding to each recommended data to obtain the sorted recommended data; the recommended data corresponding to the greater object interaction interest degree is located in the front position in the sorted recommended data.

[0042] The sorted recommended data is recommended to the interactive object.

[0043] According to any one of the embodiments of the present application, the sorting of the recommended data in the recommended sequence according to the second screening sequence to obtain the sorted recommended data and the recommendation of the sorted recommended data to the interactive object include:

[0044] The second screening sequence is subjected to content clustering processing to obtain X content clustering clusters; one content clustering cluster contains one or more content data in the second screening sequence; X is an integer greater than or equal to 1.

[0045] The X content clustering clusters are subjected to content data filtering processing, respectively, to obtain X filtered content clustering clusters.

[0046] The X filtered content clustering clusters are subjected to merging and screening processing to obtain a third screening sequence.

[0047] According to the third screening sequence, the recommended data in the recommended sequence is sorted to obtain sorted recommended data, and the sorted recommended data is recommended to the interactive object.

[0048] According to any one of the embodiments of the present application, the X content cluster groups include a target content cluster group;

[0049] The content data filtering processing on the X content cluster groups respectively to obtain X filtered content cluster groups includes:

[0050] The time decay weight corresponding to each content data in the target content cluster group is determined respectively;

[0051] If the number of content data in the target content cluster group is less than or equal to a first number threshold, the target content cluster group is filtered according to the time decay weight corresponding to each content data in the target content cluster group to obtain a target filtered content cluster group;

[0052] If the number of content data in the target content cluster group is greater than the first number threshold, the cluster center vector corresponding to the target content cluster group is determined according to the time decay weight corresponding to each content data included in the target content cluster group and the content vector corresponding to each content data included in the target content cluster group;

[0053] The vector similarity between the cluster center vector and the content vector of each content data included in the target content cluster group is determined, and the target content cluster group is filtered according to the vector similarity to obtain a target filtered content cluster group.

[0054] According to any one of the embodiments of the present application, the filtering processing on the target content cluster group according to the time decay weight corresponding to each content data in the target content cluster group to obtain a target filtered content cluster group includes:

[0055] From the target content cluster group, content data with a time decay weight less than a time decay weight threshold is deleted to obtain an intermediate content cluster group;

[0056] If the intermediate content cluster group is not empty, the content data with the maximum time decay weight in the intermediate content cluster group is determined as the target filtered content cluster group;

[0057] If the intermediate content cluster group is empty, the empty cluster is determined as the target filtered content cluster group.

[0058] In a second aspect, a data processing apparatus is provided, including:

[0059] The first splitting module is configured to perform first-granularity content splitting on the content data in the original interaction sequence corresponding to the interaction object, to obtain S first-granularity content data in the content data in the original interaction sequence; S is an integer greater than 1.

[0060] The first screening module is configured to obtain a first similarity between the query data and the content data in the original interaction sequence, obtain a second similarity between the query data and the S first-granularity content data respectively, and perform content screening on the original interaction sequence according to the first similarity and the S second similarities, to obtain a first screened sequence.

[0061] The second splitting module is configured to perform second-granularity content splitting on recommendation data in a recommendation sequence associated with the query data, to obtain R second-granularity content data in the recommendation data in the recommendation sequence; R is an integer greater than 1.

[0062] The second screening module is configured to obtain a third similarity between the recommendation data in the recommendation sequence and the content data in the first screened sequence, and perform content screening on the first screened sequence according to the third similarity, the R second-granularity content data in the recommendation data in the recommendation sequence, and granularity content data in the first screened sequence that has the same granularity type as the R second-granularity content data, to obtain a second screened sequence.

[0063] The sorting module is configured to sort the recommendation data in the recommendation sequence according to the second screened sequence, to obtain sorted recommendation data, and recommend the sorted recommendation data to the interaction object.

[0064] According to any one of the embodiments of the present application, the number of content data in the original interaction sequence is N; N is an integer greater than 1.

[0065] The first screening module is configured to obtain a first similarity between the query data and the content data in the original interaction sequence, obtain a second similarity between the query data and the S first-granularity content data respectively, and perform content screening on the original interaction sequence according to the first similarity and the S second similarities, to obtain a first screened sequence.

[0066] The first screening module is configured to obtain a first similarity between the query data and the content data in the original interaction sequence, obtain a second similarity between the query data and the S first-granularity content data respectively, and perform content screening on the original interaction sequence according to the first similarity and the S second similarities, to obtain a first screened sequence.

[0067] The first screening module is configured to obtain a first similarity between the query data and the content data in the original interaction sequence, obtain a second similarity between the query data and the S first-granularity content data respectively, and perform content screening on the original interaction sequence according to the first similarity and the S second similarities, to obtain a first screened sequence.

[0068] vector cosine similarity between the query data and S first granularity content data in the kth content data is determined as S second similarities corresponding to the kth content data;

[0069] When the N content data in the original interaction sequence are traversed, the original interaction sequence is content filtered according to the first similarity and the S second similarities corresponding to each content data, to obtain a first filtered sequence.

[0070] According to any one of the embodiments of the present application, the first filtering module is specifically configured to:

[0071] The N content data are first sorted according to the first similarity corresponding to each content data, to obtain a first sorted sequence, and the first M content data in the first sorted sequence are determined as a first filtered subsequence; the M is a preset filtering parameter;

[0072] The S second similarities corresponding to each content data are traversed, and the jth second similarity corresponding to each content data is obtained; j is a positive integer less than or equal to S;

[0073] The N content data are second sorted according to the jth second similarity corresponding to each content data, to obtain a second sorted sequence, and the first M content data in the second sorted sequence are determined as a jth second filtered subsequence;

[0074] When the S second similarities corresponding to each content data are traversed, the first filtered subsequence and the S second filtered subsequences are merged, to obtain the first filtered sequence.

[0075] According to any one of the embodiments of the present application, the first filtering module is specifically configured to:

[0076] The query vector corresponding to the query data is obtained, the content vector corresponding to the content data in the original interaction sequence and the granularity content vector corresponding to the S first granularity content data are obtained;

[0077] The first similarity between the query data and the content data in the original interaction sequence is obtained by performing cosine similarity calculation on the query vector and the content vector;

[0078] performing cosine similarity calculation on the query vector and the S granularity content vectors respectively to obtain second similarities between the query data and the S first granularity content data respectively;

[0079] performing content screening on the original interaction sequence according to the first similarity and the S second similarities to obtain a first screening sequence.

[0080] According to any one of the embodiments of the present application, the query vector corresponding to the query data is obtained, the content vector corresponding to the content data in the original interaction sequence and the granularity content vector corresponding to the S first granularity content data are obtained, and the first screening module is specifically configured to:

[0081] performing single granularity encoding processing on the query data based on a multi-modal vector encoding model to obtain a query vector corresponding to the query data;

[0082] obtaining global identification data corresponding to the content data in the original interaction sequence, performing splicing processing on the S first granularity content data, S granularity content delimiters and the global identification data to obtain multi-granularity content data corresponding to the content data in the original interaction sequence; the end of one first granularity content data in the multi-granularity content data is connected with one granularity content delimiter, and the global identification data is connected at the end of the last granularity content delimiter;

[0083] performing multi-granularity encoding processing on the multi-granularity content data based on the multi-modal vector encoding model to obtain a global vector corresponding to the global identification data and granularity content vectors corresponding to the S first granularity content data respectively, and determining the global vector as the content vector corresponding to the content data in the original interaction sequence.

[0084] According to any one of the embodiments of the present application, the number of recommendation data in the recommendation sequence is H, and the H recommendation data include recommendation data P i , H is a positive integer, i is a positive integer less than or equal to H; the number of the first screening sequence is H, and one second screening sequence corresponds to one recommendation data in the recommendation sequence; the number of content data in the first screening sequence is L, and the L is an integer greater than 1;

[0085] The third similarity between the recommendation data in the recommendation sequence and the content data in the first screening sequence is obtained, and the first screening sequence is screened according to the third similarity, R second granularity content data in the recommendation data in the recommendation sequence and granularity content data in the first screening sequence having the same granularity type as the R second granularity content data, to obtain a second screening sequence, and the second screening module is specifically configured to:

[0086] obtaining the jth content data from the L content data in the first screening sequence; j is a positive integer less than or equal to L;

[0087] determining the vector cosine similarity between the recommendation data P i and the jth content data as a third similarity corresponding to the jth content data;

[0088] generating R granularity content data pairs according to the R second granularity content data in the recommendation data P i and the jth content data; one granularity content data pair includes one second granularity content data in the recommendation data P i and the granularity content data in the jth content data having the same granularity type as the one second granularity content data;

[0089] determining the vector cosine similarity of the R granularity content data pairs respectively as a fourth similarity corresponding to the jth content data;

[0090] when the L content data in the first screening sequence are traversed, performing content screening on the first screening sequence according to the third similarity and the fourth similarity corresponding to the L content data respectively, to obtain a second screening sequence corresponding to the recommendation data P i .

