Recommendation apparatus, storage medium, and computer program product
By extending the similarity between the vector of the target content and the vectors of multiple contents accessed by the first user, multiple similar contents are determined and recommended, which solves the problem of single recommended content in the existing technology, realizes efficient personalized recommendation, and improves Improve user stickiness and recommendation effect.
Patent Information
- Application Number
- CN202310886564.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-07-18
- Publication Date
- 2025-01-21
AI Technical Summary
In the prior art, the single recommendation content is caused by poor recommendation results. Especially in the cold start stage of new users and new content, it is difficult to achieve efficient and personalized recommendations.
By performing similarity expansion between the vector of the target content and the vectors of multiple contents accessed by the first user, multiple similar contents are determined and recommended as recommended contents.
It effectively improves the richness of recommended content, enhances user stickiness, and improves recommendation effects.
Smart Images
Figure CN119336975A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a recommendation device, a storage medium and a computer program product. Background Art
[0002] The CS (Cold Start) problem in the recall phase of the APP recommendation system (i.e., the phase of retrieving content that users are potentially interested in from a massive content library based on some of their characteristics) is becoming increasingly prominent. As new users (users) and new content (items) continue to arrive, if there is no efficient CS module in the recommendation system, new users and new content will not receive efficient personalized recommendations, which will greatly reduce user stickiness and content richness, leading to a decrease in the number of users and revenue.
[0003] In existing CS solutions, new content is usually recommended to all users when it arrives. This method recommends a single piece of content, which will inevitably reduce user stickiness and lead to poor recommendation results. Summary of the invention
[0004] The present invention provides a recommendation device storage medium and a computer program product, which are used to solve the defect that the recommendation content in the prior art is single, resulting in poor recommendation effect.
[0005] In a first aspect, the present invention provides a recommendation device, comprising: a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the following steps:
[0006] Determine a content feature sequence according to the order in which the first user accesses the first contents and the attribute features of the first contents, and determine a vector of each feature in the content feature sequence;
[0007] Determining a target vector of the target content according to the attribute characteristics of the target content and the vectors of the characteristics;
[0008] Determine a first vector of each first content according to the attribute characteristics of each first content, and determine a plurality of similar contents according to the similarities between the target vector and the first vector of each first content;
[0009] The multiple similar contents are recommended as recommended contents.
[0010] In one embodiment, the processor is specifically configured to:
[0011] A second user is determined according to the plurality of similar contents, and at least one of the plurality of similar contents is recommended to the second user, wherein the plurality of similar contents includes the target content.
[0012] In one embodiment, the processor is further configured to filter the second user:
[0013] Acquire multiple access contents of the first candidate user;
[0014] Clustering the plurality of access contents according to a preset clustering algorithm to generate target attribute features of the first candidate user;
[0015] Performing similarity matching between the target attribute feature and the features of the plurality of similar contents;
[0016] When the sum of the similarities between the target attribute feature and the features of the plurality of similar contents is greater than or equal to a similarity threshold, the first candidate user is determined to be the second user.
[0017] In one embodiment, the processor is further configured to filter the second user:
[0018] Obtaining user behavior of the second candidate user;
[0019] According to a preset screening strategy, whether matching the user behavior with at least one of the plurality of similar contents reflects a strong behavior;
[0020] When the user behavior and at least one of the plurality of similar contents reflect a strong behavior, the second candidate user is determined to be the second user.
[0021] In one embodiment, the processor is further configured to: filter the first user by any of the following methods:
[0022] taking a user who has visited the target content as the first user;
[0023] A user who has performed operations on the target content on a network platform is regarded as the first user.
[0024] In one embodiment, the processor is specifically configured to implement any of the following:
[0025] The first content whose similarity between the target vector and the first vector exceeds a similarity threshold, and the target content, are used as the multiple similar contents;
[0026] The first content having the greatest similarity between the target vector and the first vector and the target content are used as the multiple similar contents.
