CN109460519B - Browsing object recommendation method and device, storage medium and server - Google Patents
Browsing object recommendation method and device, storage medium and server Download PDFInfo
- Publication number
- CN109460519B CN109460519B CN201811621490.1A CN201811621490A CN109460519B CN 109460519 B CN109460519 B CN 109460519B CN 201811621490 A CN201811621490 A CN 201811621490A CN 109460519 B CN109460519 B CN 109460519B
- Authority
- CN
- China
- Prior art keywords
- browsing
- browsing object
- user
- vector
- target user
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A browsing object recommendation method and device, a storage medium and a server are provided, and the method comprises the following steps: determining a dense vector of each browsing object according to reference data, wherein the reference data comprises object identifiers of a plurality of browsing objects arranged according to a historical browsing sequence, and the dense vector at least describes a browsing association relation between the browsing object and each other browsing object; clustering the dense vectors of all browsing objects to obtain k browsing object central vectors, wherein each browsing object central vector represents a browsing object type, and k is a positive integer greater than 1; determining browsing object interest vectors of the target user according to the browsing records of the target user and the k browsing object center vectors, wherein the browsing object interest vectors are used for representing the browsing object types which are interested by the target user; and recommending the browsing object to the target user based on the browsing object interest vector. By the scheme of the invention, the personalized browsing object display can be carried out aiming at the user, and the clicking rate of the browsing object is improved.
Description
Technical Field
The invention relates to the technical field of machine learning, in particular to a browsing object recommendation method and device, a storage medium and a server.
Background
With the continuous development of internet technology, more and more users start browsing various information on the internet, such as news, videos, commodity information, and the like.
In order to improve user experience, when pushing information of various browsing objects to a user, the pushed information needs to be more in line with the interest of the user, so as to obtain clicks of the user as much as possible. However, in the prior art, when recommending a browsing object to a user, the accuracy of recommendation still needs to be improved.
Disclosure of Invention
The technical problem to be solved by the invention is how to determine the browsing object which is interested by the user so as to display the personalized browsing object aiming at the user and improve the accuracy of recommendation.
In order to solve the above technical problem, an embodiment of the present invention provides a browsing object recommendation method, including: determining a dense vector of each browsing object according to reference data, wherein the reference data comprises object identifiers of a plurality of browsing objects arranged according to a historical browsing sequence, and the dense vector at least describes a browsing association relation between the browsing object and each other browsing object; clustering the dense vectors of all browsing objects to obtain k browsing object central vectors, wherein each browsing object central vector represents a browsing object type, and k is a positive integer greater than 1; determining browsing object interest vectors of the target user according to the browsing records of the target user and the k browsing object center vectors, wherein the browsing object interest vectors are used for representing the browsing object categories which are interested by the target user; recommending the browsing object to the target user based on the browsing object interest vector.
Optionally, the determining the browsing object interest vector of the target user according to the browsing record of the target user and the k browsing object center vectors includes: determining a user browsing object and user browsing time browsed by the target user according to the browsing record of the target user; determining a browsing object vector of the target user according to the user browsing object and the k browsing object center vectors, and determining a browsing object time sequence vector of the user browsing object according to the user browsing time and preset comparison time; and weighting the browsing object vector and the browsing object time sequence vector of the user to obtain the browsing object interest vector of the target user.
Optionally, the determining the browsing object vector of the target user according to the user browsing object and the k browsing object center vectors includes: similarity calculation is carried out on the user browsing objects and the k browsing object center vectors, so that a browsing object category to which each user browsing object browsed by the target user belongs is obtained; counting the number of the user browsing objects in each browsing object category; normalizing the statistical result to obtain the user browsing weight of the target user to the k browsing object categories; and weighting the k browsing object center vectors by using the user browsing weight to obtain the browsing object vector of the target user.
Optionally, the determining a browsing object timing vector of the user browsing object according to the user browsing time and a preset comparison time includes: determining the time influence weight of each browsing object category represented by the k browsing object center vectors based on the user browsing time and the preset comparison time; and weighting the k browsing object center vectors by using the time influence weights to obtain browsing object time sequence vectors of the user browsing objects.
Optionally, the time influence weight of each browsing object category characterized by the k browsing object center vectors is determined as follows: the longer the interval time between the user browsing time and the preset comparison time is, the smaller the time influence weight of the browsing object category is; the shorter the interval time between the user browsing time and the preset comparison time is, the larger the time influence weight of the browsing object category is.
Optionally, the weighting the browsing object vector and the browsing object timing vector of the target user includes: and weighting the browsing object vector and the browsing object time sequence vector of the target user by using a preset harmonic factor, wherein the preset harmonic factor is greater than or equal to 0 and less than or equal to 1.