[0091] According to any one of the embodiments of the present application, the number of the second screening sequences is the same as the number of the recommendation data in the recommendation sequence; one second screening sequence corresponds to one recommendation data in the recommendation sequence;

[0092] The sorting module is specifically configured to:

[0093] performing object interaction prediction processing on each recommendation data in the recommendation sequence according to each second screening sequence respectively, to obtain an object interaction interest degree corresponding to each recommendation data;

[0094] sorting the recommendation data in the recommendation sequence according to the object interaction interest degree corresponding to each recommendation data, to obtain sorted recommendation data; in the sorted recommendation data, the recommendation data with a higher position corresponds to a greater object interaction interest degree;

[0095] recommending the sorted recommendation data to the interactive object.

[0096] According to any one of the embodiments of the present application, the sorting module is specifically configured to:

[0097] The second filtering sequence is subjected to content clustering processing to obtain X content clustering clusters; one content clustering cluster contains one or more content data in the second filtering sequence; X is an integer greater than or equal to 1;

[0098] The X content clustering clusters are subjected to content data filtering processing respectively to obtain X filtered content clustering clusters;

[0099] The X filtered content clustering clusters are subjected to merging and filtering processing to obtain a third filtering sequence;

[0100] According to the third filtering sequence, the recommended data in the recommendation sequence are sorted to obtain sorted recommended data, and the sorted recommended data are recommended to the interactive object.

[0101] According to any one of the embodiments of the present application, the X content clustering clusters contain a target content clustering cluster;

[0102] The clustering module is specifically configured to:

[0103] The time decay weight corresponding to each content data in the target content clustering cluster is determined respectively;

[0104] If the number of content data in the target content clustering cluster is less than or equal to a first quantity threshold, the target content clustering cluster is subjected to filtering processing according to the time decay weight corresponding to each content data in the target content clustering cluster respectively to obtain a target filtered content clustering cluster;

[0105] If the number of content data in the target content clustering cluster is greater than the first quantity threshold, the cluster center vector corresponding to the target content clustering cluster is determined according to the time decay weight corresponding to each content data contained in the target content clustering cluster and the content vector corresponding to each content data contained in the target content clustering cluster;

[0106] The vector similarity between the cluster center vector and the content vector of each content data contained in the target content clustering cluster is determined, and the target content clustering cluster is subjected to filtering processing according to the vector similarity to obtain a target filtered content clustering cluster.

[0107] According to the time decay weight corresponding to the content data in the target content clustering cluster, the target content clustering cluster is filtered to obtain a target filtered content clustering cluster.

[0108] From the target content clustering cluster, content data corresponding to a time decay weight less than a time decay weight threshold is deleted to obtain an intermediate content clustering cluster.

[0109] If the intermediate content clustering cluster is not empty, content data corresponding to the maximum time decay weight in the intermediate content clustering cluster is determined as the target filtered content clustering cluster.

[0110] If the intermediate content clustering cluster is empty, the empty cluster is determined as the target filtered content clustering cluster.

[0111] In a third aspect, an electronic device is provided, including a processor and a memory, the memory being configured to store computer program code, the computer program code including computer instructions, and when the processor executes the computer instructions, the electronic device performs the first aspect and any one of the embodiments thereof.

[0112] In a fourth aspect, another electronic device is provided, including a processor, a sending device, an input device, an output device and a memory, the memory being configured to store computer program code, the computer program code including computer instructions, and when the processor executes the computer instructions, the electronic device performs the first aspect and any one of the embodiments thereof.

[0113] In a fifth aspect, a computer readable storage medium is provided, the computer readable storage medium storing a computer program, the computer program including program instructions, and when the program instructions are executed by a processor, the processor performs the first aspect and any one of the embodiments thereof.

[0114] In a sixth aspect, a computer program product is provided, the computer program product including a computer program or instructions, and when the computer program or instructions are run on a computer, the computer performs the first aspect and any one of the embodiments thereof.

[0115] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, but not limiting the present application.

[0116] In the present application, the content data in the original interaction sequence corresponding to the interactive object can be first granularity content split to obtain S first granularity content data in the content data in the original interaction sequence, then the first similarity between the query data and the content data in the original interaction sequence is obtained, the second similarity between the query data and the S first granularity content data is obtained, the original interaction sequence is content screened according to the first similarity and the S second similarities, and a first screened sequence is obtained; then the recommendation data in the recommendation sequence associated with the query data is granularity content split to obtain R second granularity content data in the recommendation data in the recommendation sequence, then the third similarity between the recommendation data in the recommendation sequence and the content data in the first screened sequence is obtained, and the first screened sequence is content screened according to the third similarity, the R second granularity content data in the recommendation data in the recommendation sequence and the granularity content data in the first screened sequence having the same granularity type as the R second granularity content data, and a second screened sequence is obtained; finally, the recommendation data in the recommendation sequence is sorted according to the second screened sequence, the sorted recommendation data is obtained, and the sorted recommendation data is recommended to the interactive object. Through the method provided by the present application, the dimension of sequence screening is expanded by introducing S first granularity content data and R second granularity content data, effectively solving the problem that the characteristics of local data are difficult to be reflected by the overall representation of content data, and when calculating the similarity corresponding to different granularity content data, the corresponding modal type can be flexibly processed, effectively solving the problem that when the content data is represented as a whole, the cross-modal information is not well understood, thereby reducing the omission of truly relevant content data, and effectively improving the relevance of the screened sequence and the accuracy of sorting the recommendation data. BRIEF DESCRIPTION OF DRAWINGS

[0117] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the background art, the drawings needed to be used in the embodiments of the present application or the background art will be described below.

[0118] The drawings herein are incorporated into the specification and form part of the specification, which illustrate the embodiments consistent with the present application, and together with the specification, serve to explain the technical solutions of the present application.

[0119] Figure 1 A network architecture schematic diagram is provided for the embodiments of the present application;

[0120] Figure 2 A flowchart of a data processing method is provided for the embodiments of the present application;

[0121] Figure 3 A flowchart of a data sorting method is provided for the embodiments of the present application;

[0122] Figure 4 A schematic diagram of an overall flow of a note sequence screening provided by an embodiment of the present application is shown in FIG. 1.

[0123] Figure 5 A schematic diagram of a structure of a content recommendation device provided by an embodiment of the present application is shown in FIG. 2.

[0124] Figure 6 A schematic diagram of a hardware structure of an electronic device provided by an embodiment of the present application is shown in FIG. 3. DETAILED DESCRIPTION

[0125] In order to make the personnel in the technical field better understand the present application scheme, the technical scheme in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by the person skilled in the art without creative labor are within the scope of protection of the present application.

[0126] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish different objects, not to describe a specific order. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0127] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily mutually exclusive of other embodiments. It is explicitly and implicitly understood that the embodiments described herein can be combined.

[0128] The embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0129] Please refer to Figure 1 , Figure 1 A schematic diagram of a network architecture provided by an embodiment of the present application is shown in FIG. 4. As shown in Figure 1 the network architecture can include a server 2000 and a terminal cluster. The terminal cluster can specifically include one or more terminals, and the number of terminals in the terminal cluster will not be limited here. As shown in Figure 1As shown, the multiple terminals may specifically include terminal 3000a, terminal 3000b, terminal 3000c, ..., terminal 3000n; terminal 3000a, terminal 3000b, terminal 3000c, ..., terminal 3000n can be directly or indirectly connected to server 2000 via wired or wireless communication, so that each terminal can interact with server 2000 through the network connection. It is understood that the above network connection is not limited to a specific connection method; it can be directly or indirectly connected via wired communication, wireless communication, or other methods. This application does not impose any restrictions on this.

[0130] Each terminal in the terminal cluster can include: smartphones, tablets, laptops, desktop computers, smart voice interaction devices, smart home appliances (e.g., smart TVs), wearable devices, in-vehicle terminals, aircraft, and other smart terminals with data processing capabilities. It should be understood that, as... Figure 1 Each terminal in the terminal cluster shown can have an application client installed. When the application client runs on each terminal, it can interact with the aforementioned... Figure 1 Data exchange occurs between the 2000 servers shown.

[0131] Specifically, the application client can include in-vehicle clients, smart home clients, entertainment clients (e.g., game clients), multimedia clients (e.g., video clients), social clients, and information clients (e.g., news clients) with data processing capabilities. In this embodiment, the application client can be integrated into a client (e.g., a social client), or it can be a standalone client (e.g., a news client). This embodiment does not limit the type of application client.

[0132] Among them, such as Figure 1 The server 2000 shown can be the server corresponding to the application client. The server 2000 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0133] For ease of understanding, the embodiments of this application may be described in detail below. Figure 1 One of the multiple terminals shown is selected as the interactive terminal. For example, in the embodiments of this application, a terminal can be selected as the interactive terminal. Figure 1The terminal 3000a shown serves as an interactive terminal, which can be equipped with an application client capable of data processing. In this case, the interactive terminal can interact with the server 2000 via the application client. Furthermore, the user corresponding to the interactive terminal can be referred to as the interaction object.

[0134] like Figure 1 As shown, the interactive terminal can generate a content recommendation request for the query data through a running application client, and then send the content recommendation request to the server 2000. After receiving the content recommendation request, the server 2000 can obtain a recommendation sequence associated with the query data according to preset recommendation rules. This recommendation sequence may contain one or more recommended data. Typically, after obtaining the recommendation sequence, the server 2000 needs to obtain the most relevant content data to the query data from the original interaction sequence of the interactive object's historical interactions, and sort the recommended data in the recommendation sequence, placing the recommended data that is predicted to be more likely to be of interest to the interactive object at the top, in order to improve the interactive object's satisfaction with the returned recommendation sequence.