[0027] In one embodiment, the mandatory behavior includes at least one of the following:
[0028] The time spent browsing content exceeds the time threshold;
[0029] The number of times the content is accessed exceeds the threshold;
[0030] Like content;
[0031] Collection content;
[0032] Purchase content.
[0033] In one embodiment, the processor is specifically configured to implement any of the following:
[0034] Determine the cosine similarity between the target vector and the first vector of each first content;
[0035] Determine the distance similarity between the target vector and the first vector of each first content.
[0036] In one embodiment, the processor is specifically configured to:
[0037] Establishing a vector index of each feature in the content feature sequence;
[0038] Determining, according to the attribute features of the target content and the vector index, a vector corresponding to each attribute feature of the target content;
[0039] The vectors corresponding to the attribute features of the target content are combined to obtain a target vector of the target content.
[0040] In a second aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed, implements the method steps executed by the processor as described in the first aspect.
[0041] In a third aspect, the present invention further provides a computer program product, comprising a computer program, which, when executed, implements the method steps executed by the processor as described in the first aspect.
[0042] The recommendation device, storage medium and computer program product provided by the embodiments of the present invention can effectively improve the richness of the recommended content by performing similarity expansion between the vector of the target content and the vector of multiple contents accessed by the first user, and recommend the multiple similar contents, thereby improving the stickiness of users and enhancing the recommendation effect. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0044] Figure 1 is a flowchart of the steps of the method performed by the recommendation device provided in an embodiment of the present invention;
[0045] Figure 2 is a schematic diagram of the structure of a recommendation system using the recommendation device provided by an embodiment of the present invention;
[0046] Figure 3 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0047] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0048] Figure 1 is a flow chart of the steps of the method performed by the recommendation device provided in an embodiment of the present invention. Figure 1 The recommendation device provided in the embodiment of the present invention may include a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the following steps:
[0049] Step 110: Determine a content feature sequence according to the order in which the first user accesses the first contents and the attribute features of the first contents, and determine a vector of each feature in the content feature sequence;
[0050] Step 120: determining a target vector of the target content according to the attribute features of the target content and the vectors of each feature;
[0051] Step 130: determining a first vector of each first content according to the attribute characteristics of each first content, and determining a plurality of similar contents according to the similarities between the target vector and the first vector of each first content;
[0052] Step 140: recommend multiple similar contents as recommended contents.
[0053] It should be noted that the recommendation device provided in the embodiment of the present invention may specifically be a computer device, such as a mobile phone, a tablet computer, a laptop computer, a PDA, etc.
[0054] Before executing step 110, the recommendation device may use a computing engine such as Spark to clean the attribute feature data of each first content accessed by the first user from a data warehouse such as Hive. Alternatively, the recommendation device may directly call the attribute feature data of each first content accessed by the first user stored in the database through an interface.
[0055] The embodiment of the present invention does not specifically limit the specific manner of obtaining the attribute feature of the first content.
[0056] In step 110 , the recommendation device may generate a content feature sequence by using the order in which the first user accesses each first content and the attribute features of each first content.
[0057] The first user may be any user. The first content may be any content, such as various commodities, books, videos, music, etc.
[0058] For example, the recommendation device can use a computing engine such as Spark to generate a content feature sequence by calculating the order in which user A visits content a, content b, content c, ..., content n, etc., and the attribute characteristics of each content an, such as id, publisher, publishing time, number of pictures, video length, topic, category, at least one of the type tags, etc.
[0059] The type label can be a classification label of the content. For example, when the content is an online novel, the type label can be "fantasy", "martial arts", "cool writing", "romance", etc.; when the content is a video work, the type label can be "science fiction", "secondary", "information", "game", etc.