Optionally, the recommending a browsing object to the target user based on the browsing object interest vector includes: determining each candidate browsing object to be recommended; calculating the similarity of each candidate browsing object and the browsing object interest vector of the target user; sequencing each candidate browsing object according to the similarity calculation result; and recommending a preset number of candidate browsing objects ranked in the front to the user according to the ranking result.
Optionally, before determining the dense vector of each browsing object according to the reference data, the method further includes: and extracting the reference data from the browsing records of the users.
Optionally, the determining the dense vector of each browsing object according to the reference data includes: and determining the reference data as input data of a depth migration algorithm, and determining a dense vector of each browsing object by using the depth migration algorithm.
In order to solve the above technical problem, an embodiment of the present invention further provides a browsing object recommending apparatus, including: the first determination module is suitable for determining a dense vector of each browsing object according to reference data, wherein the reference data comprises object identifications of a plurality of browsing objects arranged according to a historical browsing sequence, and the dense vector at least describes a browsing incidence relation between the browsing object and other browsing objects; the clustering module is suitable for clustering the dense vectors of the browsing objects to obtain k browsing object central vectors, each browsing object central vector represents a browsing object category, and k is a positive integer greater than 1; the second determination module is suitable for determining the browsing object interest vector of the target user according to the browsing object browsing record of the target user and the k browsing object central vectors; and the recommending module is suitable for recommending the browsing object to the target user based on the browsing object interest vector.
To solve the above technical problem, an embodiment of the present invention further provides a storage medium having stored thereon computer instructions, where the computer instructions execute the steps of the above method when executed.
In order to solve the above technical problem, an embodiment of the present invention further provides a server, including a memory and a processor, where the memory stores computer instructions executable on the processor, and the processor executes the computer instructions to perform the steps of the above method.
Compared with the prior art, the technical scheme of the embodiment of the invention has the following beneficial effects:
the embodiment of the invention provides a browsing object recommendation method, which comprises the following steps: determining a dense vector of each browsing object according to reference data, wherein the reference data comprises object identifiers of a plurality of browsing objects arranged according to a historical browsing sequence, and the dense vector at least describes a browsing association relation between the browsing object and each other browsing object; clustering the dense vectors of all browsing objects to obtain k browsing object central vectors, wherein each browsing object central vector represents a browsing object type, and k is a positive integer greater than 1; determining browsing object interest vectors of the target user according to the browsing records of the target user and the k browsing object center vectors, wherein the browsing object interest vectors are used for representing the browsing object categories which are interested by the target user; recommending the browsing object to the target user based on the browsing object interest vector. According to the technical scheme provided by the embodiment of the invention, the reference data can be utilized to obtain the dense vector at least describing the browsing incidence relation between each browsing object and each other browsing object, so that the browsing object categories represented by the browsing object interest vectors can be obtained by analyzing (for example, clustering) the dense vectors, and the browsing objects can be recommended to the target user after the browsing object interest vectors interested by the target user are determined. After the browsing object which is interested by the user is determined, personalized browsing object display can be performed for the user, and the clicking rate of the browsing object is improved.
Further, according to the browsing record of the target user, determining a user browsing object and user browsing time browsed by the target user; determining a browsing object vector of the target user according to the user browsing object and the k browsing object center vectors, and determining a browsing object time sequence vector of the user browsing object according to the user browsing time and preset comparison time; and weighting the browsing object vector and the browsing object time sequence vector of the user to obtain the browsing object interest vector of the target user. By the embodiment of the invention, the browsing object vector and the browsing object time sequence vector can be determined based on the browsing record of the target user, the browsing object interest vector of the target user is obtained by weighting, and the possibility of recommending the interested browsing object for the target user is further provided.
Further, the recommending browsing objects to the target user based on the browsing object interest vector comprises: determining each candidate browsing object to be recommended; calculating the similarity of each candidate browsing object and the browsing object interest vector of the target user; sequencing each candidate browsing object according to the similarity calculation result; and recommending a preset number of candidate browsing objects ranked in the front to the user according to the ranking result. Through the technical scheme provided by the embodiment of the invention, the similarity between the interest vector of the browsing object and the candidate browsing object can be calculated, and then the most interesting browsing object is recommended to the target user according to the similarity calculation result, so that the possibility is further provided for improving the click rate of the user.
Drawings
Fig. 1 is a schematic flowchart of a browsing object recommendation method according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart of one embodiment of step S103 shown in FIG. 1;
fig. 3 is a schematic structural diagram of a browsing object recommending apparatus according to an embodiment of the present invention.