[0135] To improve the relevance of the selected sequences to the query and recommendation data, this application proposes a data processing method that introduces more fine-grained content data to filter the original interaction sequences. The data processing method can be specifically as follows: First-level content splitting is performed on the content data in the original interaction sequence corresponding to the interactive object, resulting in S first-level content data items in the original interaction sequence; S is an integer greater than 1. First similarity is obtained between the query data and the content data in the original interaction sequence; second similarity is obtained between the query data and each of the S first-level content data items; content filtering is performed on the original interaction sequence based on the first similarity and the S second similarities to obtain a first filtering sequence. Second-level content splitting is performed on the recommended data in the recommendation sequence associated with the query data, resulting in R second-level content data items in the recommended data in the recommendation sequence; R is an integer greater than 1. Third similarity is obtained between the recommended data in the recommendation sequence and the content data in the first filtering sequence; content filtering is performed on the first filtering sequence based on the third similarity, the R second-level content data items in the recommended data in the recommendation sequence, and the granular content data in the first filtering sequence that has the same granularity type as the R second-level content data items, to obtain a second filtering sequence. Finally, the recommended data in the recommendation sequence is sorted according to the second filtering sequence to obtain sorted recommended data, and the sorted recommended data is recommended to the interactive object.

[0136] Further, the embodiment of the present application provides a data processing method, wherein the execution subject of the data processing method is a data processing device, wherein the data processing device can be any kind of electronic device capable of executing the technical solutions disclosed in the method embodiments of the present application. Optionally, the data processing device can be one of the following: a computer, a server. It should be understood that the method embodiments of the present application can also be implemented in the manner that the processor executes computer program codes. The data processing method provided by the embodiment of the present application can be used to screen out a second screening sequence related to the recommended data, and sort the recommended sequence according to the second screening sequence to obtain the sorted recommended data for sending to the interactive object. In the embodiment of the present application, the content data can include: text, video, image, audio, graphic notes and other data.

[0137] Please refer to Figure 2 , Figure 2 The flowchart of the data processing method provided by the embodiment of the present application is shown in the figure.

[0138] S101, the content data in the original interaction sequence corresponding to the interactive object is first granularity content split, and S first granularity content data in the content data in the original interaction sequence is obtained; S is an integer greater than 1.

[0139] Specifically, the interactive object refers to a user who initiates a content recommendation request, which can be a query request initiated by the user when using a business application, which can contain query data. Wherein the business application can be a video application, a note application, a music application and any application with content recommendation function, and the query data can be a search word or a search short drama input by the user, which is used to represent the user's query demand. The original interaction sequence corresponding to the interactive object contains all content data of the interactive object in the process of using the business application. It can be understood that the longer the interactive object uses the business application, the longer the length of the original interaction sequence may be, which can reach tens of thousands or hundreds of thousands. Wherein the interactive behavior can include reading, liking and other behaviors.

[0140] Specifically, the content data often contains multiple granularity data, for example, when the content data is a note, the granularity content data can be a title, a picture or a note body, etc., which can be understood as the partial content data contained in the content data. A feasible implementation process of performing first granularity content splitting on the content data in the original interaction sequence corresponding to the interaction object to obtain S first granularity content data in the content data in the original interaction sequence can be: performing first granularity content splitting on each content data in the original interaction sequence corresponding to the interaction object according to a preset granularity splitting rule to obtain S first granularity content data corresponding to each content data in the original interaction sequence respectively. It can be understood that the preset granularity splitting rule can be determined based on actual conditions, for example, a feasible preset granularity splitting rule can be to split the content data according to the modal type, that is, to determine the data of different modal types as different granularity content data, in addition, for the main modal data contained in the content data, the main modal data refers to the modal data with the largest data amount or the most important modal data in the content data, which can be split in a finer granularity according to the structure of the main modal data. For example, when the content data is a note, it can be split into text modal data and picture modal data according to the modal type, and the text modal data as the main modal data of the note can be split into a title and a body based on its structure, therefore, when performing granularity content splitting on a note, S granularity content data that can be obtained can be a title, a body and a picture.

[0141] S102, obtaining a first similarity between the query data and the content data in the original interaction sequence, obtaining a second similarity between the query data and the S first granularity content data respectively, and performing content screening on the original interaction sequence according to the first similarity and the S second similarities to obtain a first screened sequence.

[0142] Specifically, the number of content data in the original interaction sequence is N; N is an integer greater than 1; the first similarity between the query data and the content data in the original interaction sequence is obtained, the second similarity between the query data and the S first granularity content data is obtained, and the original interaction sequence is content filtered according to the first similarity and the S second similarities to obtain a feasible implementation process of the first filtered sequence, which can be: traversing the N content data in the original interaction sequence, obtaining the kth content data; k is a positive integer less than or equal to N; the vector cosine similarity between the query data and the kth content data is determined as the first similarity corresponding to the kth content data; the vector cosine similarity between the query data and the S first granularity content data in the kth content data is determined as the S second similarities corresponding to the kth content data; when the N content data in the original interaction sequence is traversed, the original interaction sequence is content filtered according to the first similarity and the S second similarities corresponding to each content data to obtain the first filtered sequence. Wherein, the vector cosine similarity between the query data and the kth content data refers to the cosine similarity between the query vector corresponding to the query data and the content vector corresponding to the kth content data. Wherein, the vector cosine similarity between the query data and the S first granularity content data in the kth content data includes the vector cosine similarity between the query data and the first granularity content data S a in the kth content data, a is a positive integer less than or equal to S, and the vector cosine similarity between the query data and the first granularity content data S a in the kth content data refers to the cosine similarity between the query vector corresponding to the query data and the granularity content vector corresponding to the first granularity content data S a in the kth content data.

[0143] Specifically, a feasible implementation process of content screening on the original interaction sequence according to the first similarity and the S second similarities corresponding to each content data can be: performing first sorting processing on the N content data according to the first similarity corresponding to each content data to obtain a first sorting sequence, determining the first M content data in the first sorting sequence as a first screening subsequence; M is a preset screening parameter; traversing the S second similarities corresponding to each content data to obtain the jth second similarity corresponding to each content data; j is a positive integer less than or equal to S; performing second sorting processing on the N content data according to the jth second similarity corresponding to each content data to obtain a second sorting sequence, determining the first M content data in the second sorting sequence as a jth second screening subsequence; when the S second similarities corresponding to each content data are traversed, performing merging processing on the first screening subsequence and the S second screening subsequences to obtain the first screening sequence. For ease of understanding, assuming that the original interaction sequence contains [note 1, note 2, …, note 10000], the S granularity content data corresponding to a note can contain a picture, a title, and a body, then the first similarity corresponding to a note refers to a vector cosine similarity between the note and the query data (referred to as “query data-note similarity”), and the S second similarities corresponding to the note include a vector cosine similarity between the query data and the picture (referred to as “query data-picture similarity”), a vector cosine similarity between the query data and the title (referred to as “query data-title similarity”), and a vector cosine similarity between the query data and the body (referred to as “query data-body similarity”), then the original interaction sequence can be screened four times, and 300 notes are screened out each time. In the first screening, the original interaction sequence is sorted from large to small according to the query data-note similarity corresponding to each note, that is, the smaller the corresponding query data-note similarity, the higher the sorting position of the note, and then 300 notes are taken out from the sorted original interaction sequence as the first screened subsequence; similarly, the original interaction sequence is screened three times according to the query data-picture similarity, the query data-title similarity, and the query data-body similarity corresponding to each note, and a subsequence containing 300 notes is obtained each time, and then the four screened subsequences are merged, and the repeated notes are removed, thereby obtaining the first screening sequence.

[0144] Optionally, the value of M can be different for different granularity screening, for example, in the process of screening the original interaction sequence based on the query data-note similarity, the value of M can be 1000, and in the process of screening the original interaction sequence based on the query data-picture similarity, the value of M can be 300.

[0145] Specifically, one feasible implementation process of obtaining the first similarity between the query data and the content data in the original interaction sequence, obtaining the second similarity between the query data and the S first granularity content data respectively, and performing content screening on the original interaction sequence according to the first similarity and the S second similarities to obtain the first screened sequence can be: obtaining a query vector corresponding to the query data, obtaining a content vector corresponding to the content data in the original interaction sequence, and obtaining granularity content vectors corresponding to the S first granularity content data respectively; performing cosine similarity calculation on the query vector and the content vector to obtain the first similarity between the query data and the content data in the original interaction sequence; performing cosine similarity calculation on the query vector and the S granularity content vectors respectively to obtain the second similarity between the query data and the S first granularity content data respectively; and performing content screening on the original interaction sequence according to the first similarity and the S second similarities to obtain the first screened sequence.