[0060] After determining the content feature sequence, the recommendation device can determine the vector of each feature in the content feature sequence. For example, the recommendation device can input the content feature sequence into an algorithm implementation tool such as Tensorflow, process the content feature sequence using a model with a text-conversion function such as Word2vec, obtain the vector of each feature in the content feature sequence, and cache the processing result in a database such as Redis.
[0061] It can be understood that since the content feature sequence can be obtained based on the same user accessing different contents, and the correlation and similarity between the contents accessed by the same user are relatively strong, the use of the content feature sequence in the embodiment of the present application can help match similar content, thereby improving the recommendation efficiency and recommendation effect.
[0062] In step 120 , the recommendation device may determine a target vector of the target content according to the attribute features of the target content and the vectors of each feature in the content feature sequence obtained in step 110 .
[0063] For example, the recommendation device can first use tools such as Flink to read the attribute features of the target content in the Kafka system, and match the vectors of each feature corresponding to the attribute features of the target content from the content feature sequence according to the attribute features of the target content, and generate the target vector of the target content according to the corresponding vectors of each feature. Subsequently, the recommendation device can cache the target vector in a database such as Redis.
[0064] The target content may be content to be recommended.
[0065] In step 130, the recommendation device may determine the first vector of each first content according to the attribute characteristics of each first content, and determine a plurality of similar contents according to the similarity between the target vector and the first vector of each first content.
[0066] The recommendation device may read the target vector and each first vector stored in a database such as Redis, then calculate the similarity between the target vector and each first vector, and determine a plurality of similar contents from each first content according to the similarity determination result.
[0067] It should be noted that the first vector is determined according to the attribute characteristics of the first content.
[0068] For example, the recommendation device may input the attribute features of the first content into an algorithm implementation tool such as Tensorflow, and use a model with a text-to-volume function such as DSSM to process the attribute features of the first content, thereby obtaining a first vector of the first content. Alternatively, the recommendation device may also match the vectors of each feature corresponding to the attribute features of the first content from the content feature sequence, and generate the first vector of the first content based on the corresponding vectors of each feature.
[0069] In step 140 , the recommendation device may recommend a plurality of similar contents as recommended contents.
[0070] The recommendation device provided in the embodiment of the present invention can effectively improve the richness of the recommended content by performing similarity expansion between the vector of the target content and the vector of multiple contents accessed by the first user to determine multiple similar contents, and recommend the multiple similar contents, thereby improving the stickiness of users and enhancing the recommendation effect.
[0071] In one embodiment, the processor is specifically configured to:
[0072] A second user is determined according to the plurality of similar contents, and at least one of the plurality of similar contents is recommended to the second user, wherein the plurality of similar contents includes the target content.
[0073] It should be noted that the second user may be a user who is interested in at least one of the multiple similar contents.
[0074] For example, the second user may be a user who has shown strong behavior towards at least one of the plurality of similar contents, wherein the strong behavior may refer to behavior showing interest in the content, such as collection behavior, purchase behavior, etc.
[0075] For example, when multiple similar contents are content a, content b, and content c (the target content can be any one of content a, content b, and content c), if the second user shows strong behavior towards content b, the recommendation system will recommend content a and content c to the second user; if the second user shows strong behavior towards both content a and content c, the recommendation system will recommend content b to the second user.
[0076] The second user may also be a user that meets the following condition: a target attribute feature of the user is determined according to the accessed content of the user, and the target attribute feature is similar to features of a plurality of similar contents.
[0077] It is understandable that the target attribute feature can reflect the regularity of the user's interest in the content. Furthermore, when the target attribute feature is similar to the feature of similar content, the user is used as the second user, which can ensure that the second user can be recommended the content that he is interested in, thereby improving the recommendation effect.
[0078] The recommendation device provided by the embodiment of the present invention can make the recommendation of content purposeful by determining a second user who is interested in multiple similar contents based on multiple similar contents and recommending at least one of the multiple similar contents to the second user, thereby improving the recommendation effect.