Detailed Description
As will be appreciated by those skilled in the art, as background art, in recommending browsing objects to a user, the accuracy of the recommendation is low in the prior art.
The embodiment of the invention provides a browsing object recommendation method, which comprises the following steps: determining a dense vector of each browsing object according to reference data, wherein the reference data comprises object identifiers of a plurality of browsing objects arranged according to a historical browsing sequence, and the dense vector at least describes a browsing association relation between the browsing object and each other browsing object; clustering the dense vectors of all browsing objects to obtain k browsing object central vectors, wherein each browsing object central vector represents a browsing object type, and k is a positive integer greater than 1; determining browsing object interest vectors of the target user according to the browsing records of the target user and the k browsing object center vectors, wherein the browsing object interest vectors are used for representing the browsing object categories which are interested by the target user; recommending the browsing object to the target user based on the browsing object interest vector.
According to the technical scheme provided by the embodiment of the invention, the reference data can be utilized to obtain the dense vector at least describing the browsing incidence relation between each browsing object and each other browsing object, so that the browsing object categories represented by the browsing object interest vectors can be obtained by analyzing (for example, clustering) the dense vectors, and the browsing objects can be recommended to the target user after the browsing object interest vectors interested by the target user are determined. After the browsing object which is interested by the user is determined, personalized browsing object display can be performed for the user, and the clicking rate of the browsing object is improved.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in detail below.
As used herein, the terms "comprising," "including," and the like are to be construed as open-ended terms, i.e., "including/including but not limited to," meaning that additional content can be included as well. In the present disclosure, the term "based on" is "based, at least in part, on".
Fig. 1 is a flowchart illustrating a browsing object recommendation method according to an embodiment of the present invention. The browsing object recommendation method can be executed by a server and can be used for carrying out personalized display on a user. In particular implementations, the server may be a single server or a server cluster comprised of multiple servers.
As a non-limiting example, the browsing object may be a commodity, or may be a video, an information message, or the like.
Specifically, the browsing object recommendation method may include the following steps:
step S101, determining a dense vector of each browsing object according to reference data, wherein the reference data comprises object identifiers of a plurality of browsing objects arranged according to a historical browsing sequence, and the dense vector at least describes a browsing incidence relation between the browsing object and other browsing objects;
step S102, clustering dense vectors of all browsing objects to obtain k browsing object center vectors, wherein each browsing object center vector represents a browsing object type, and k is a positive integer greater than 1;
step S103, determining browsing object interest vectors of the target user according to the browsing records of the target user and the k browsing object center vectors, wherein the browsing object interest vectors are used for representing the types of browsing objects which are interested by the user;
and step S104, recommending a browsing object to the target user based on the browsing object interest vector.
More specifically, in step S101, the server may obtain browsing records of each user from the internet, where the browsing records may be commodities, videos, information, and the like.
Then, the browsing records may be processed, and browsing object identification sequences of browsing objects are extracted from the browsing records of each user. Each browsing object identification may represent one or more attribute information of the browsing object, and each browsing object identification may be used to represent a single discrete data.
Specifically, the browsing record may be segmented according to a time length, and an object identifier of each browsing object in the segmented browsing record is extracted to obtain the browsing object identifier sequence. In general, the time length may be set to 1 day, 1 hour, 0.5 hour, or the like. For example, for the browsing record of each user, the browsing record of the user is segmented according to the interval time of 0.5 hour, and then the object identifier of each browsing object in each segment is extracted, so that a browsing object identifier sequence can be obtained. In practical applications, the time period can be set by those skilled in the art according to actual needs, and is not limited to the above specific scheme.
Further, the obtained multiple browsing object identification sequences (for example, the number of the browsing object identification sequences may reach more than millions) may be determined as reference data, and the object identifications in the reference data are arranged according to the historical browsing order. And vectorizing the reference data to obtain dense vectors of all browsing objects. The dense vector may describe at least browsing associations of the browsing object with other respective browsing objects. For example, the dense vector may represent attribute information of a product, in addition to describing a browsing association relationship between one product browsed by the user and other products.
In a specific implementation, the reference data may be vectorized using a deep walk (deep walk) algorithm. The DeepWalk algorithm is a network characterization learning algorithm for learning vector representations of nodes in a network or learning attribute information of a network graph. The vectors obtained by the Deepwalk algorithm can express each node in the network and the relationship among the nodes, namely, the closer the distance of the vectors corresponding to the nodes with closer relationship in the original network in the space of the nodes. In the embodiment of the invention, the vector representation mode of each browsing object in the reference data and the browsing incidence relation of each browsing object browsed by the user can be obtained by adopting a Deepwalk algorithm.