[0146] Specifically, one feasible implementation process of obtaining the query vector corresponding to the query data, obtaining the content vector corresponding to the content data in the original interaction sequence, and obtaining the granularity content vectors corresponding to the S first granularity content data respectively can be: performing single-granularity encoding processing on the query data based on a multi-modal vector encoding model to obtain the query vector corresponding to the query data; obtaining global identification data corresponding to the content data in the original interaction sequence, and performing splicing processing on the S first granularity content data, the S granularity content separators, and the global identification data to obtain multi-granularity content data corresponding to the content data in the original interaction sequence; the end of one first granularity content data in the multi-granularity content data is connected with one granularity content separator, and the global identification data is connected at the end of the last granularity content separator; performing multi-granularity encoding processing on the multi-granularity content data based on the multi-modal vector encoding model to obtain a global vector corresponding to the global identification data and granularity content vectors corresponding to the S first granularity content data respectively, and determining the global vector as the content vector corresponding to the content data in the original interaction sequence. The multi-modal vector encoding model can be a model trained in advance using a multi-modal large language model structure. The global identification data can be data obtained by performing main modal identification processing on non-main modal data in the content data, for example, when the content data is a note, the main modal data of the note is text data, and the non-text modal data (such as video modal data, audio modal data, etc.) in the note can be processed by text recognition to obtain text recognition data, which is used as the global identification data corresponding to the note. The multi-granularity content data can be represented as “first granularity content data 1 [sep] first granularity content data 2 [sep] … first granularity content data S [sep] global identification data”.

[0147] The feasible implementation process of obtaining the global vector corresponding to the global identification data and the granularity content vectors corresponding to the S first granularity content data based on the multi-modal vector encoding model for multi-granularity content data processing can be: performing data preprocessing on the multi-granularity content data to obtain a multi-granularity content identification sequence; the multi-granularity content identification sequence includes granularity content identification sequences corresponding to the S first granularity content data, granularity content separator identifications corresponding to the S granularity content separators, and a global identification data identification sequence corresponding to the global identification data; performing vector encoding processing on the multi-granularity content identification sequence through the multi-modal vector encoding model to obtain a multi-granularity segmentation hidden layer vector matrix; the multi-granularity segmentation hidden layer vector matrix includes a hidden layer vector corresponding to each identification in the multi-granularity content identification sequence; obtaining the hidden layer vectors corresponding to the S granularity content separator identifications from the multi-granularity segmentation hidden layer vector matrix as the granularity content vectors corresponding to the S granularity content data; and obtaining the hidden layer vector corresponding to the last identification from the multi-granularity segmentation hidden layer vector matrix as the global vector corresponding to the global identification data.

[0148] S103, performing second granularity content splitting on the recommendation data in the recommendation sequence associated with the query data to obtain R second granularity content data in the recommendation data in the recommendation sequence; the R is an integer greater than 1.

[0149] Specifically, the preset granularity splitting rule relied on by the second granularity content splitting can be the same as the preset granularity splitting rule relied on by the first granularity content splitting, or can be adjusted according to business needs, such as reducing unnecessary granularity splitting, or increasing some finer granularity splitting, which is not limited in the present application. For example, when the content data is a note, the S first granularity content data corresponding thereto can include a title, a body and a picture. Considering that the effect of evaluating the relevance of the recommendation data and the content data by only using the title is poor, the title can not be used in the second screening stage, and therefore, the R second granularity content data corresponding thereto can include the body and the picture.

[0150] S104, obtaining a third similarity between the recommendation data in the recommendation sequence and the content data in the first screening sequence, and performing content screening on the first screening sequence according to the third similarity, the R second granularity content data in the recommendation data in the recommendation sequence, and the granularity content data in the first screening sequence having the same granularity type as the R second granularity content data, to obtain a second screening sequence.

[0151] Specifically, the number of recommendation data in the recommendation sequence is H, and the H recommendation data includes recommendation data P i, H is a positive integer, i is a positive integer less than or equal to H; the number of the second screening sequence is H, and one second screening sequence corresponds to one recommended data in the recommended sequence; the number of the content data in the first screening sequence is L, and L is an integer greater than 1; a third similarity between the recommended data in the recommended sequence and the content data in the first screening sequence is obtained, and the first screening sequence is content-screened according to the third similarity, R second-granularity content data in the recommended data in the recommended sequence and granularity content data in the first screening sequence having the same granularity type as the R second-granularity content data, to obtain a second screening sequence, which can be: traversing the L content data in the first screening sequence, obtaining the jth content data; j is a positive integer less than or equal to L; determining the vector cosine similarity between the recommended data P i and the jth content data as the third similarity corresponding to the jth content data; generating R granularity content data pairs according to the R second-granularity content data in the recommended data P i and the jth content data; one granularity content data pair contains one second-granularity content data in the recommended data P i and granularity content data in the jth content data having the same granularity type as the one second-granularity content data; determining the vector cosine similarity corresponding to the R granularity content data pairs as the fourth similarity corresponding to the jth content data; when the L content data in the first screening sequence are traversed, the first screening sequence is content-screened according to the third similarity and the fourth similarity corresponding to the L content data, to obtain the second screening sequence corresponding to the recommended data P i . Wherein, the vector cosine similarity between the recommended data P i and the jth content data refers to the cosine similarity between the recommended vector corresponding to the recommended data P i and the content vector corresponding to the jth content data. To better understand the generation of the R granularity content data pairs according to the R second-granularity content data in the recommended data P i and the jth content data, it is assumed that the recommended data P i is a note A1, the jth content data is a note A2, and the R second-granularity content data contained in the note A1 are the picture of the note A1 and the text of the note A1, then the R granularity content data pairs are [the picture of the note A1, the picture of the note A2] and [the text of the note A1, the text of the note A2]. Wherein, the vector cosine similarity corresponding to one granularity content data pair refers to the cosine similarity between the granularity content vectors corresponding to the two granularity content data contained in the granularity content data pair, for example, for [the picture of the note A1, the picture of the note A2], the corresponding vector cosine similarity is the cosine similarity between the picture vector corresponding to the picture of the note A1 and the picture vector corresponding to the picture of the note A2.

[0152] Specifically, the first screening sequence is content-screened according to the third similarity and the fourth similarity corresponding to the L content data, to obtain the recommendation data P i The screening process of the corresponding second screening sequence can refer to the process of content-screening the original interaction sequence according to the first similarity and the S second similarities corresponding to each content data, to obtain the first screening sequence, that is, the process of content-screening the first screening sequence according to the third similarity and the fourth similarity corresponding to the L content data, to obtain the recommendation data P i A feasible implementation process of the corresponding second screening sequence can be: performing third sorting processing on the L content data according to the third similarity corresponding to each content data, to obtain a third sorting sequence, determining the first P content data in the third sorting sequence as a third screening subsequence; P is a preset screening parameter; traversing the fourth similarity corresponding to each content data to obtain the kth fourth similarity corresponding to each content data; k is a positive integer less than or equal to R; performing fourth sorting processing on the L content data according to the kth fourth similarity corresponding to each content data, to obtain a fourth sorting sequence, determining the first P content data in the fourth sorting sequence as the kth fourth screening subsequence; when the fourth similarity corresponding to each content data is traversed, performing merging processing on the third screening subsequence and the R fourth screening subsequences, to obtain the second screening sequence.

[0153] S105, sorting the recommendation data in the recommendation sequence according to the second screening sequence, to obtain sorted recommendation data, and recommending the sorted recommendation data to the interaction object.

[0154] Specifically, the number of the second screening sequences is the same as the number of the recommendation data in the recommendation sequence; one second screening sequence corresponds to one recommendation data in the recommendation sequence; and a feasible implementation process of sorting the recommendation data in the recommendation sequence according to the second screening sequence to obtain sorted recommendation data and recommending the sorted recommendation data to the interaction object can be: performing object interaction prediction processing on each recommendation data in the recommendation sequence according to each second screening sequence, to obtain the object interaction interest degree corresponding to each recommendation data; sorting the recommendation data in the recommendation sequence according to the object interaction interest degree corresponding to each recommendation data, to obtain the sorted recommendation data; in the sorted recommendation data, the recommendation data with a higher position has a greater object interaction interest degree; and the sorted recommendation data is recommended to the interaction object.

[0155] In the embodiments of the present application, the content data in the original interaction sequence corresponding to the interaction object can be first granularity content split to obtain S first granularity content data in the content data in the original interaction sequence, then the first similarity between the query data and the content data in the original interaction sequence is obtained, the second similarity between the query data and the S first granularity content data is obtained, the original interaction sequence is content screened according to the first similarity and the S second similarities, and a first screened sequence is obtained; then the recommendation data in the recommendation sequence associated with the query data is granularity content split to obtain R second granularity content data in the recommendation data in the recommendation sequence, then the third similarity between the recommendation data in the recommendation sequence and the content data in the first screened sequence is obtained, and the first screened sequence is content screened according to the third similarity, the R second granularity content data in the recommendation data in the recommendation sequence, and the granularity content data in the first screened sequence having the same granularity type as the R second granularity content data, and a second screened sequence is obtained; finally, the recommendation data in the recommendation sequence is sorted according to the second screened sequence, sorted recommendation data is obtained, and the sorted recommendation data is recommended to the interaction object. Through the method provided in the embodiments of the present application, the dimension of sequence screening is expanded by introducing S first granularity content data and R second granularity content data, effectively solving the problem that the characteristics of local data are difficult to be reflected by the overall representation of content data, and when calculating the similarity corresponding to different granularity content data, the corresponding modal type can be flexibly processed, effectively solving the problem that when the content data is represented as a whole, the focus is on a certain modal type data, and the cross-modal information is not well understood, thereby reducing the omission of truly relevant content data, and effectively improving the relevance of the screened sequence and the accuracy of sorting the recommendation data.