[0079] In one embodiment, the processor is further configured to screen the second user:
[0080] Acquire multiple access contents of the first candidate user;
[0081] Clustering the multiple access contents according to a preset clustering algorithm to generate target attribute features of the first candidate user;
[0082] Perform similarity matching between target attribute features and features of multiple similar contents;
[0083] When the sum of similarities between the target attribute feature and the features of the plurality of similar contents is greater than or equal to a similarity threshold, the first candidate user is determined to be the second user.
[0084] The recommendation device can obtain the attribute characteristic information of each access content in the history of the first candidate user, or obtain the attribute characteristic information of each access content of the first candidate user within a period of time (for example, within a week or a month, etc.), and cluster the obtained attribute characteristic information of each access content to obtain the target attribute characteristics of the first candidate user.
[0085] For example, for the first candidate user A, the recommendation device can obtain the attribute feature information of each accessed content within a week, and cluster the attribute feature information of each accessed content by the first candidate user A within a week through a preset clustering algorithm, such as a K-means clustering algorithm, a GMM Gaussian mixture model clustering algorithm, etc., to obtain the target attribute feature of the first candidate user A. The target attribute feature can be a clustering result of various attribute features such as the publisher, category, and type label of the content.
[0086] Next, the recommendation device may calculate similarity between the target attribute feature and features of a plurality of similar contents.
[0087] For example, the recommendation device may calculate the similarity between the target attribute feature and the feature of one of the multiple similar contents to obtain a similarity value a1; calculate the similarity between the target attribute feature and the feature of another similar content in the multiple similar contents to obtain a similarity value a2, and so on.
[0088] After obtaining the similarity values a1, a2, ..., a n After that, we can get the sum of similarities M = a1 + a2 + ... + a n ; If M is greater than or equal to the similarity threshold M0, the first candidate user A can be determined as the second user.
[0089] If M is less than the similarity threshold M0, it can be determined that the first candidate user A is not the second user.
[0090] It should be noted that when determining the sum M of the similarities between the target attribute feature and the features of multiple similar contents, different weights can be set according to different similar contents. For example, a weight m1 can be set for the similarity value a1 between the target attribute feature and one of the multiple similar contents, and a weight m2 can be set for the similarity value a2 between the target attribute feature and another of the multiple similar contents, and the sum M of the similarities between the target attribute feature and the features of multiple similar contents can be calculated as M=m1a1+m2a2+…+m n a n .
[0091] Among them, each weight can be preset according to the importance of each similar content, and the embodiment of the present invention does not make any specific limitation to this.
[0092] It is understandable that by clustering the accessed contents of the first candidate user, the regularity of the contents that the first candidate user is interested in can be determined. Furthermore, the sum of the similarities between the target attribute characteristics of the first candidate user obtained by clustering and the similar contents can reflect the possibility that the similar contents are the contents of interest to the first candidate user.
[0093] Therefore, by recommending similar content to the first candidate user when the first candidate user meets the condition that the sum of the similarities between his target attribute characteristics and the characteristics of multiple similar contents is greater than or equal to the similarity threshold, it can be ensured that the first candidate user can be recommended the content that he is interested in, thereby improving the recommendation effect.
[0094] In one embodiment, the processor may also be configured to filter the second user:
[0095] Obtaining user behavior of the second candidate user;
[0096] According to a preset screening strategy, whether the matching user behavior and at least one of the multiple similar contents reflects a strong behavior;
[0097] When the user behavior and at least one of the plurality of similar contents reflect a strong behavior, the second user to be selected is determined to be the second user.
[0098] The recommendation device may obtain the user behavior of the second candidate user in history, or obtain the user behavior of the second candidate user within a period of time (e.g., within a week or a month, etc.), wherein the user behavior refers to the operation behavior of the second candidate user on each content, such as liking, collecting, purchasing, etc. the content.
[0099] The recommendation device may determine whether the second candidate user exhibits strong behavior towards multiple similar contents according to a preset screening strategy.