In a specific implementation, the dense vector obtained by using the deep walk algorithm may be a K-dimensional vector, for example, K ═ 32 or 64. The elements of the K-dimensional vector may be floating point numbers, and logical operations may be performed based on the K-dimensional vector, and may represent an association relationship for a user to browse a plurality of browsing objects. Those skilled in the art will appreciate that in particular implementations, the user's browsing objects may be represented using the dense vector without regard to the particular meaning of each dimension of data.
Taking the browsing object as a commodity as an example, the server may extract a plurality of commodity identification sequences from the browsing records of the respective users to obtain a group of commodity identification sequences. Each item identification represents a discrete datum. The group of commodity identification sequences can be vectorized through the Deepwalk algorithm, so that dense vectors of all commodities can be obtained. The dense vector may represent an association of each item viewed by the user.
In step S102, the dense vectors of the respective browsing objects may be clustered. In a specific implementation, a K-means (K-means) clustering algorithm may be adopted to cluster the dense vectors of the respective browsing objects.
The K-means clustering algorithm adopts the distance as an evaluation index of similarity, and considers that the closer the distance between two objects is, the greater the similarity of the two objects is, so that browsing objects with high similarity can be clustered into a class. After clustering is completed by adopting a clustering algorithm, k browsing object center vectors can be obtained. Each browsing object center vector may characterize a browsing object class, k being a positive integer greater than 1.
In step S103, before recommending browsing objects for the target user, the server may determine browsing object interest vectors of the target user according to the browsing records of the target user and the k browsing object center vectors, so as to obtain a browsing object category in which the target user is interested.
Fig. 2 is a schematic flow chart of an embodiment of step S103 in fig. 1. Specifically, the step S103 may include:
step S1031, determining a user browsing object and user browsing time browsed by the target user according to the browsing record of the target user;
step S1032, determining a browsing object vector of the target user according to the user browsing object and the k browsing object center vectors, and determining a browsing object timing sequence vector of the user browsing object according to the user browsing time and a preset comparison time;
step S1033, weighting the browsing object vector and the browsing object timing vector of the user to obtain a browsing object interest vector of the target user.
Specifically, in step S1031, the server may save a browsing record of the target user. And analyzing the browsing record to obtain user browsing objects browsed by the target user and user browsing time of each user browsing object.
In a specific implementation, assuming that the browsing object is a commodity, as shown in table 1, the browsing record of the target user at least may include: the method comprises the steps of user identification, commodity identification browsed by a user and initial browsing time for browsing commodities pointed by the commodity identification. Table 1 records the items (indicated by item identifiers) and the initial browsing time viewed by user a, and the items (indicated by item identifiers) and the initial browsing time viewed by user B. The commodity can be used as a user browsing object, and the initial browsing time of the commodity can be used as the user browsing time.
TABLE 1
In step S1032, the server may determine a browsing object vector of the target user according to the user browsing object and the k browsing object center vectors.
Specifically, similarity calculation may be performed between each user browsing object of the target user and the k browsing object center vectors, so as to obtain a browsing object category to which each user browsing object browsed by the target user belongs.
Secondly, the number of the user browsing objects in each browsing object category can be counted. In specific implementation, the histogram may be used to count the number of each browsing object. For example, the histogram statistic may be a vector Hist, and then the vector Hist may be expressed as Hist ═ n1,...,ni,…,nK]i∈K,ni≥0，niIndicating the number of browsing objects in each browsing object category.
Then, the statistical result may be normalized to obtain the user browsing weights of the user for the k browsing object categories. In a specific implementation, the following formula can be used:
wherein, HistwRepresenting the browsing weight vector, k representing the number of browsing object center vectors, alphaiElements representing the browsing weight vector.
Further, the k browsing object center vectors may be weighted by using the user browsing weights, so as to obtain a browsing object vector of the target user. In specific implementation, the formula User can be adoptedHist＝α1·ω1+…+αi·ωi+…+αk·ωkWherein, UserHistRepresenting said browsing object vector, ωiIndicating the ith browsing object center vector, and k indicating the number of browsing object center vectors.
In a specific implementation, the browsing object timing vector of the user browsing object may be determined according to the user browsing time and a preset comparison time. Specifically, the time interval between the browsing time of the user and the preset comparison time may be calculated, and the time influence weight of each browsing object category represented by the k browsing object center vectors may be determined.
As a non-limiting example, the preset contrast time may be set as a current time, and if the interval time between the user browsing time and the current time is longer, the time influence weight of the browsing object category may be set to be smaller; on the contrary, if the interval time between the browsing time of the user and the current time is shorter, the time influence weight of the browsing object category is larger.