[0156] Further, please refer to Figure 3 , Figure 3 The flowchart of a data sorting method provided in the embodiments of the present application. It can be understood that the data sorting method can be a specific and feasible embodiment of step S105 in the above-mentioned Figure 2 corresponding embodiments. As shown in Figure 3 , the data sorting method specifically includes the following steps:

[0157] S201, performing content clustering processing on the second screened sequence to obtain X content clustering clusters; one content clustering cluster contains one or more content data in the second screened sequence; and X is an integer greater than or equal to 1.

[0158] Specifically, the second screened sequence refers to the second screened sequence obtained after the steps S101-S104 in the above-mentioned Figure 2 corresponding embodiments, and the generation thereof will not be repeated herein.

[0159] Specifically, the content clustering processing can be implemented using a DBSACN algorithm (Density-Based Spatial Clustering of Applications with Noise). The DBSACN algorithm is a typical density clustering algorithm, has the advantage of resisting noise, and can determine the number of clustering clusters according to the actual data distribution, without needing to limit the cluster K value in advance. Alternatively, the content clustering processing can also be implemented using other interest clustering algorithms, which are not limited in the present application.

[0160] S202, respectively, the X content clustering clusters are subjected to content data filtering processing, and X filtered content clustering clusters are obtained.

[0161] Specifically, assuming that the X content clustering clusters include a target content clustering cluster, a feasible implementation process of respectively subjecting the X content clustering clusters to content data filtering processing to obtain X filtered content clustering clusters can be illustrated by taking the target content clustering cluster as an example, which can be: determining a time decay weight corresponding to each content data in the target content clustering cluster; if the number of content data in the target content clustering cluster is less than or equal to a first quantity threshold, filtering processing is performed on the target content clustering cluster according to the time decay weight corresponding to each content data in the target content clustering cluster, and a target filtered content clustering cluster is obtained; if the number of content data in the target content clustering cluster is greater than the first quantity threshold, a cluster center vector corresponding to the target content clustering cluster is determined according to the time decay weight corresponding to each content data included in the target content clustering cluster and the content vector corresponding to each content data included in the target content clustering cluster; a vector similarity between the cluster center vector and the content vector of each content data included in the target content clustering cluster is determined, and filtering processing is performed on the target content clustering cluster according to the vector similarity, and a target filtered content clustering cluster is obtained. The time decay weight is determined by a time difference between the current system time and the interaction time of the interactive object on the content data, and the greater the time difference, the smaller the time decay weight, so as to realize the quantitative measurement that the recent interaction behavior is more important than the historical interaction behavior. A feasible calculation process of the time decay weight can be represented by formula (1):

[0162] w = e -0.3Δt Formula (1)

[0163] wherein w is the time decay weight, and Δt refers to the time difference between the current system time and the interaction time of the interactive object on the content data.

[0164] Wherein, according to the time decay weight corresponding to each content data in the target content clustering cluster and the content vector corresponding to each content data in the target content clustering cluster, a feasible implementation process of determining the cluster center vector corresponding to the target content clustering cluster can be represented by formula (2):

[0165]

[0166] Wherein, center refers to the cluster center vector of the target content clustering cluster, w d refers to the time decay weight corresponding to the dth content data in the target content clustering cluster, emb d refers to the content vector corresponding to the dth content data in the target content clustering cluster, and c refers to the number of content data included in the target content clustering cluster.

[0167] Wherein, the first quantity threshold can be set based on actual conditions, and is usually small, for example, the first quantity threshold can be 2.

[0168] Specifically, according to the time decay weight corresponding to each content data in the target content clustering cluster, a feasible implementation process of filtering the target content clustering cluster to obtain the target filtered content clustering cluster can be: from the target content clustering cluster, deleting the content data corresponding to the time decay weight less than the time decay weight threshold to obtain an intermediate content clustering cluster; if the intermediate content clustering cluster is not empty, determining the content data corresponding to the maximum time decay weight in the intermediate content clustering cluster as the target filtered content clustering cluster; if the intermediate content clustering cluster is empty, determining the empty cluster as the target filtered content clustering cluster. The time decay weight threshold can be set based on actual conditions, for example, the time decay weight threshold is 0.01, assuming that the target content clustering cluster includes content data 1 and content data 2, wherein the time decay weight corresponding to content data 1 is 0.02, and the time decay weight corresponding to content data 2 is 0.03, since both are greater than the time decay weight threshold, neither is deleted, and the obtained intermediate content clustering cluster still includes content data 1 and content data 2, at this time, content data 2 is determined as the target filtered content clustering cluster, that is, content data 1 is filtered out from the intermediate content clustering cluster.

[0169] Specifically, a feasible implementation process of filtering the target content clustering cluster to obtain the target filtered content clustering cluster according to the vector similarity can be: according to the vector similarity, the content data in the target content clustering cluster is sorted, the content data corresponding to the large vector similarity is arranged in front, the sorted target content clustering cluster is obtained, and then from the head of the sorted target content clustering cluster, the content data corresponding to the target percentage is reserved. Wherein, the target percentage can be set based on actual conditions, for example, the target percentage can be 33%.

[0170] S203. Perform merging and filtering on the X clusters of filtered content to obtain the third filtering sequence.

[0171] Specifically, a feasible implementation process for merging and filtering X clusters of filtered content to obtain a third filtering sequence can be as follows: First, merge the X clusters of filtered content to obtain a total cluster of filtered content. Then, sort the content data in the total cluster of filtered content according to the interaction time of the interactive object, prioritizing content data with interaction times closer to the current time. This results in a sorted total cluster of filtered content. Next, starting from the beginning, select the content data corresponding to the target sequence value as the third filtering sequence. The target sequence value can be determined based on actual conditions; for example, the target sequence value could be 300.

[0172] S204. Sort the recommended data in the recommendation sequence according to the third filtering sequence to obtain sorted recommended data, and recommend the sorted recommended data to the interactive object.

[0173] Specifically, the implementation process of step S204 can be referred to the detailed description of step S203 above, and will not be repeated here.

[0174] The method provided in this application embodiment addresses the second screening sequence, which has a long sequence length due to increased content data diversity. By performing interest clustering on the second screening sequence and filtering out behavioral noise, the sequence length can be compressed. This reduces the computational burden of subsequent processing while ensuring that the obtained third screening sequence does not affect subsequent prediction of user interests.

[0175] Furthermore, to better understand the data processing method provided in the embodiments of this application, a note-taking example based on content data will be used for explanation. Please refer to... Figure 4 , Figure 4 This is a schematic diagram illustrating the overall process of note sequence filtering provided in an embodiment of this application. Figure 4 As shown, in order to obtain relevant sequences that can be used to predict the interest of interactive objects in target recommended notes, the original interaction sequence with a length of tens of thousands can be filtered three times to obtain the final third filtered sequence with a length of 300.

[0176] like Figure 4 As shown, in the first filtering process, the similarity between the query data and the original interaction sequence can be calculated in four dimensions: query data-note dimension, query data-title dimension, query data-image dimension, and query data-text dimension. The similarity scores in these four dimensions are then used to filter the original interaction sequence.

[0177] For the query data-note dimension, the cosine similarity between the query vector output by the multimodal vector encoding model and the content vector corresponding to the note can be calculated to obtain the query data-note similarity, and the notes corresponding to the top 1k (top 1000) similarity scores can be retained.

[0178] For the query data-title dimension, the query vector output by the multimodal vector encoding model and the title vector corresponding to the note can be used to calculate the cosine similarity to obtain the query data-title similarity, and the notes corresponding to the top 300 similarity scores can be retained.

[0179] For the query data-image dimension, the query vector output by the multimodal vector encoding model can be used to calculate the cosine similarity between the query vector and the image vector corresponding to the note to obtain the query data-image similarity, and the notes corresponding to the top 300 similarity scores can be retained.

[0180] For the query data-text dimension, the query vector output by the multimodal vector encoding model and the text vector corresponding to the note can be used to calculate the cosine similarity to obtain the query data-text similarity, and the notes corresponding to the top 400 (top 4000) similarity scores can be retained.

[0181] It is understandable that after filtering through the above four dimensions, the first filtering can yield a first filtered sequence of length 2000. Then, a second filtering can be performed on the first filtered sequence.

[0182] like Figure 4 As shown, in the second screening process, the similarity between the recommended notes and the first screening sequence can be calculated in three dimensions: recommended note-to-note, recommended note-to-image, and recommended note-to-text. The similarity scores in these three dimensions are then used to further filter the first screening sequence.

[0183] For the recommended note-to-note dimension, the cosine similarity between the content vectors corresponding to the recommended notes and the content vectors corresponding to the notes can be calculated using the multimodal vector encoding model to obtain the recommended note-to-note similarity, and the notes corresponding to the top 400 similarity scores can be retained.