[0100] The preset screening strategy may include determining whether the second candidate user exhibits strong behavior on the similar content according to whether the operation data of the second candidate user on the similar content reaches a threshold corresponding to the operation.
[0101] For example, when the time for which the second candidate user browses a similar content exceeds a time threshold, it can be considered that the second candidate user has shown strong behavior towards the similar content. Or, when the number of times the second candidate user visits a similar content exceeds a number threshold, it can be considered that the second candidate user has shown strong behavior towards the similar content.
[0102] The preset screening strategy may also include determining whether the second candidate user exhibits a strong behavior on the similar content according to whether the operation of the second candidate user on the similar content is a specific operation.
[0103] For example, when the second candidate user has liked a certain similar content, it can be considered that the second candidate user has shown strong behavior towards the similar content. Alternatively, when the second candidate user has collected a certain similar content, it can be considered that the second candidate user has shown strong behavior towards the similar content. Alternatively, when the second candidate user has purchased a certain similar content, it can be considered that the second candidate user has shown strong behavior towards the similar content. The recommendation device provided in the embodiment of the present application determines whether the user is the second user by judging whether the user has shown strong behavior towards at least one of multiple similar contents. It can accurately determine whether the user is interested in multiple similar contents, so as to specifically recommend other similar contents of interest to the user, thereby ensuring the recommendation effect.
[0104] In one embodiment, the processor may further be configured to: filter the first user by any of the following methods:
[0105] The user who has visited the target content is regarded as the first user;
[0106] The users who have operated the target content on the network platform are regarded as the first users.
[0107] It should be noted that, since the content feature sequence of the first content can be obtained based on the same user accessing different contents, and the correlation and similarity between the contents accessed by the same user are relatively strong, the use of the content feature sequence in the embodiment of the present application can help match similar content, thereby improving the recommendation efficiency and recommendation effect.
[0108] Furthermore, when a user has visited the target content, it means that the user is more likely to be interested in the target content. Therefore, taking the user as the first user can not only ensure that the correlation and similarity between the contents visited by the same user are relatively strong, but also ensure that the correlation and similarity between the contents visited by the user and the target content are relatively strong, thereby further ensuring the recommendation effect.
[0109] Similarly, when a user has operated on the target content on the network platform, it can also indicate that the user is more likely to be interested in the target content. Therefore, taking the user as the first user can ensure that the relevance and similarity between the content accessed by the user and the target content are relatively strong, thereby further ensuring the recommendation effect.
[0110] Among them, the network platform can refer to various types of network-based platforms, such as social platforms, review platforms, and sharing platforms.
[0111] In one embodiment, the processor may be specifically configured to implement any of the following:
[0112] 1. The first content whose similarity between the target vector and the first vector exceeds a similarity threshold, and the target content are regarded as a plurality of similar contents;
[0113] The similarity threshold may be, for example, 80%, 90%, etc., and its specific value may be adjusted according to actual conditions, which is not specifically limited in the embodiment of the present invention.
[0114] For example, the recommendation device calculates the similarity between the target vector A of the target content a and the first vector B of the first content b, the first vector C of the first content c, ..., the first vector N of the first content n, and determines that the similarity between the target vector A and the first vector B and the similarity between the target vector A and the first vector C exceed the similarity threshold. The recommendation device then regards the target content a, the first content b, and the first content c as multiple similar contents.
[0115] 2. The first content whose similarity between the target vector and the first vector is the greatest, and the target content, are regarded as a plurality of similar contents.
[0116] For example, the recommendation device calculates the similarity between the target vector A of the target content a and the first vector B of the first content b, the first vector C of the first content c, ..., and the first vector N of the first content n, and determines that the similarity between the target vector A and the first vector N is the greatest. The recommendation device then regards the target content a and the first content n as multiple similar contents.