In a specific implementation, the time influence weight may be calculated assuming that the user browsing time is inversely proportional to the preset contrast time. In practical applications, those skilled in the art may change more embodiments according to actual needs, and details are not described herein.
Further, the time influence weights may be used to weight the k browsing object center vectors, so that browsing object timing vectors of the user browsing objects may be obtained. In a specific implementation, assume βjRepresenting the time influence weight, omega, of the jth browsing object vector in the browsing object class corresponding to the ith browsing object center vectoriRepresenting the ith browsing object center vector, k representing the number of browsing object center vectors, niRepresenting the browsing object which is subordinate to the ith browsing object central vector, and then the time sequence vector User of the browsing objectDecayThe formula of (1) is as follows:
in step S1033, the browsing object vector and the browsing object timing vector of the user may be weighted to obtain a browsing object interest vector of the target user.
In a specific implementation, the browsing object vector and the browsing object timing vector of the target user may be weighted by using a preset harmonic factor.
The preset harmonic factor can be set according to the interest in the browsing object or the importance degree of the browsing time. The preset harmonic factor is greater than or equal to 0 and less than or equal to 1. Preferably, the reconciliation factor may be set to 0.5, which indicates that the browsing objects and browsing time of the target user are equal in proportion to the browsing objects interested in analyzing the user.
In a specific implementation, the browsing object interest vector may be represented by the following formula:
UserEmbedding＝p·UserHist+(1-p)·UserDecay,0≤p≤1；
wherein p represents the harmonic factor, UserHistRepresenting said vector of browsing objects, UserDecayRepresenting the browsing object timing vector. UserEmbeddingRepresenting the browsing object interest vector.
In step S104, browsing objects, such as commodities, videos, information, and the like, may be recommended to the target user based on the browsing object interest vector.
In a specific implementation, the browsing object that is most interested in by the target user may be recommended to the target user according to the browsing object interest vector.
As a variation, if there are N candidate browsing objects to be recommended, the similarity between each candidate browsing object and the browsing object interest vector of the target user may be calculated, so as to obtain a similarity calculation result between the candidate browsing object and the browsing object interest vector of the target user. Thereafter, the candidate browsing objects may be sorted according to the similarity calculation result, for example, sorted in order of high similarity to low similarity. Further, a preset number of candidate browsing objects ranked in the top can be recommended to the user according to the ranking result.
Further, if a plurality of browsing objects are allowed to be recommended simultaneously, after the browsing objects to be recommended are determined, the sorted goods to be recommended can be displayed to the user according to the sorting result.
As a non-limiting example, suppose a commodity is recommended to the target user A, and the candidate commodities to be recommended include a mobile phone, a hat, a refrigerator, and a basketball. After the browsing object interest vector of the target user A is obtained by the embodiment of the invention, similarity calculation can be respectively carried out on the mobile phone, the hat, the refrigerator and the basketball and the browsing object interest vector of the target user A, and the similarity calculation results are arranged according to the sequence of similarity from high to low, wherein the sequence of the similarity calculation results is the refrigerator, the mobile phone, the basketball and the hat. If only one kind of commodity can be recommended to the target user A, the refrigerator information can be pushed. If two commodities can be recommended to the target user A, a refrigerator and a mobile phone can be recommended, and the refrigerator can be placed at the display position of a page.
Therefore, by the technical scheme provided by the embodiment of the invention, the dense vector expression mode of the browsing object of the user can be obtained by utilizing the reference data, the browsing object interested by the user is determined based on the dense vector capable of carrying out logic operation, and further, the browsing objects such as commodities, videos, information and the like can be pertinently recommended to the user, thereby being beneficial to improving the click rate of the user.
Fig. 3 is a schematic structural diagram of a browsing object recommending apparatus according to an embodiment of the present invention. The browsing object recommending apparatus 3 may be implemented by a server to implement the technical solutions of the methods shown in fig. 1 and fig. 2.
Specifically, the browsing object recommendation apparatus 3 may include: a first determining module 31, adapted to determine a dense vector of each browsing object according to reference data, where the reference data includes object identifiers of multiple browsing objects arranged according to a historical browsing order, and the dense vector at least describes a browsing association relationship between the browsing object and each other browsing object; the clustering module 32 is adapted to cluster the dense vectors of the browsing objects to obtain k browsing object center vectors, each browsing object center vector represents a browsing object category, and k is a positive integer greater than 1; the second determining module 33 is adapted to determine a browsing object interest vector of the target user according to the browsing object browsing record of the target user and the k browsing object center vectors; a recommending module 34 adapted to recommend a browsing object to the target user based on the browsing object interest vector.