[0184] For the recommended notes image-to-image dimension, the cosine similarity between the image vectors corresponding to the recommended notes output by the multimodal vector encoding model and the image vectors corresponding to the notes can be calculated to obtain the recommended notes image-to-image similarity, and the notes corresponding to the top 300 similarity scores can be retained.

[0185] For the recommended note text-text dimension, the cosine similarity between the text vector corresponding to the recommended note output by the multi-modal vector encoding model and the text vector corresponding to the note can be calculated to obtain the recommended note text-text similarity, and the top 300 (top 300) similarity corresponding notes are retained.

[0186] It can be understood that after the above three dimensions of screening, the second screening sequence with a length of 1000 can be obtained. Then, the second screening sequence can be subjected to third screening. The third screening can also be understood as sequence compression. Since the previous two screening processes have screened the sequence in multiple dimensions, the length of the obtained sequence has improved diversity and relevance, but compared with single-dimensional screening, the length of the sequence is longer. In order to balance the pressure on the calculation resources of the subsequent modules after the length of the sequence is lengthened, the second screening sequence can be subjected to sequence compression to compress its length to the length of the sequence corresponding to the single-dimensional screening.

[0187] As shown in Figure 4 , the third screening mainly includes the following three steps: interest clustering, time sequence constraint and gating merging.

[0188] Among them, the interest clustering refers to using the DBSACN algorithm to cluster the second screening sequence, thereby obtaining a plurality of clustering clusters.

[0189] Among them, the time sequence constraint refers to defining a time decay weight for the notes in each clustering cluster, and the definition process can refer to the above formula (1) to realize the quantitative measurement that the recent behavior is more important than the historical behavior. In addition, for all notes in a clustering cluster, the cluster center vector with time sequence constraint is redefined, and the definition process can refer to the above formula (2).

[0190] Among them, the gating merging refers to filtering and merging the notes in all clustering clusters according to the following gating rules: first, for the clustering cluster with only 1 note, if the time sequence weight w<0.01, it is considered as a noise point and is filtered; second, for the clustering cluster with 2 notes, if the time sequence weight w<0.01, it is considered as a noise point and is filtered; if there are >=1 notes after filtering, the note with the more recent time is retained and the others are filtered; then, for the clustering cluster with >2 notes, the cosine similarity distance between each note and the cluster center with time sequence constraint is calculated and sorted, and the top 33% of notes are retained and the others are filtered; finally, the remaining sequence is arranged in time sequence, and if the remaining sequence length exceeds 300, the top 300 is taken for truncation to obtain the third screening sequence.

[0191] The method provided by the embodiment of the present application expands the sample (such as a note) screening mechanism, introduces more fine-grained cross-modal vectors for merging and screening, effectively enriches the diversity of the sequence, and at the same time, in order to reduce the introduced behavior noise, adopts gated temporal perception merging to aggregate and compress the sequence length to the length before expansion, does not introduce more calculation resources for subsequent modules, and ensures the real-time calculation efficiency on line.

[0192] Those skilled in the art can understand that, in the above method of the specific embodiment, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process, and the specific execution order of each step should be determined according to its function and possible inherent logic.

[0193] The above describes the method of the embodiment of the present application in detail, and the device of the embodiment of the present application is provided below.

[0194] Please refer to Figure 5 , Figure 5 The structure diagram of a content recommendation device provided by the embodiment of the present application is shown. The content recommendation device 1 comprises a first splitting module 11, a first screening module 12, a second splitting module 13, a second screening module 14 and a sorting module 15, and specifically:

[0195] The first splitting module 11 is configured to perform first-granularity content splitting on content data in an original interaction sequence corresponding to an interaction object, to obtain S first-granularity content data in the content data in the original interaction sequence; the S is an integer greater than 1.

[0196] The first screening module 12 is configured to obtain a first similarity between query data and the content data in the original interaction sequence, obtain a second similarity between the query data and the S first-granularity content data respectively, and perform content screening on the original interaction sequence according to the first similarity and the S second similarities, to obtain a first screening sequence.

[0197] The second splitting module 13 is configured to perform second-granularity content splitting on recommendation data in a recommendation sequence associated with the query data, to obtain R second-granularity content data in the recommendation data in the recommendation sequence; the R is an integer greater than 1.

[0198] The second screening module 14 is configured to obtain a third similarity between the recommendation data in the recommendation sequence and the content data in the first screening sequence, and perform content screening on the first screening sequence according to the third similarity, the R second-granularity content data in the recommendation data in the recommendation sequence and granularity content data in the first screening sequence having the same granularity type as the R second-granularity content data, to obtain a second screening sequence.

[0199] The sorting module 15 is configured to sort the recommendation data in the recommendation sequence according to the second screening sequence, to obtain sorted recommendation data, and to recommend the sorted recommendation data to the interactive object.

[0200] According to any of the embodiments of the present application, the number of content data in the original interactive sequence is N; N is an integer greater than 1;

[0201] The first similarity between the query data and the content data in the original interactive sequence is obtained, the second similarity between the query data and the S first-granularity content data in the original interactive sequence is obtained, the original interactive sequence is content-screened according to the first similarity and the S second similarities, and a first screening sequence is obtained.

[0202] The N content data in the original interactive sequence is traversed, and the kth content data is obtained; k is a positive integer less than or equal to N;

[0203] The vector cosine similarity between the query data and the kth content data is determined as the first similarity corresponding to the kth content data;

[0204] The vector cosine similarity between the query data and the S first-granularity content data in the kth content data is determined as the S second similarities corresponding to the kth content data;

[0205] When the N content data in the original interactive sequence is traversed, the original interactive sequence is content-screened according to the first similarity and the S second similarities corresponding to each content data, and a first screening sequence is obtained.

[0206] According to any of the embodiments of the present application, the original interactive sequence is content-screened according to the first similarity and the S second similarities corresponding to each content data, and a first screening sequence is obtained.

[0207] The N content data is first sorted according to the first similarity corresponding to each content data, and a first sorting sequence is obtained, and the first M content data in the first sorting sequence is determined as a first screening sub-sequence; M is a preset screening parameter.

[0208] The S second similarities corresponding to each content data are traversed, and the jth second similarity corresponding to each content data is obtained; j is a positive integer less than or equal to S;

[0209] According to the jth second similarity corresponding to each content data, the N content data are subjected to second sorting processing, a second sorting sequence is obtained, and the first M content data in the second sorting sequence are determined as a jth second screening subsequence;

[0210] When the S second similarities corresponding to each content data are traversed, the first screening subsequence and the S second screening subsequences are subjected to merging processing, and a first screening sequence is obtained.

[0211] According to any one of the embodiments of the present application, the first similarity between the query data and the content data in the original interaction sequence is obtained, the second similarity between the query data and the S first granularity content data is obtained, the original interaction sequence is subjected to content screening according to the first similarity and the S second similarities, and a first screening sequence is obtained. The first screening module 12 is specifically used for:

[0212] The query vector corresponding to the query data is obtained, the content vector corresponding to the content data in the original interaction sequence and the granularity content vector corresponding to the S first granularity content data are obtained;

[0213] The cosine similarity between the query vector and the content vector is calculated to obtain the first similarity between the query data and the content data in the original interaction sequence;

[0214] The cosine similarity between the query vector and the S granularity content vectors is calculated to obtain the second similarity between the query data and the S first granularity content data;

[0215] The original interaction sequence is subjected to content screening according to the first similarity and the S second similarities, and a first screening sequence is obtained.

[0216] According to any one of the embodiments of the present application, the query vector corresponding to the query data is obtained, the content vector corresponding to the content data in the original interaction sequence and the granularity content vector corresponding to the S first granularity content data are obtained, and the first screening module 12 is specifically used for:

[0217] The query data is subjected to single granularity encoding processing based on a multi-modal vector encoding model to obtain a query vector corresponding to the query data;

[0218] obtaining global identification data corresponding to the content data in the original interaction sequence, and performing splicing processing on the S first granularity content data, the S granularity content separators, and the global identification data to obtain multi-granularity content data corresponding to the content data in the original interaction sequence; a first granularity content data in the multi-granularity content data is connected with a granularity content separator at the end, and the global identification data is connected at the end of the last granularity content separator;

[0219] performing multi-granularity encoding processing on the multi-granularity content data based on the multi-modal vector encoding model to obtain a global vector corresponding to the global identification data and granularity content vectors corresponding to the S first granularity content data respectively, and determining the global vector as a content vector corresponding to the content data in the original interaction sequence.