[0117] The recommendation device provided by the embodiment of the present invention can determine multiple similar contents in multiple ways, and can adapt to the content recommendation needs in various scenarios, thereby ensuring the purpose of accurately recommending the content that the user is interested in and ensuring the recommendation effect.
[0118] In one embodiment, the mandatory behavior may include at least one of the following:
[0119] 1. The duration of browsing content exceeds the duration threshold;
[0120] The duration threshold may be, for example, 30 seconds, 60 seconds, etc., and its specific value may be adjusted according to actual conditions, and the embodiment of the present invention does not specifically limit this.
[0121] For example, when the second user browses one of the multiple similar contents for a time period exceeding a time threshold, such as 30 seconds, the recommendation device may determine that the second user has shown strong behavior towards the content and recommend other content among the multiple similar contents except the content to the second user.
[0122] 2. The number of times the content is accessed exceeds the threshold;
[0123] The number threshold may be, for example, 2 times, 5 times, etc., and its specific value may be adjusted according to actual conditions, and the embodiment of the present invention does not specifically limit this.
[0124] For example, when the second user visits one of the multiple similar contents more than a threshold number, such as 2 times, the recommendation device can determine that the second user has shown strong behavior towards the content and recommend other content among the multiple similar contents except the content to the second user.
[0125] 3. Like the content;
[0126] When the second user has liked one of the multiple similar contents, the recommendation device may determine that the second user has shown a strong behavior towards the content, and recommend other contents among the multiple similar contents except the content to the second user.
[0127] 4. Collect content;
[0128] When the second user has collected one of the multiple similar contents, the recommendation device may determine that the second user has shown strong behavior towards the content, and recommend other contents of the multiple similar contents except the content to the second user.
[0129] 5. Purchase content.
[0130] When the second user has purchased one of the multiple similar contents, the recommendation device may determine that the second user has shown strong behavior towards the content, and recommend other contents except the content among the multiple similar contents to the second user.
[0131] It should be noted that when the recommendation device determines the second user based on the user behavior of the second candidate user, that is, when multiple similar contents include content in which the second user exhibits strong behavior, the recommendation device can still recommend content in which the second user exhibits strong behavior to the second user, thereby further improving the recommendation effect.
[0132] For example, when user A has liked content A (one of multiple similar contents) (strong behavior), it indicates that user A has a strong willingness to purchase content A. After determining that user A is the second user, the recommendation device can recommend content A to user A, thereby further increasing the probability that user A purchases content A, thereby improving the recommendation effect. Alternatively, when user A has purchased content A (one of multiple similar contents) (strong behavior), the recommendation device can recommend content A to user A after determining that user A is the second user, thereby facilitating user A to repurchase content A, thereby improving the recommendation effect.
[0133] The recommendation device provided in the embodiment of the present application determines whether the user has shown strong behavior towards the content by judging whether the time the user has browsed a certain content among multiple similar contents exceeds a threshold, whether the number of times the user has visited a certain content exceeds a threshold, whether the user has liked, collected, or purchased a certain content, etc., and can accurately determine whether the user is interested in multiple similar contents, so as to specifically recommend other similar contents of interest to the user, thereby ensuring the recommendation effect.
[0134] In one embodiment, the processor may be specifically configured to implement any of the following:
[0135] 1. Determine the cosine similarity between the target vector and the first vector of each first content;
[0136] After determining the target vector of the target content and the first vectors of each first content, the recommendation device may determine the cosine value of the angle between the target vector and each first vector, and then determine the similarity between the target vector and each first vector according to each cosine value.
[0137] 2. Determine the distance similarity between the target vector and the first vector of each first content;
[0138] After determining the target vector of the target content and the first vectors of each first content, the recommendation device can determine the distance similarity between the target vector and each first vector, such as Minkowski distance similarity, Manhattan distance similarity, Euclidean distance similarity, Chebyshev distance similarity, etc.