In a specific implementation, the second determining module 33 may include: the first determining sub-module 331 is adapted to determine a user browsing object and a user browsing time browsed by the target user according to the browsing record of the target user; a second determining submodule 332, adapted to determine a browsing object vector of the target user according to the user browsing object and the k browsing object center vectors, and determine a browsing object timing vector of the user browsing object according to the user browsing time and a preset comparison time; the weighting submodule 333 is adapted to weight the browsing object vector and the browsing object timing vector of the user to obtain a browsing object interest vector of the target user.
In a specific implementation, the second determining sub-module 332 may include a calculating unit 3321, adapted to perform similarity calculation on the user browsing object and the k browsing object center vectors, so as to obtain a browsing object category to which each user browsing object browsed by the target user belongs; a counting unit 3322, adapted to count the number of the browsing objects of the user in each browsing object category; a normalization unit 3323, adapted to normalize the statistical result to obtain user browsing weights of the target user for k browsing object categories; a first weighting unit 3324, adapted to weight the k browsing object center vectors by using the user browsing weights to obtain browsing object vectors of the target user.
The second determination submodule 332 may further include: a determining unit 3325, adapted to determine a time influence weight of each browsing object category characterized by the k browsing object center vectors based on the user browsing time and a preset comparison time; a second weighting unit 3326, adapted to weight the k browsing object center vectors by using the time influence weights to obtain browsing object timing vectors of the user browsing objects.
In a specific implementation, the second weighting unit 3326 may determine the time influence weight of each browsing object category characterized by the k browsing object center vectors as follows: the longer the interval time between the user browsing time and the preset comparison time is, the smaller the time influence weight of the browsing object category is; the shorter the interval time between the user browsing time and the preset comparison time is, the larger the time influence weight of the browsing object category is.
In a specific implementation, the second determining sub-module 332 is further adapted to weight the browsing object vector and the browsing object timing vector of the target user by using a preset harmonic factor, where the preset harmonic factor is greater than or equal to 0 and less than or equal to 1.
In a specific implementation, the recommendation module 34 may include: a third determining submodule 341 adapted to determine each candidate browsing object to be recommended; a calculating sub-module 342 adapted to calculate a similarity of each candidate browsing object with the browsing object interest vector of the target user; the sorting submodule 343, is suitable for sorting every candidate and browsing the target according to the calculation result of the degree of similarity; and the recommending sub-module 344 is adapted to recommend the top preset number of candidate browsing objects to the user according to the sorting result.
The browsing object recommendation apparatus 3 may further include: an extraction module 35 adapted to extract the reference data from the browsing records of the respective users before determining the dense vector of each browsing object based on the reference data.
In a specific implementation, the first determining module 31 may include: a fourth determination sub-module 311 adapted to determine the reference data as input data for a depth walk algorithm with which to determine a dense vector for each browsing object.
For more contents of the operation principle and the operation mode of the browsing object recommending apparatus 3 shown in fig. 3, reference may be made to the related descriptions in fig. 1 and fig. 2, and details are not repeated here.
Further, the embodiment of the present invention also discloses a storage medium, on which computer instructions are stored, and when the computer instructions are executed, the technical solution of the method in the embodiment shown in fig. 1 and fig. 2 is executed. Preferably, the storage medium may include a computer-readable storage medium such as a non-volatile (non-volatile) memory or a non-transitory (non-transient) memory. The computer readable storage medium may include ROM, RAM, magnetic or optical disks, and the like.
Further, an embodiment of the present invention further discloses a server, which includes a memory and a processor, where the memory stores computer instructions capable of being executed on the processor, and the processor executes the computer instructions to execute the technical solutions of the methods in the embodiments shown in fig. 1 and fig. 2.
Although the present invention is disclosed above, the present invention is not limited thereto. Various changes and modifications may be effected therein by one skilled in the art without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (10)
1. A browsing object recommendation method is characterized by comprising the following steps:
determining a dense vector of each browsing object according to reference data, wherein the reference data comprises object identifiers of a plurality of browsing objects arranged according to a historical browsing sequence, and the dense vector at least describes a browsing association relation between the browsing object and each other browsing object;
clustering the dense vectors of all browsing objects to obtain k browsing object central vectors, wherein each browsing object central vector represents a browsing object type, and k is a positive integer greater than 1;
determining browsing object interest vectors of a target user according to browsing records of the target user and the k browsing object center vectors, wherein the browsing object interest vectors are used for representing browsing object categories which are interested by the target user, and the browsing records of the target user are browsing records of the target user within a period of time;
recommending a browsing object to the target user based on the browsing object interest vector;
wherein the determining the browsing object interest vector of the target user according to the browsing record of the target user and the k browsing object center vectors comprises:
determining a user browsing object and user browsing time browsed by the target user according to the browsing record of the target user;
determining a browsing object vector of the target user according to the user browsing object and the k browsing object center vectors, and determining a browsing object time sequence vector of the user browsing object according to the user browsing time and preset comparison time;
weighting the browsing object vector and the browsing object time sequence vector of the user to obtain a browsing object interest vector of the target user;
wherein the weighting the browsing object vector and the browsing object timing vector of the target user comprises:
weighting the browsing object vector and the browsing object time sequence vector of the target user by using a preset harmonic factor, wherein the preset harmonic factor is greater than or equal to 0 and less than or equal to 1, and the preset harmonic factor is set according to the interest of the browsing object or the importance degree of the browsing time.