[0220] According to any one of the embodiments of the present application, the number of recommendation data in the recommendation sequence is H, and the H recommendation data includes recommendation data P i , H is a positive integer, and i is a positive integer less than or equal to H; the number of the second screening sequence is H, and one second screening sequence corresponds to one recommendation data in the recommendation sequence; the number of content data in the first screening sequence is L, and the L is an integer greater than 1;

[0221] The third similarity between the recommendation data in the recommendation sequence and the content data in the first screening sequence is obtained, and the first screening sequence is content-screened according to the third similarity, R second granularity content data in the recommendation data in the recommendation sequence, and granularity content data in the first screening sequence having the same granularity type as the R second granularity content data, to obtain a second screening sequence, and the second screening module 14 is specifically configured to:

[0222] traverse the L content data in the first screening sequence to obtain the jth content data; j is a positive integer less than or equal to L;

[0223] determine the vector cosine similarity between the recommendation data P i and the jth content data as the third similarity corresponding to the jth content data;

[0224] generate R granularity content data pairs according to the R second granularity content data in the recommendation data P i and the jth content data; one granularity content data pair includes one second granularity content data in the recommendation data P i and granularity content data in the jth content data having the same granularity type as the one second granularity content data;

[0225] The R pieces of granular content data are respectively corresponding to a vector cosine similarity, and the fourth similarity corresponding to the jth content data is determined;

[0226] When the L pieces of content data in the first screening sequence are traversed, the first screening sequence is screened according to the third similarity and the fourth similarity corresponding to the L pieces of content data, and the recommended data P is obtained i The second screening sequence corresponds.

[0227] According to any one of the embodiments of the present application, the number of the second screening sequences is the same as the number of the recommended data in the recommended sequence; one second screening sequence corresponds to one recommended data in the recommended sequence;

[0228] The sorting module 15 is specifically configured to:

[0229] According to each second screening sequence, each recommended data in the recommended sequence is subjected to object interaction prediction processing, and the object interaction interest degree corresponding to each recommended data is obtained;

[0230] According to the object interaction interest degree corresponding to each recommended data, the recommended data in the recommended sequence is sorted, and the sorted recommended data is obtained; in the sorted recommended data, the recommended data with a higher position corresponds to a greater object interaction interest degree;

[0231] The sorted recommended data is recommended to the interactive object.

[0232] According to any one of the embodiments of the present application, the sorting module 15 is specifically configured to:

[0233] The second screening sequence is subjected to content clustering processing, and X content clustering clusters are obtained; one content clustering cluster contains one or more content data in the second screening sequence; X is an integer greater than or equal to 1;

[0234] The X content clustering clusters are respectively subjected to content data filtering processing, and X filtered content clustering clusters are obtained;

[0235] The X filtered content clustering clusters are subjected to merging and screening processing, and a third screening sequence is obtained;

[0236] According to the third screening sequence, the recommended data in the recommended sequence is sorted to obtain sorted recommended data, and the sorted recommended data is recommended to the interactive object.

[0237] According to any of the embodiments of the present application, the X content cluster groups include a target content cluster group;

[0238] The X content cluster groups are respectively subjected to content data filtering processing to obtain X filtered content cluster groups, and the clustering module is specifically configured to:

[0239] Determine a time decay weight corresponding to each content data in the target content cluster group;

[0240] If the number of content data in the target content cluster group is less than or equal to a first number threshold, the target content cluster group is subjected to filtering processing according to the time decay weight corresponding to each content data in the target content cluster group to obtain a target filtered content cluster group.

[0241] If the number of content data in the target content cluster group is greater than the first number threshold, a cluster center vector corresponding to the target content cluster group is determined according to the time decay weight corresponding to each content data included in the target content cluster group and the content vector corresponding to each content data included in the target content cluster group.

[0242] Determine a vector similarity between the cluster center vector and the content vector of each content data included in the target content cluster group, and the target content cluster group is subjected to filtering processing according to the vector similarity to obtain a target filtered content cluster group.

[0243] According to any of the embodiments of the present application, the target content cluster group is subjected to filtering processing according to the time decay weight corresponding to each content data in the target content cluster group to obtain a target filtered content cluster group, and the clustering module is specifically configured to:

[0244] From the target content cluster group, content data with a time decay weight less than a time decay weight threshold is deleted to obtain an intermediate content cluster group.

[0245] If the intermediate content cluster group is not empty, the content data with the largest time decay weight in the intermediate content cluster group is determined as the target filtered content cluster group; if the intermediate content cluster group is empty, the empty cluster is determined as the target filtered content cluster group.

[0246] In the embodiments of the present application, the data processing apparatus can perform first granularity content splitting on the content data in the original interaction sequence corresponding to the interaction object, obtain S first granularity content data in the content data in the original interaction sequence, then obtain a first similarity between the query data and the content data in the original interaction sequence, obtain a second similarity between the query data and the S first granularity content data respectively, perform content screening on the original interaction sequence according to the first similarity and the S second similarities, and obtain a first screened sequence; then perform granularity content splitting on the recommendation data in the recommendation sequence associated with the query data, obtain R second granularity content data in the recommendation data in the recommendation sequence, then obtain a third similarity between the recommendation data in the recommendation sequence and the content data in the first screened sequence, and perform content screening on the first screened sequence according to the third similarity, the R second granularity content data in the recommendation data in the recommendation sequence, and the granularity content data in the first screened sequence having the same granularity type as the R second granularity content data, and obtain a second screened sequence; finally, sort the recommendation data in the recommendation sequence according to the second screened sequence, obtain sorted recommendation data, and recommend the sorted recommendation data to the interaction object. Through the method provided in the embodiments of the present application, the dimension of sequence screening is expanded by introducing S first granularity content data and R second granularity content data, effectively solving the problem that the characteristics of local data are difficult to be reflected by the overall representation of content data, and when calculating the similarity corresponding to different granularity content data, the corresponding modal type can be flexibly processed, effectively solving the problem that when the content data is represented as a whole, the focus is on a certain modal type data, and the cross-modal information is not well understood, thereby reducing the omission of truly relevant content data, and effectively improving the relevance of the screened sequence and the accuracy of sorting the recommendation data.

[0247] Figure 6 A hardware structure schematic diagram of an electronic device is provided in the embodiments of the present application. The electronic device 4 includes a processor 41 and a memory 42. Optionally, the electronic device 4 further includes an input device 43 and an output device 44. The processor 41, the memory 42, the input device 43 and the output device 44 are coupled through a connector, which includes various interfaces, transmission lines or buses, etc., and the embodiments of the present application do not make any limitation on this. It should be understood that in various embodiments of the present application, coupling means mutual contact in a specific way, including direct connection or indirect connection through other devices, for example, connection through various interfaces, transmission lines, buses, etc.

[0248] The processor 41 can include one or more processors, for example, including one or more central processing units (CPUs), which can be single core CPUs or multi-core CPUs in the case of a CPU. Alternatively, the processor 41 can be a processor group composed of multiple CPUs, and the multiple processors are coupled to each other through one or more buses. Alternatively, the processor can also be other types of processors, etc., and the embodiments of the present application are not limited thereto.

[0249] The memory 42 can be used to store computer program instructions, and various types of computer program codes for executing the scheme of the present application. Alternatively, the memory includes, but is not limited to, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read only memory (EPROM), or a compact disc read-only memory (CD-ROM), which is used for related instructions and data.

[0250] The input device 43 is used to input data and / or signals, and the output device 44 is used to output data and / or signals. The input device 43 and the output device 44 can be independent devices, or can be an integral device.

[0251] It can be understood that, in the embodiments of the present application, the memory 42 can be used not only to store related instructions, but also to store related data, and the embodiments of the present application do not limit the data stored in the memory.

[0252] It can be understood that, Figure 6 Only a simplified design of an electronic device is shown. In actual applications, the electronic device can also include other necessary elements, including but not limited to any number of input / output devices, processors, memories, etc., and all electronic devices that can implement the embodiments of the present application are within the protection scope of the present application.

[0253] Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0254] Those skilled in the art can clearly understand the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments for the convenience and brevity of description, which will not be repeated here. Those skilled in the art can also clearly understand that each embodiment of the present application describes each focus, and for the convenience and brevity of description, the same or similar parts can not be described in different embodiments. Therefore, the parts not described or not described in detail in an embodiment can be referred to the description of other embodiments.

[0255] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0256] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0257] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0258] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable apparatus. The computer instructions can be stored in or transmitted by a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through a wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that includes one or more available media sets. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a digital versatile disc (DVD)), or a semiconductor medium (such as a solid state disk (SSD)), etc.

[0259] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments can be instructed by a computer program to complete the relevant hardware, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-mentioned method embodiments. The aforementioned storage medium includes a read-only memory (ROM) or a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

Claims

1. A data processing method, characterized in that, include: The content data in the original interaction sequence corresponding to the interaction object is split into first-granularity content data to obtain S first-granularity content data in the content data of the original interaction sequence; S is an integer greater than 1; Obtain the first similarity between the query data and the content data in the original interaction sequence, obtain the second similarity between the query data and the S first-granularity content data respectively, and filter the content of the original interaction sequence according to the first similarity and the S second similarities to obtain the first filtered sequence; The recommended data in the recommendation sequence associated with the query data is split into second-granularity content data to obtain R second-granularity content data in the recommended data of the recommendation sequence; where R is an integer greater than 1. Obtain the third similarity between the recommended data in the recommended sequence and the content data in the first filtering sequence. Based on the third similarity, R second-granularity content data in the recommended data of the recommended sequence and the granularity content data in the first filtering sequence that has the same granularity type as the R second-granularity content data, perform content filtering on the first filtering sequence to obtain the second filtering sequence. The recommended data in the recommended sequence is sorted according to the second filtering sequence to obtain sorted recommended data, and the sorted recommended data is recommended to the interactive object.