[0139] The recommendation device provided by the embodiment of the present invention determines the similarity between the target content and each first content in a variety of ways, and can adapt to the similarity calculation requirements in various scenarios, thereby ensuring the purpose of accurately recommending the content of interest to the user and ensuring the recommendation effect.
[0140] In one embodiment, the processor may be specifically configured to:
[0141] Establishing a vector index of each feature in the content feature sequence;
[0142] Determine the vector corresponding to each attribute feature of the target content according to the attribute features of the target content and the vector index;
[0143] The vectors corresponding to the attribute features of the target content are combined to obtain the target vector of the target content.
[0144] After determining the vector of each feature of the content feature sequence, the recommendation device may establish an index of the feature and the vector, ie, a vector index, and cache the vector index in a database such as Redis.
[0145] Afterwards, the recommendation device can read the vector index in a database such as Redis, and match the vectors corresponding to each attribute feature from the vector index according to the attribute features of the target content, and combine the vectors corresponding to each attribute feature to obtain the target vector of the target content.
[0146] The specific method of combination may be vector splicing, such as horizontal splicing of vectors.
[0147] The recommendation device provided by the embodiment of the present invention can ensure the accuracy of the target vector by establishing an index of the vector of each feature in the content feature sequence and obtaining the target vector of the target content according to the attribute characteristics of the target content and the index, thereby improving the accuracy of determining similar content.
[0148] The following uses an example of a recommended device provided by an embodiment of the present invention as an example to further illustrate the technical solution of the present invention:
[0149] This example may be a recommendation system including the recommendation device provided in the embodiment of the present application. The structure of the recommendation system is as follows: Figure 2 As shown, it includes: data module, Word2vec module, DSSM module, and expansion module.
[0150] The data module is used to clean the historical data of the first content, obtain the attribute characteristics of the first content, and generate a content characteristic sequence.
[0151] The DSSM module is used to train the DSSM model based on the attribute feature data of the historical content, and input the attribute features of the first content into the trained DSSM model to obtain each first vector.
[0152] The Word2vec module is used to train the Word2vec model based on the content feature sequence, and input the attribute features of the target content into the trained Word2vec model to obtain the target vector of the target content;
[0153] The expansion module is used to calculate the similarity between the target vector and each first vector, so as to determine a plurality of similar contents for recommendation to the user.
[0154] The recommendation device provided by the embodiment of the present invention is described below. The recommendation device described below and the recommendation method described above can refer to each other and can achieve the same technical effects, which will not be described in detail here.
[0155] Figure 3 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 3As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330 and a communication bus 340, wherein the processor 310, the communication interface 320 and the memory 330 communicate with each other through the communication bus 340. The processor 310 may call the logic instructions in the memory 330 to execute the method steps executed by the processor of each embodiment above, for example, including:
[0156] Determine a content feature sequence according to the order in which the first user accesses the first contents and the attribute features of the first contents, and determine a vector of each feature in the content feature sequence;
[0157] Determining a target vector of the target content according to the attribute characteristics of the target content and the vectors of the characteristics;
[0158] Determine a first vector of each first content according to the attribute characteristics of each first content, and determine a plurality of similar contents according to the similarities between the target vector and the first vector of each first content;
[0159] The multiple similar contents are recommended as recommended contents. In addition, the logic instructions in the above-mentioned memory 330 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.
[0160] On the other hand, the present invention further provides a computer program product, the computer program product comprising a computer program, the computer program may be stored on a non-transitory computer-readable storage medium, and when the computer program is executed, the computer can execute the method steps executed by the processor of each of the above embodiments, for example, including:
[0161] Determine a content feature sequence according to the order in which the first user accesses the first contents and the attribute features of the first contents, and determine a vector of each feature in the content feature sequence;
[0162] Determining a target vector of the target content according to the attribute characteristics of the target content and the vectors of the characteristics;
[0163] Determine a first vector of each first content according to the attribute characteristics of each first content, and determine a plurality of similar contents according to the similarities between the target vector and the first vector of each first content;
[0164] The multiple similar contents are recommended as recommended contents.
[0165] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed, implements the method steps executed by the processor of each of the above embodiments, for example, including:
[0166] Determine a content feature sequence according to the order in which the first user accesses the first contents and the attribute features of the first contents, and determine a vector of each feature in the content feature sequence;
[0167] Determining a target vector of the target content according to the attribute characteristics of the target content and the vectors of the characteristics;
[0168] Determine a first vector of each first content according to the attribute characteristics of each first content, and determine a plurality of similar contents according to the similarities between the target vector and the first vector of each first content;
[0169] The multiple similar contents are recommended as recommended contents. The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.
[0170] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0171] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A recommendation device, characterized in that: include: A memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the following steps: Determine a content feature sequence according to the order in which the first user accesses the first contents and the attribute features of the first contents, and determine a vector of each feature in the content feature sequence; Determining a target vector of the target content according to the attribute characteristics of the target content and the vectors of the characteristics; Determine a first vector of each first content according to the attribute characteristics of each first content, and determine a plurality of similar contents according to the similarities between the target vector and the first vector of each first content; The multiple similar contents are recommended as recommended contents.
2. The recommendation device according to claim 1, characterized in that: The processor is specifically used for: A second user is determined according to the plurality of similar contents, and at least one of the plurality of similar contents is recommended to the second user, wherein the plurality of similar contents includes the target content.
3. The recommendation device according to claim 2, characterized in that: The processor is further configured to filter the second user: Acquire multiple access contents of the first candidate user; Clustering the plurality of access contents according to a preset clustering algorithm to generate target attribute features of the first candidate user; Performing similarity matching between the target attribute feature and the features of the plurality of similar contents; When the sum of the similarities between the target attribute feature and the features of the plurality of similar contents is greater than or equal to a similarity threshold, the first candidate user is determined to be the second user.
4. The recommendation device according to claim 2, characterized in that: The processor is further configured to filter the second user: Obtaining user behavior of the second candidate user; According to a preset screening strategy, whether matching the user behavior with at least one of the plurality of similar contents reflects a strong behavior; When the user behavior and at least one of the plurality of similar contents reflect a strong behavior, the second candidate user is determined to be the second user.
5. The recommendation device according to claim 1, characterized in that: The processor is further configured to: filter the first user by any of the following methods: taking a user who has visited the target content as the first user; A user who has performed operations on the target content on a network platform is regarded as the first user.
6. The recommendation device according to claim 1, characterized in that: The processor is specifically configured to implement any of the following: The first content whose similarity between the target vector and the first vector exceeds a similarity threshold, and the target content, are used as the multiple similar contents; The first content having the greatest similarity between the target vector and the first vector and the target content are used as the multiple similar contents.
7. The recommendation device according to claim 4, characterized in that: The mandatory behavior includes at least one of the following: The time spent browsing content exceeds the time threshold; The number of times the content is accessed exceeds the threshold; Like content; Collection content; Purchase content.
8. The recommendation device according to claim 1, characterized in that: The processor is specifically configured to implement any of the following: Determine the cosine similarity between the target vector and the first vector of each first content; Determine the distance similarity between the target vector and the first vector of each first content.
9. The recommendation device according to any one of claims 1 to 8, characterized in that: The processor is specifically used for: Establishing a vector index of each feature in the content feature sequence; Determining, according to the attribute features of the target content and the vector index, a vector corresponding to each attribute feature of the target content; The vectors corresponding to the attribute features of the target content are combined to obtain a target vector of the target content.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the method steps executed by the processor according to any one of claims 1 to 9 are implemented.
11. A computer program product, comprising a computer program, characterized in that When the computer program is executed, the method steps executed by the processor according to any one of claims 1 to 9 are implemented.