2. The browsing object recommendation method according to claim 1, wherein the determining the browsing object vector of the target user according to the user browsing object and the k browsing object center vectors comprises:
similarity calculation is carried out on the user browsing objects and the k browsing object center vectors, so that a browsing object category to which each user browsing object browsed by the target user belongs is obtained;
counting the number of the user browsing objects in each browsing object category;
normalizing the statistical result to obtain the user browsing weight of the target user to the k browsing object categories;
and weighting the k browsing object center vectors by using the user browsing weight to obtain the browsing object vector of the target user.
3. The browsing object recommendation method according to claim 1, wherein the determining a browsing object timing vector of the user browsing object according to the user browsing time and a preset comparison time comprises:
determining the time influence weight of each browsing object category represented by the k browsing object center vectors based on the user browsing time and the preset comparison time;
and weighting the k browsing object center vectors by using the time influence weights to obtain browsing object time sequence vectors of the user browsing objects.
4. The browsing object recommendation method according to claim 3, wherein the time impact weight of each browsing object category characterized by said k browsing object center vectors is determined as follows:
the longer the interval time between the user browsing time and the preset comparison time is, the smaller the time influence weight of the browsing object category is;
the shorter the interval time between the user browsing time and the preset comparison time is, the larger the time influence weight of the browsing object category is.
5. The browsing object recommendation method according to any one of claims 1 to 4, wherein the recommending a browsing object to the target user based on the browsing object interest vector comprises:
determining each candidate browsing object to be recommended;
calculating the similarity of each candidate browsing object and the browsing object interest vector of the target user;
sequencing each candidate browsing object according to the similarity calculation result;
and recommending a preset number of candidate browsing objects ranked in the front to the user according to the ranking result.
6. The browsing object recommendation method according to any one of claims 1 to 4, further comprising, before determining the dense vector of each browsing object based on the reference data:
and extracting the reference data from the browsing records of the users.
7. The browsing object recommendation method according to any one of claims 1 to 4, wherein the determining a dense vector for each browsing object according to the reference data comprises:
and determining the reference data as input data of a depth migration algorithm, and determining a dense vector of each browsing object by using the depth migration algorithm.
8. A browsing object recommendation apparatus, comprising:
the first determination module is suitable for determining a dense vector of each browsing object according to reference data, wherein the reference data comprises object identifications of a plurality of browsing objects arranged according to a historical browsing sequence, and the dense vector at least describes a browsing incidence relation between the browsing object and other browsing objects;
the clustering module is suitable for clustering the dense vectors of the browsing objects to obtain k browsing object central vectors, each browsing object central vector represents a browsing object category, and k is a positive integer greater than 1;
a second determining module, adapted to determine a browsing object interest vector of a target user according to a browsing object browsing record of the target user and the k browsing object center vectors, where the browsing record of the target user is a browsing record of the target user within a period of time;
a recommending module, adapted to recommend a browsing object to the target user based on the browsing object interest vector; wherein the second determining module comprises:
the first determining submodule is suitable for determining a user browsing object browsed by the target user and user browsing time according to the browsing record of the target user;
the second determining submodule is suitable for determining a browsing object vector of the target user according to the user browsing object and the k browsing object central vectors, and determining a browsing object time sequence vector of the user browsing object according to the user browsing time and preset comparison time;
the weighting submodule is suitable for weighting the browsing object vector and the browsing object time sequence vector of the user to obtain a browsing object interest vector of the target user;
wherein the weighting submodule performs the steps of:
weighting the browsing object vector and the browsing object time sequence vector of the target user by using a preset harmonic factor, wherein the preset harmonic factor is greater than or equal to 0 and less than or equal to 1, and the preset harmonic factor is set according to the interest of the browsing object or the importance degree of the browsing time.