2. The method according to claim 1, characterized in that, The number of content data in the original interaction sequence is N; where N is an integer greater than 1. The process of obtaining a first similarity between the query data and the content data in the original interaction sequence, obtaining a second similarity between the query data and the S first-granularity content data, and filtering the original interaction sequence based on the first similarity and the S second similarities to obtain a first filtering sequence includes: Iterate through the N content data in the original interaction sequence to obtain the k-th content data; k is a positive integer less than or equal to N; The vector cosine similarity between the query data and the kth content data is determined as the first similarity corresponding to the kth content data. The vector cosine similarity between the query data and the S first-granularity content data in the k-th content data is determined as the S second similarity corresponding to the k-th content data. When all N content data in the original interaction sequence have been traversed, the original interaction sequence is filtered based on the first similarity and S second similarities corresponding to each content data to obtain the first filtered sequence.

3. The method according to claim 2, characterized in that, The step of filtering the original interaction sequence based on the first similarity and S second similarities corresponding to each content data to obtain a first filtered sequence includes: The N content data are sorted according to the first similarity corresponding to each content data to obtain a first sorting sequence. The first M content data in the first sorting sequence are determined as the first filtering subsequence; M is a preset filtering parameter. Iterate through the S second similarities corresponding to each content data, and obtain the j-th second similarity corresponding to each content data; j is a positive integer less than or equal to S; The N content data are sorted according to the j-th second similarity corresponding to each content data to obtain a second sorting sequence. The first M content data in the second sorting sequence are determined as the j-th second filtering subsequence. When all S second similarities corresponding to each content data have been traversed, the first filtering subsequence and the S second filtering subsequences are merged to obtain the first filtering sequence.

4. The method according to claim 1, characterized in that, The process of obtaining a first similarity between the query data and the content data in the original interaction sequence, obtaining a second similarity between the query data and the S first-granularity content data, and filtering the original interaction sequence based on the first similarity and the S second similarities to obtain a first filtering sequence includes: Obtain the query vector corresponding to the query data, obtain the content vector corresponding to the content data in the original interaction sequence, and obtain the granular content vectors corresponding to the S first granularity content data respectively; The cosine similarity between the query vector and the content vector is calculated to obtain the first similarity between the query data and the content data in the original interaction sequence. The cosine similarity is calculated between the query vector and the S granularity content vectors respectively to obtain the second similarity between the query data and the S first granularity content data respectively; The original interaction sequence is filtered based on the first similarity and S second similarities to obtain a first filtered sequence.

5. The method according to claim 4, characterized in that, The step of obtaining the query vector corresponding to the query data, obtaining the content vector corresponding to the content data in the original interaction sequence, and obtaining the granular content vectors corresponding to the S first-granularity content data respectively include: The query data is processed using a multimodal vector coding model to obtain a query vector corresponding to the query data. Obtain the global identifier data corresponding to the content data in the original interaction sequence, and concatenate the S first-granularity content data, S granularity content separators, and the global identifier data to obtain multi-granularity content data corresponding to the content data in the original interaction sequence; a granularity content separator is connected to the end of one of the first-granularity content data in the multi-granularity content data, and the global identifier data is connected to the end of the last granularity content separator; Based on the multimodal vector coding model, the multi-granularity content data is processed by multi-granularity coding to obtain the global vector corresponding to the global identifier data and the granular content vectors corresponding to the S first granularity content data respectively. The global vector is then determined as the content vector corresponding to the content data in the original interaction sequence.

6. The method according to claim 1, characterized in that, The number of recommended data in the recommended sequence is H, and the H recommended data contain recommended data P. i H is a positive integer, and i is a positive integer less than or equal to H; the number of the second screening sequences is H, and one second screening sequence corresponds to one recommended data in the recommended sequence; The number of content data in the first filtering sequence is L, where L is an integer greater than 1; The step of obtaining a third similarity between the recommended data in the recommended sequence and the content data in the first filtering sequence, and then filtering the first filtering sequence based on the third similarity, R second-granularity content data in the recommended data of the recommended sequence, and granular content data in the first filtering sequence that has the same granularity type as the R second-granularity content data, to obtain a second filtering sequence, includes: Iterate through the L content data in the first filtering sequence to obtain the j-th content data; j is a positive integer less than or equal to L; The recommended data P i The vector cosine similarity between the j-th content data and the j-th content data is determined as the third similarity corresponding to the j-th content data; Based on the recommended data P i R second-granularity content data and the j-th content data are used to generate R granularity content data pairs; each granularity content data pair contains the recommendation data P. i The second granularity content data in the j-th content data and the granular content data in the j-th content data that has the same granularity type as the second granularity content data; The cosine similarity of the vectors corresponding to the R granularity content data pairs is determined as the fourth similarity corresponding to the j-th content data. When all L content data points in the first filtering sequence have been traversed, the first filtering sequence is further filtered based on the third and fourth similarities corresponding to the L content data points, resulting in the recommended data P. i The corresponding second screening sequence.

7. The method according to claim 1, characterized in that, The number of the second filtering sequences is the same as the number of recommended data in the recommended sequences; one second filtering sequence corresponds to one recommended data in the recommended sequences. The step of sorting the recommended data in the recommended sequence according to the second filtering sequence to obtain sorted recommended data, and recommending the sorted recommended data to the interactive object, includes: Each recommended data in the recommendation sequence is subjected to object interaction prediction processing according to each second screening sequence to obtain the object interaction interest degree corresponding to each recommended data. The recommended data in the recommendation sequence are sorted according to the object interaction interest level corresponding to each recommended data to obtain the sorted recommended data; in the sorted recommended data, the earlier the recommended data is, the greater the object interaction interest level. The sorted recommendation data is then recommended to the interactive object.

8. The method according to claim 1, characterized in that, The step of sorting the recommended data in the recommended sequence according to the second filtering sequence to obtain sorted recommended data, and recommending the sorted recommended data to the interactive object, includes: The second filtering sequence is subjected to content clustering processing to obtain X content clusters; each content cluster contains one or more content data from the second filtering sequence; X is an integer greater than or equal to 1. The content data is filtered for each of the X content clusters to obtain X filtered content clusters; The X clusters of filtered content are merged and filtered to obtain the third filtering sequence; The recommended data in the recommended sequence is sorted according to the third filtering sequence to obtain sorted recommended data, and the sorted recommended data is recommended to the interactive object.

9. The method according to claim 8, characterized in that, The X content clusters include the target content cluster; The process of filtering content data on each of the X content clusters to obtain X filtered content clusters includes: Determine the time decay weight corresponding to each content data in the target content cluster; If the number of content data in the target content cluster is less than or equal to the first quantity threshold, then the target content cluster is filtered according to the time decay weight corresponding to the content data in the target content cluster to obtain the target filtered content cluster. If the number of content data in the target content cluster is greater than the first quantity threshold, then the cluster center vector corresponding to the target content cluster is determined according to the time decay weight corresponding to each content data in the target content cluster and the content vector corresponding to each content data in the target content cluster. Determine the vector similarity between the cluster center vector and the content vector of each content data contained in the target content cluster, and perform filtering processing on the target content cluster based on the vector similarity to obtain the target filtered content cluster.

10. The method according to claim 9, characterized in that, The step of filtering the target content clusters based on the time decay weights corresponding to the content data in the target content clusters to obtain target filtered content clusters includes: From the target content cluster, delete the content data whose time decay weight is less than the time decay weight threshold to obtain the intermediate content cluster; If the intermediate content cluster is not an empty cluster, then the content data with the largest time decay weight in the intermediate content cluster is determined as the target filtered content cluster. If the intermediate content cluster is an empty cluster, then the empty cluster is determined as the target filtered content cluster.

11. A data processing apparatus, characterized in that, include: The first splitting module is used to perform first-granularity content splitting on the content data in the original interaction sequence corresponding to the interaction object, to obtain S first-granularity content data in the content data of the original interaction sequence. S is an integer greater than 1; The first filtering module is used to obtain a first similarity between the query data and the content data in the original interaction sequence, obtain a second similarity between the query data and the S first-granularity content data respectively, and filter the content of the original interaction sequence according to the first similarity and the S second similarities to obtain a first filtering sequence. The second splitting module is used to perform second-granularity content splitting on the recommended data in the recommended sequence associated with the query data, to obtain R second-granularity content data in the recommended data of the recommended sequence; where R is an integer greater than 1. The second filtering module is used to obtain a third similarity between the recommended data in the recommended sequence and the content data in the first filtering sequence, and to perform content filtering on the first filtering sequence based on the third similarity, R second-granularity content data in the recommended data in the recommended sequence and granular content data in the first filtering sequence that have the same granularity type as the R second-granularity content data, to obtain the second filtering sequence. The sorting module is used to sort the recommended data in the recommended sequence according to the second filtering sequence, obtain the sorted recommended data, and recommend the sorted recommended data to the interactive object.

12. An electronic device, characterized in that, include: A processor and a memory, the memory being used to store computer program code, the computer program code including computer instructions, wherein, when the processor executes the computer instructions, the electronic device performs the method as described in any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 10.

14. A computer program product, characterized in that, The computer program product includes a computer program or instructions; when the computer program or instructions are executed on a computer, the computer causes the computer to perform the method according to any one of claims 1 to 10.

Images