9. A storage medium having stored thereon computer instructions, wherein the computer instructions when executed perform the steps of the browsing object recommendation method of any one of claims 1 to 7.
10. A server comprising a memory and a processor, the memory having stored thereon computer instructions executable on the processor, wherein the processor executes the computer instructions to perform the steps of the browsing object recommendation method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811621490.1A CN109460519B (en) | 2018-12-28 | 2018-12-28 | Browsing object recommendation method and device, storage medium and server |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811621490.1A CN109460519B (en) | 2018-12-28 | 2018-12-28 | Browsing object recommendation method and device, storage medium and server |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109460519A CN109460519A (en) | 2019-03-12 |
| CN109460519B true CN109460519B (en) | 2021-07-06 |
Family
ID=65615309
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811621490.1A Active CN109460519B (en) | 2018-12-28 | 2018-12-28 | Browsing object recommendation method and device, storage medium and server |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109460519B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110851737B (en) * | 2019-11-13 | 2024-03-12 | 哈工大机器人湖州国际创新研究院 | Recommendation method, recommendation device, electronic equipment and computer storage medium |
| CN113704596A (en) * | 2020-05-21 | 2021-11-26 | 北京沃东天骏信息技术有限公司 | Method and apparatus for generating a set of recall information |
| CN111738754A (en) * | 2020-06-01 | 2020-10-02 | 上海晶赞融宣科技有限公司 | Object recommendation method and device, storage medium and computer equipment |
| CN113763080A (en) * | 2020-08-26 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method and device for determining recommended article, electronic equipment and storage medium |
| CN112732927A (en) * | 2020-12-31 | 2021-04-30 | 平安资产管理有限责任公司 | Content similarity analysis method and device based on knowledge graph |
| CN113452463A (en) * | 2021-06-09 | 2021-09-28 | 东风柳州汽车有限公司 | Intelligent pushing method, device and equipment for vehicle-mounted radio station programs and storage medium |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103838756A (en) * | 2012-11-23 | 2014-06-04 | 阿里巴巴集团控股有限公司 | Method and device for determining pushed information |
| CN104615779B (en) * | 2015-02-28 | 2017-08-11 | 云南大学 | A kind of Web text individuations recommend method |
| CN105022840B (en) * | 2015-08-18 | 2018-06-05 | 新华网股份有限公司 | A kind of news information processing method, news recommend method and relevant apparatus |
| CN105608111B (en) * | 2015-12-11 | 2019-02-15 | 中国银联股份有限公司 | To the method and system of terminal user's recommended |
-
2018
- 2018-12-28 CN CN201811621490.1A patent/CN109460519B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109460519A (en) | 2019-03-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109460519B (en) | Browsing object recommendation method and device, storage medium and server | |
| CN105224699B (en) | News recommendation method and device | |
| CN107424043B (en) | Product recommendation method and device and electronic equipment | |
| CN107220365B (en) | Accurate recommendation system and method based on collaborative filtering and association rule parallel processing | |
| CN108121737B (en) | Method, device and system for generating business object attribute identifier | |
| US10423648B2 (en) | Method, system, and computer readable medium for interest tag recommendation | |
| CN106919619B (en) | Commodity clustering method and device and electronic equipment | |
| CN111444395B (en) | Method, system and equipment for obtaining relation expression between entities and advertisement recall system | |
| CN108717407B (en) | Entity vector determination method and device, and information retrieval method and device | |
| CN107888950A (en) | A kind of method and system for recommending video | |
| CN107464132B (en) | Similar user mining method and device and electronic equipment | |
| CN112395487B (en) | Information recommendation method and device, computer readable storage medium and electronic equipment | |
| CN112052387B (en) | Content recommendation method, device and computer readable storage medium | |
| CN109409928A (en) | A kind of material recommended method, device, storage medium, terminal | |
| CN107832338B (en) | Method and system for recognizing core product words | |
| CA3059929C (en) | Text searching method, apparatus, and non-transitory computer-readable storage medium | |
| CN107908616B (en) | Method and device for predicting trend words | |
| US20210073890A1 (en) | Catalog-based image recommendations | |
| CN111444304A (en) | Search ranking method and device | |
| CN111460251A (en) | Data content personalized push cold start method, device, equipment and storage medium | |
| CN110532351A (en) | Recommend word methods of exhibiting, device, equipment and computer readable storage medium | |
| CN111400586A (en) | Group display method, terminal, server, system and storage medium | |
| CN106033455B (en) | Method and equipment for processing user operation information | |
| CN113837842A (en) | Commodity recommendation method and equipment based on user behavior data | |
| CN115496566B (en) | Regional specialty recommendation method and system based on big data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |