CN116340610A - Real-time recommendation method and device - Google Patents
Real-time recommendation method and device Download PDFInfo
- Publication number
- CN116340610A CN116340610A CN202111598697.3A CN202111598697A CN116340610A CN 116340610 A CN116340610 A CN 116340610A CN 202111598697 A CN202111598697 A CN 202111598697A CN 116340610 A CN116340610 A CN 116340610A
- Authority
- CN
- China
- Prior art keywords
- time
- real
- user
- behavior
- preference
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000012163 sequencing technique Methods 0.000 claims abstract description 4
- 230000006399 behavior Effects 0.000 claims description 104
- 239000013598 vector Substances 0.000 claims description 32
- 230000003542 behavioural effect Effects 0.000 claims description 12
- 230000003993 interaction Effects 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 9
- 235000019633 pungent taste Nutrition 0.000 claims 1
- 238000004458 analytical method Methods 0.000 description 10
- 230000002452 interceptive effect Effects 0.000 description 8
- 230000000670 limiting effect Effects 0.000 description 7
- 239000011159 matrix material Substances 0.000 description 7
- 238000012545 processing Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 239000000284 extract Substances 0.000 description 5
- 238000001914 filtration Methods 0.000 description 4
- 238000003058 natural language processing Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000000354 decomposition reaction Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 239000002245 particle Substances 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000002829 reductive effect Effects 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 101100256916 Caenorhabditis elegans sid-1 gene Proteins 0.000 description 1
- 101100256918 Caenorhabditis elegans sid-2 gene Proteins 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000002238 attenuated effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 238000002790 cross-validation Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 239000004984 smart glass Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9538—Presentation of query results
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a real-time recommendation method and device and a computer readable storage medium. The real-time recommendation method comprises the following steps: acquiring behavior data of at least one user on at least one behavior object; determining the real-time preference score of each user on each behavior object according to the behavior data and the time difference from the release time of each behavior object to the current time; ranking each of the real-time preference scores to determine at least one real-time preference object for each of the users; counting total scores of the real-time preference scores of the behavior objects, and sequencing the total scores to determine a popular recommendation list; and matching each behavior object in the popular recommendation list with each real-time preference object of each user respectively so as to determine recommended content of each user.
Description
Technical Field
The present invention relates to a recommendation technology of application content, and in particular, to a real-time recommendation method, a real-time recommendation device, and a computer readable storage medium.
Background
The recommendation system based on the Internet of vehicles (Internet of Vehicle, IOV) can acquire operation data of a user about vehicle-mounted applications through the Internet of vehicles, and recommend application contents of interest to the user according to the operation data. The prior art of the Internet of vehicles recommendation system mainly carries out offline analysis based on a recall ordering algorithm according to buried point behavior data of a user in a target application, constructs an input matrix of a collaborative filtering algorithm according to the behavior data of the user in a recall stage, recalls preference data of the user according to similar users and similar preferences, and constructs a logistic regression algorithm model according to recommended feedback results in the ordering stage to obtain preference data of most interest of the user.
However, this offline analysis model based on recall ordering algorithm involves a large amount of data processing load, and is long in running time, and can only be updated once a day at present. Although the update frequency can meet the application requirements of music, radio stations, oiling reminding and the like with low real-time requirements, the update frequency can not meet the requirements of applications with relatively quick change frequencies such as news, videos and the like on the real-time performance of recommended contents.
In order to overcome the above-mentioned drawbacks of the prior art, there is a need in the art for a recommendation technique for application content for analyzing real-time preferences of users and recommending the content of most interest to the users.
Disclosure of Invention
The following presents a simplified summary of one or more aspects in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that is presented later.
In order to overcome the above-mentioned drawbacks of the prior art, the present invention provides a real-time recommendation method, a real-time recommendation device, and a computer-readable storage medium.
Specifically, the real-time recommendation method provided according to the first aspect of the present invention includes the following steps: acquiring behavior data of at least one user on at least one behavior object; determining the real-time preference score of each user on each behavior object according to the behavior data and the time difference from the release time of each behavior object to the current time; ranking each of the real-time preference scores to determine at least one real-time preference object for each of the users; counting total scores of the real-time preference scores of the behavior objects, and sequencing the total scores to determine a popular recommendation list; and matching each behavior object in the popular recommendation list with each real-time preference object of each user respectively so as to determine recommended content of each user. By executing the steps, the real-time recommendation method can analyze the real-time preference of the user to determine the popular content of the target application, and recommend the content which is most interesting to the user currently to the user by combining the real-time preference of the user and the popular content of the target application.
In addition, the real-time recommendation device provided in the second aspect of the present invention includes a memory and a processor. The processor is connected to the memory and configured to implement the above-mentioned real-time recommendation method provided by the first aspect of the present invention. By implementing the real-time recommendation method, the real-time recommendation device can analyze the real-time preference of the user to determine the popular content of the target application, and recommend the content which is most interesting to the user currently to the user by combining the real-time preference of the user and the popular content of the target application.
Further, the above-described computer-readable storage medium according to the third aspect of the present invention stores thereon the instructions by the computer. The computer instructions, when executed by a processor, implement the above-described real-time recommendation method provided by the first aspect of the present invention. By implementing the real-time recommendation method, the computer-readable storage medium can analyze the real-time preferences of the user to determine popular content of the target application, and recommend the content of most interest to the user in combination with the real-time preferences of the user and the popular content of the target application.
Drawings
The above features and advantages of the present invention will be better understood after reading the detailed description of embodiments of the present disclosure in conjunction with the following drawings. In the drawings, the components are not necessarily to scale and components having similar related features or characteristics may have the same or similar reference numerals.
Fig. 1 illustrates an architecture diagram of a real-time recommendation device provided according to some embodiments of the present invention.
Fig. 2 illustrates a flow diagram of a real-time recommendation method provided in accordance with some embodiments of the present invention.
Fig. 3 illustrates a flow diagram for determining similar users provided in accordance with some embodiments of the invention.
Fig. 4 illustrates a flow diagram for determining similar content provided in accordance with some embodiments of the invention.
Detailed Description
Further advantages and effects of the present invention will become apparent to those skilled in the art from the disclosure of the present specification, by describing the embodiments of the present invention with specific examples. While the description of the invention will be presented in connection with a preferred embodiment, it is not intended to limit the inventive features to that embodiment. Rather, the purpose of the invention described in connection with the embodiments is to cover other alternatives or modifications, which may be extended by the claims based on the invention. The following description contains many specific details for the purpose of providing a thorough understanding of the present invention. The invention may be practiced without these specific details. Furthermore, some specific details are omitted from the description in order to avoid obscuring the invention.
In the description of the present invention, it should be noted that, unless explicitly specified and limited otherwise, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be either fixedly connected, detachably connected, or integrally connected, for example; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the above terms in the present invention will be understood in specific cases by those of ordinary skill in the art.
In addition, the terms "upper", "lower", "left", "right", "top", "bottom", "horizontal", "vertical" as used in the following description should be understood as referring to the orientation depicted in this paragraph and the associated drawings. This relative terminology is for convenience only and is not intended to be limiting of the invention as it is described in terms of the apparatus being manufactured or operated in a particular orientation.
It will be understood that, although the terms "first," "second," "third," etc. may be used herein to describe various elements, regions, layers and/or sections, these elements, regions, layers and/or sections should not be limited by these terms and these terms are merely used to distinguish between different elements, regions, layers and/or sections. Accordingly, a first component, region, layer, and/or section discussed below could be termed a second component, region, layer, and/or section without departing from some embodiments of the present invention.
As described above, the offline analysis model based on the recall ordering algorithm involves a large amount of data processing load, and has long running time, and can only be updated once a day at present. Although the update frequency can meet the application requirements of music, radio stations, oiling reminding and the like with low real-time requirements, the update frequency can not meet the requirements of applications with relatively quick change frequencies such as news, videos and the like on the real-time performance of recommended contents.
In order to overcome the above-mentioned drawbacks of the prior art, the present invention provides a real-time recommendation method, a real-time recommendation device, and a computer readable storage medium, which can analyze the real-time preference of a user to determine the popular content of a target application, and recommend the content of most interest to the user in combination with the real-time preference of the user and the popular content of the target application.
In some non-limiting embodiments, the above-mentioned real-time recommendation method provided by the first aspect of the present invention may be implemented by the above-mentioned real-time recommendation device provided by the second aspect of the present invention. The real-time recommending device can be configured on the server side of the target application in the form of hardware equipment and/or software programs. The target applications include, but are not limited to, various applications (applications) such as news applications, video applications, etc., applet (Applet) and Web page program (Web App) running in various user terminals such as car set, mobile phone, tablet computer, palm top computer (Personal Digital Assistant, PDA), notebook computer, personal computer (Personal Computer, PC), smart watch, smart bracelet, smart glasses, etc.
Further, please refer to fig. 1. Fig. 1 illustrates an architecture diagram of a real-time recommendation device provided according to some embodiments of the present invention.
As shown in fig. 1, the real-time recommendation device 10 according to the second aspect of the present invention is provided with a memory 11 and a processor 12. The memory 11 includes, but is not limited to, the above-described computer-readable storage medium provided by the third aspect of the present invention, on which computer instructions are stored. The processor 12 is connected to the memory 11 and is configured to execute computer instructions stored on the memory 11 to implement the above-described real-time recommendation method provided by the first aspect of the present invention.
The operation of the real-time recommendation device 10 described above will be described below in connection with a real-time recommendation method for some car news applications. It will be appreciated by those skilled in the art that these real-time recommendation methods for news applications are merely some non-limiting embodiments provided by the present invention, and are intended to clearly illustrate the general concepts of the present invention and to provide some embodiments that are convenient to the public, rather than for limiting the kinds of target applications described above, or for limiting the overall functionality or overall operation of the real-time recommendation device 10 described above. Similarly, the real-time recommendation device 10 is just one non-limiting embodiment provided by the present invention, and does not limit the execution subject of each step in the real-time recommendation method.
Referring to fig. 2, fig. 2 is a flow chart illustrating a real-time recommendation method according to some embodiments of the invention.
As shown in fig. 2, in the process of implementing the real-time recommendation method, the real-time recommendation device 10 may first acquire behavior data B of at least one user i on at least one behavior object j ij 。
In some embodiments of the present invention, a buried point may be preset in a news application running on a car systemProgram for collecting behavior data B of user i on news j in news application ij . Further, the behavior data B ij Preference data input by a user when using the news application and object data corresponding to each operation instruction may be included. The object data is index information for each news in the news application including, but not limited to, content tags, titles, and/or index numbers for each news. The preference data includes, but is not limited to, at least one operational record of searching, browsing, subscribing, collecting, praying, clicking, barrage, uninteresting operations, indicating the personal preference of the user for each news.
Responsive to collecting news application-related behavior data B ij The embedded point program provides the data to the vehicle system, and the vehicle system uploads the data to the real-time recommendation device 10 at the news application server through Kafka software. The real-time recommendation device 10 may use a link stream processing framework to stream the behavior data uploaded by each user terminal, consume the data uploaded by the Kafka software in real time, and clean the collected data to screen dirty data with null values and abnormal values, while only retaining the embedded point data B, such as user i search news j, user i browse news j, user i subscription news j, user i collection news j, user i praise news j, user i point treading news j, user i in news j bullet screen, user i is not interested in news j, etc., where the user i is suitable for user preference analysis ij Thereby improving the accuracy of the real-time recommendation result.
As shown in fig. 2, the behavior data B of at least one user i on at least one behavior object j is acquired ij The real-time recommendation device 10 can then calculate the behavior data B based on the acquired behavior data ij And the release time t of each behavior object j j0 Time difference Δt to current time t j Determining the real-time preference score S of each user i on each behavior object j ij 。
In some embodiments, the real-time recommendation device 10 may first determine the initial heat score S of the corresponding behavior object j based on the acquired object data 0j . Specifically, the real-time recommendation device 10 may first obtain the behavior data B from the acquired behavior data ij Chinese extraction fingerObject data such as a content tag (label) of the behavior object j is displayed, and whether the behavior object j is used for the first time by the user i is determined based on the extracted object data. If the behavior object j is used by the user i for the first time, the real-time recommendation device 10 can determine the object class to which the behavior object j belongs according to the content tag, and determine the initial popularity score S of the behavior object j according to the object class to which the behavior object j belongs 0j 。
Referring to Table 1, table 1 shows an initial heat score table provided in accordance with some embodiments of the invention. As shown in table 1, in some embodiments, each news j in a news application may be divided into multiple categories of sports, entertainment, finance, internationally, socially, cultural, weather, etc., according to the domain, wherein each category is configured with a corresponding initial heat score according to how much users of the wide news application prefer for its entirety. The real-time recommendation device 10 may query the initial heat score table shown in table 1 according to the object category to which the behavior object j belongs to determine the initial heat score S of the behavior object j 0j 。
TABLE 1
| Object class | Initial heat score |
| Sports | 1.5 |
| Entertainment device | 1.5 |
| Finance and economics | 1.2 |
| International | 1.2 |
| Society | 1.2 |
| Culture | 0.8 |
| Weather of | 0.6 |
In determining the initial heat score S of the behavior object j 0j The real-time recommendation device 10 can then calculate the release time t of the behavior object j j0 Time difference Δt to current time t j ＝t-t j0 Determining a time decay heat score S of a behavioral object j tj . Specifically, the real-time recommendation device 10 may compare the time difference Δt j Substitution of a time decay function based on Newton's cooling coefficient
To determine a time decay heat score S for a behavioural object j tj 。
In determining the initial heat score S of the behavior object j 0j Time decay heat score S tj The real-time recommendation device 10 may then differencing the two to determine the real-time preference score S of user i for behavior object j ij ＝S 0j -S tj . By introducing the initial heat score S 0j The invention can effectively solve the problem of cold start of the real-time recommending function. Even if the user i uses the behavior object j for the first time (i.e. lacks corresponding preference data in the recommendation system), the real-time recommendation device 10 may assign an initial score to the behavior object j in combination with the overall preference of other users for news in the field as a data base for real-time recommendation. By introducing the time-decay heat score S tj The invention can further combine the characteristic that the news heat can be attenuated with time, thereby further improvingAccurately characterizing the real-time preferences of the user.
Further, in some embodiments of the present invention, the real-time recommendation device 10 may also obtain behavior data B from the acquired behavior data B ij Further extracting preference data indicating at least one operation of searching, browsing, subscribing, collecting, praying, clicking, barrage, uninteresting and the like of the behavior object j by the user i, and determining the interaction heat score S of the behavior object j according to the preference data uj Combining the interactive heat score S uj To determine the real-time preference score S of user i for behavior object j ij 。
Referring to Table 2, table 2 illustrates an interactive heat score table provided in accordance with some embodiments of the invention. As shown in table 2, in some embodiments, the operations of searching, browsing, subscribing, collecting, praying, clicking, barrage, uninteresting, and the like may each correspond to a positive score or a negative score, where the positive score indicates that the user i likes the corresponding news j, the negative score indicates that the user i dislikes the corresponding news j, and the absolute value of the score indicates the likeness/dislikeness of the news j corresponding to the user i.
TABLE 2
| Interactive operation | Interactive heat score |
| Searching | 3 |
| Browsing | 1 |
| Collecting and storing | 4 |
| Praise to be praise | 4 |
| Point stepping | -4 |
| Bullet screen | 2 |
| Not of interest | -4 |
The real-time recommendation device 10 can count one or more operation records of the user i on each news j, and respectively determine the interaction heat score S of each behavior object according to the score of the user i on each operation record of each behavior object j uj ＝3*s Searching +1*s Browsing +4*s Collecting and storing +4*s Praise to be praise -4*s Point stepping +2*s Bullet screen -4*s Not of interest 。
The real-time recommendation device 10 may then incorporate the initial heat score S 0j Interactive heat score S uj Time decay heat score S tj To determine the real-time preference score S of user i for behavior object j ij ＝S 0j +S uj -S tj . By further introducing the above-mentioned interactive heat score S uj The method and the device can further combine the characteristic that news heat can change along with the behavior of the user, so that the real-time preference of the user can be more accurately represented.
Those skilled in the art will appreciate that the above described combination initial heat score S 0j Interactive heat score S uj Time decay heat score S tj To determine real-time preference scores S ij The present invention provides a non-limiting embodiment only, and is intended to clearly illustrate the main concept of the present invention and to provide some specific embodiments for public implementation, not to limit the scope of the present invention.
Alternatively, in other embodiments, if the determination indicates that the behavioral object j is not being used by the user i for the first time, the real-time recommendation device 10 may skip the determining the initial heat score S 0j Directly according to the interactive heat score S uj Time decay heat score S tj To determine real-time preference scores S ij Thereby eliminating the initial heat score S based on the overall preference degree 0j The influence of the recommendation result is improved, the individuation and the accuracy of the recommendation result are improved, the data processing load of the real-time recommendation process is reduced, and the instantaneity of the recommendation result is improved.
As shown in FIG. 2, in determining the real-time preference score S of user i for each behavior object j ij Thereafter, the real-time recommendation device 10 may compare the real-time preference scores S of the behavior objects j ij Ordering and constructing the real-time preference list P of user i from at least one (e.g., 10) behavior object j with higher score i . The real-time preference list P i Containing at least one real-time preference object m for user i. The real-time recommendation device 10 may then also calculate the real-time preference score S based on the user ID, news ID, and real-time preference score S ij Constructing preference score field "uid (i), sid (j), score (S) ij ) ", it is stored in the Hbase database and its row key (rowkey) is set to reverse (uid) as the data base for the real-time recommendation function.
In addition, in determining the real-time preference score S of each user i for each behavior object j ij The real-time recommendation device 10 can also count the total score S of the real-time preference scores of the behavior objects j j ＝∑S ij And for each total point S j Ordering is performed to determine at least one (e.g., 50) behavior object j in which the total score is higher. The real-time recommendation device 10 may then determine the current hot recommendation list R for the news application with the higher total score of at least one (e.g., 50) behavioral object j.
Then, the real-time recommendation device 10 may match each behavior object n in the popular recommendation list R with each real-time preference object m of each user i, so as to determine the recommended content of each user i.
For example, the real-time recommendation device 10 may first extract at least one keyword KW from the content of each behavior object n of the popular recommendation list R based on the Jieba chinese word segmentation component n And extracts at least one keyword KW from the content of each real-time preference object m of user i m . The real-time recommendation device 10 may then individually select the keyword sets KW n And each keyword set KW m Performing intersection operation and counting keywords kw in each intersection nm Is a number of (3). Still further, the real-time recommendation device 10 may select the keyword kw from each intersection nm Ordering the behavior objects n in the popular recommendation list R, and determining at least one (e.g. 10) behavior object n with a large keyword intersection number as at least one recommendation object R of the user i 1 。
The method and the device for recommending the content of the target application determine the popular recommendation list R of the target application by analyzing the real-time preferences of a plurality of users, and recommend the content which is most interesting currently to the user i by combining the real-time preferences of the user i and the popular recommendation list R of the target application.
In addition, compared with the scheme that the existing offline analysis model only carries out content recommendation based on historical behavior data of a previous period of time (for example: 30 days), the method can further combine all the historical behavior data of the user and calculate the time decay heat score S tj Characterizing the decay characteristic of content heat over time, thus enabling more historical behavioral data to be utilized to provide at least one recommendation r that better meets the personalized needs of user i 1 。
Further, in some embodiments of the present invention, for target applications with high real-time requirements, such as news applications and video applications, the real-time recommendation device 10 may preferably record updated contents of the target applications and construct the target applications with the updated contents in response to completing one content recommendationNew content list N. Thereafter, the real-time recommending apparatus 10 may further acquire a new content list N and extract at least one candidate o included therein at the time of the next content recommendation. Then, the real-time recommendation device 10 may match each candidate object o in the new content recommendation list N with each real-time preference object m of the user i as described above, and rank each candidate object o in the new content list N according to the number of keyword intersections to determine at least one recommendation object r of the user i 2 。
By further matching each real-time preference object m of the user i with each candidate object o in the new content recommendation list N, the method and the device can further combine the new content referenced by the target to conduct content recommendation, so that the real-time performance of the recommended content can be further improved.
Still further, in some embodiments of the present invention, the real-time recommendation device 10 may further determine a similar user u of the user i based on the near real-time collaborative filtering algorithm model, and determine at least one recommendation object r of the user i according to at least one real-time preference object v of the similar user u 3 。
Referring to fig. 3, fig. 3 is a flow chart illustrating a method for determining similar users according to some embodiments of the invention.
As shown in fig. 3, in determining the similar user u of the user i, the real-time recommendation device 10 may first obtain preference data of at least one operation of searching, browsing, subscribing, collecting, praying, clicking, barrage, uninteresting, etc. on each behavior object j by the user i and the multiple candidate users w, and determine the interaction heat score S of each behavior object j by each user i, w according to the preference data as described above uij S and S uwj . The real-time recommendation device 10 can then calculate the interaction heat score S of each user i, w for each behavior object j uij S and S uwj Construction of scoring matrices for user-news, i.e
Still further, the real-time recommender 10 may use a pre-built Spark-based ALS matrix decomposition algorithm model, scoring matrix S of the user-news according to algorithm parameters set by cross-validation algorithm u Matrix decomposition is performed so as to obtain the interaction heat scores S of the user i on the behavior objects j uij Determining a first behavioral feature vector BF for user i 1 And according to the interaction heat scores S of the candidate users w on the behavior objects j uwj Determining a second behavior feature vector BF of each candidate user w 2w 。
After that, the real-time recommendation device 10 can calculate the first behavior feature vectors BF 1 And each second behavior feature vector BF 2w Cosine similarity cos iw And according to the user ID, the candidate user ID and cosine similarity cos iw Constructing similarity field "ui 1 (i), ui 2 (w), score (cos) iw ) ", it is stored in Hbase database as a data basis for determining similar users.
Still further, at least one recommendation object r for user i is determined based on at least one real-time preference object v for similar user u 3 In this case, the real-time recommendation device 10 can directly query the corresponding cosine similarity cos according to the user ID of the user i and the user IDs of the candidate users w iw And according to cosine similarity cos iw Ordering the candidate users w to obtain cosine similarity cos iw The higher at least one (e.g., 10) candidate users w are determined to be similar users u to user i. Thereafter, the real-time recommendation device 10 can acquire the real-time preference list P of each similar user u u And randomly acquire one or more real-time preference objects v therefrom as at least one recommendation object r of the user i 3 。
Compared with an offline analysis model based on a recall ordering algorithm, the user-news scoring matrix S is filled with the matrix-decomposed user feature vectors and content feature vectors u In order to determine the proposal of the recommended object, the invention directly calculates the first behavior feature vector BF after matrix decomposition 1 And each second behavior feature vector BF 2w The cosine distance between the two candidate users is calculated to calculate the similarity between the user i and each candidate user w, so that the real time can be greatly reducedThe data processing load of the device 10 is recommended so as to meet the real-time requirement of more than 1500 updated contents per day for news applications. Further, compared with the bias analysis scheme for determining similar users based on the basic information of the users, the bias analysis method and the bias analysis device can calculate the similarity based on the historical behavior data of the user i and each candidate user w, so that the true preference of the users can be more accurately represented, and the content which meets the personalized requirements of the users can be recommended to the users.
Furthermore, in some embodiments of the present invention, the real-time recommendation device 10 may further determine at least one similar content c of each real-time preference object m of the user i based on the near real-time collaborative filtering algorithm model, and determine the at least one similar content c as the at least one recommendation object r of the user i 4 。
Referring to fig. 4, fig. 4 is a flow chart illustrating a method for determining similar content according to some embodiments of the invention.
As shown in fig. 4, in determining the similar content c of the real-time preference object m, the real-time recommendation apparatus 10 may extract at least one (e.g., 6) keywords from the content of the real-time preference object m based on the Jieba chinese word segmentation component using the NLP natural language processing algorithm to construct a first word vector WF of the real-time preference object m 1 . In some embodiments, the first word vector WF 1 The dimension of (c) may be determined according to the number of all keywords (e.g., 50) contained in the database of the news application, where only the dimension in which the 6 extracted keywords are located has a valid value, and the remaining keyword dimensions are all 0.
In addition, the real-time recommendation device 10 may acquire the full content list a containing all news from the database of the news application, and extract at least one (e.g., 6) keywords from the contents of each candidate object a of the full content list a, respectively, as described above, to construct the second word vector WF of each candidate object a 2a . In some embodiments, the second word vector WF 2a May also be determined based on the number of all keywords contained in the database of the news application (e.g., 50), where only the dimensions in which the 6 extracted keywords resideWith valid values and the remaining keyword dimensions are all 0.
The real-time recommendation device 10 may then calculate the first word vectors WF 1 And each second word vector WF 2a Cosine similarity cos ma And according to the real-time preference object ID, the candidate object ID and the cosine similarity cos ma Constructing similarity field "sid1 (m), sid2 (a), score (cos) ma ) ", it is stored in Hbase database as the data base for determining similar content.
Still further, at least one recommendation object r for user i is determined based on the similar content c 4 In this case, the real-time recommendation device 10 can directly query the corresponding cosine similarity cos according to the object ID of each real-time preference object m and the object ID of each candidate object a of the user i ma And according to cosine similarity cos ma Ordering the candidates a to obtain cosine similarity cos ma The higher at least one (e.g., 10) candidate object a is determined as the similar content c of each real-time preference object m of user i. Thereafter, the real-time recommendation device 10 may determine the at least one similar content c as at least one recommendation object r of the user i 4 。
Further, in some embodiments of the invention, the above-mentioned computed cosine similarity cos ma And may preferably be performed based on the importance of each keyword in the corresponding content. Specifically, after extracting at least one (e.g., 6) keywords from the content of the real-time preference object m using the NLP natural language processing algorithm, the real-time recommender 10 may calculate the importance scores (i.e., TF-IDF values) of the keywords using the TF-IDF algorithm, and construct the first word vector WF of the real-time preference object m based on the importance scores of the keywords 1 '. Similarly, after extracting at least one (e.g., 6) keywords from the content of each candidate object a, the real-time recommendation apparatus 10 may calculate the importance scores (i.e., TF-IDF values) of the keywords by TF-IDF algorithm, and construct the second word vector WF of each candidate object a based on the importance scores of the keywords 2a '. Thereafter, real-time recommendationThe apparatus 10 may calculate the first word vectors WF separately as described above 1 ' and respective second word vectors WF 2a ' cosine similarity cos ma ' and according to similarity cos ma ' order the candidates a in the full content list A to rank the cosine similarity cos therein ma ' higher at least one (e.g., 10) candidate objects a are determined as at least one recommended object r of user i 4 。
Compared with an offline analysis model based on a recall ordering algorithm, the user-news scoring matrix S is filled with the matrix-decomposed user feature vectors and content feature vectors u In order to determine the scheme of the recommended objects, the invention directly calculates the similarity between each candidate object a in the full content list A and each real-time preference object m of the user i, and determines at least one recommended object r of the user i according to at least one similar content c of each real-time preference object m 4 Therefore, the data processing load of the real-time recommendation device 10 can be greatly reduced, and the real-time requirement of the news application on the updated content of more than 1500 pieces per day can be met.
Further, in some embodiments of the present invention, at least one recommendation object R is determined from the hot recommendation list R 1 At least one recommendation object r determined from the new content list N 2 At least one recommendation object r determined from the full content list A 4 And/or determining at least one recommendation object r from at least one real-time preference object v of a similar user u 3 The real-time recommendation device 10 can then recommend each recommendation object r 1 ～r 4 And taking the union set and performing a de-duplication operation to construct a recommended content list of the user i. Thereafter, the real-time recommendation device 10 may recommend at least one recommendation object among them to the user i according to the recommendation content list.
By adopting the plurality of algorithm models to determine the recommended object of the user i, the invention can apply the quasi-real-time collaborative filtering algorithm to the real-time recommendation system on one hand, thereby meeting the real-time requirements of various target applications such as news applications, video applications and the like, and can integrate various factors such as real-time preferences of users, similar contents, popular contents, new contents and the like on the other hand, thereby improving the accuracy of the recommended contents and enabling the recommended contents to better meet the personalized requirements of the users.
While, for purposes of simplicity of explanation, the methodologies are shown and described as a series of acts, it is to be understood and appreciated that the methodologies are not limited by the order of acts, as some acts may, in accordance with one or more embodiments, occur in different orders and/or concurrently with other acts from that shown and described herein or not shown and described herein, as would be understood and appreciated by those skilled in the art.
Those of skill in the art would understand that information, signals, and data may be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips that may be referenced throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof.
Those of skill would further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
Although the real-time recommendation device 10 described in the above embodiment may be implemented by a combination of software and hardware. It will be appreciated that the real-time recommendation device 10 may also be implemented in software or hardware alone. For hardware implementation, the real-time recommendation device 10 may be implemented in one or more Application Specific Integrated Circuits (ASICs), digital Signal Processors (DSPs), programmable Logic Devices (PLDs), field Programmable Gate Arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, other electronic devices for performing the functions described above, or a selected combination of the above. For software implementation, the real-time recommendation device 10 may be implemented by separate software modules, such as program modules (procedures) and function modules (functions), running on a common chip, each of which performs one or more of the functions and operations described herein.
The various illustrative logical modules, and circuits described in connection with the embodiments disclosed herein may be implemented or performed with a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.
The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples and designs described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (16)
1. A real-time recommendation method, comprising the steps of:
acquiring behavior data of at least one user on at least one behavior object;
determining the real-time preference score of each user on each behavior object according to the behavior data and the time difference from the release time of each behavior object to the current time;
ranking each of the real-time preference scores to determine at least one real-time preference object for each of the users;
counting total scores of the real-time preference scores of the behavior objects, and sequencing the total scores to determine a popular recommendation list; and
and matching each behavior object in the popular recommendation list with each real-time preference object of each user respectively to determine recommended content of each user.
2. The real-time recommendation method of claim 1, wherein the behavior data includes object data, and the step of determining the real-time preference score of each of the users for the behavior object according to the behavior data and a time difference of a release time to a current time of each of the behavior objects includes:
determining initial heat scores of corresponding behavior objects according to the object data;
determining a time decay heat score of the behavior object according to the time difference from the release time of the behavior object to the current time; and
and determining the real-time preference score of the user on the behavior object according to the initial heat score and the time decay heat score.
3. The real-time recommendation method of claim 2, wherein the behavior data further includes preference data, and the step of determining the real-time preference score of each of the users for the behavior object according to the behavior data and a time difference of a release time to a current time of each of the behavior objects further includes: determining an interaction heat score of the behavioral object based on the preference data,
the step of determining the real-time preference score of the user for the behavior object according to the initial heat score and the time decay heat score comprises the following steps: and determining the real-time preference score of the user on the behavior object according to the initial heat score, the interaction heat score and the time decay heat score.
4. The real-time recommendation method of claim 3, wherein the preference data comprises at least one operation record of searching, browsing, subscribing, collecting, praying, clicking, barrage, uninteresting operations, wherein each operation record corresponds to a score, and the step of determining the interaction heat score of the behavioral object according to the preference data comprises:
and determining the interaction heat score of the behavior object according to the scores of the operation records of the user on the behavior object.
5. The real-time recommendation method of claim 2, wherein the step of determining initial hotness scores for corresponding behavioral objects from the object data comprises:
determining an object category of the behavior object according to the object data; and
and determining the initial heat score of the behavior object according to the object category.
6. The real-time recommendation method of claim 2, wherein the step of determining the time decay heat score of the behavioral object according to the time difference of the publication time to the current time of the behavioral object comprises:
substituting the time difference into a time decay function to determine a time decay heat score for the behavioral object.
7. The real-time recommendation method of claim 1, wherein the step of matching each of the behavior objects in the popular recommendation list with each of the real-time preference objects of the users, respectively, to determine recommended content of each of the users comprises:
performing keyword matching on each behavior object in the popular recommendation list and at least one real-time preference object of a user respectively; and
and sorting all the behavior objects in the hot recommendation list according to the number of keyword intersections so as to determine at least one recommendation object of the user.
8. The real-time recommendation method of claim 1, further comprising the steps of:
acquiring a new content list, wherein the new content list comprises at least one candidate object;
performing keyword matching on each candidate object in the new content recommendation list and at least one real-time preference object of a user respectively; and
and sorting the candidate objects in the new content list according to the number of keyword intersections to determine at least one recommended object of the user.
9. The real-time recommendation method of claim 1, further comprising the steps of:
acquiring a full content list, wherein the full content list comprises at least one candidate object;
respectively matching each candidate object in the full-content recommendation list with at least one real-time preference object of a user in a similarity manner; and
and sorting the candidate objects in the full content list according to the similarity to determine at least one recommended object of the user.
10. The real-time recommendation method of claim 9, wherein the step of matching each of the candidate objects in the full content recommendation list with at least one real-time preference object of a user, respectively, in similarity comprises:
extracting a plurality of keywords from the real-time preference object to construct a first word vector of the real-time preference object;
extracting a plurality of keywords from each candidate object in the full content list respectively to construct a second word vector of each candidate object respectively; and
and respectively calculating cosine similarity of the first word vector and each second word vector.
11. The real-time recommendation method of claim 10, wherein the step of extracting a plurality of keywords from the real-time preference object to construct a first word vector of the real-time preference object comprises: extracting a plurality of keywords from the real-time preference object; calculating importance scores of the keywords through a TF-IDF algorithm; and constructing a first word vector of the real-time preference object according to the importance scores of the keywords,
the step of respectively extracting a plurality of keywords from each candidate object in the full content list to respectively construct a second word vector of each candidate object comprises the following steps: extracting a plurality of keywords from each candidate object respectively; calculating importance scores of the keywords through a TF-IDF algorithm; and respectively constructing second word vectors of the candidate objects according to the importance scores of the keywords.
12. The real-time recommendation method of claim 1, further comprising the steps of:
determining similar users of the users; and
at least one recommended object of the user is determined according to at least one real-time preference object of the similar user.
13. The real-time recommendation method of claim 12, wherein the step of determining similar users of the user comprises:
determining a first behavior feature vector of the user according to the interaction heat scores of the user on the behavior objects;
determining a second behavior feature vector of each candidate user according to the interaction heat scores of the candidate users on each behavior object;
respectively calculating cosine similarity of the first behavior feature vector and each second behavior feature vector; and
and sequencing the candidate users according to the cosine similarity to determine at least one similar user of the users.
14. The real-time recommendation method of claim 6, further comprising the steps of:
merging at least one recommended object determined from the popular recommended list, at least one recommended object determined from a new content list, at least one recommended object determined from a full content list, and/or at least one recommended object determined according to a similar user, and performing a deduplication operation to construct a recommended content list of the user; and
and recommending at least one recommended object to the user according to the recommended content list.
15. A real-time recommendation device, comprising:
a memory; and
a processor connected to the memory and configured to implement the real-time recommendation method according to any one of claims 1 to 14.
16. A computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the real-time recommendation method according to any one of claims 1 to 14.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111598697.3A CN116340610A (en) | 2021-12-24 | 2021-12-24 | Real-time recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111598697.3A CN116340610A (en) | 2021-12-24 | 2021-12-24 | Real-time recommendation method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116340610A true CN116340610A (en) | 2023-06-27 |
Family
ID=86890187
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111598697.3A Pending CN116340610A (en) | 2021-12-24 | 2021-12-24 | Real-time recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116340610A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116546091A (en) * | 2023-07-07 | 2023-08-04 | 深圳市四格互联信息技术有限公司 | Recommendation method, device, equipment and storage medium of streaming content |
-
2021
- 2021-12-24 CN CN202111598697.3A patent/CN116340610A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116546091A (en) * | 2023-07-07 | 2023-08-04 | 深圳市四格互联信息技术有限公司 | Recommendation method, device, equipment and storage medium of streaming content |
| CN116546091B (en) * | 2023-07-07 | 2023-11-28 | 深圳市四格互联信息技术有限公司 | Recommendation method, device, equipment and storage medium of streaming content |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chu et al. | A hybrid recommendation system considering visual information for predicting favorite restaurants | |
| CN107944913B (en) | High-potential user purchase intention prediction method based on big data user behavior analysis | |
| CN108009228B (en) | Method and device for setting content label and storage medium | |
| US11907240B2 (en) | Method and system for presenting a search result in a search result card | |
| Tso-Sutter et al. | Tag-aware recommender systems by fusion of collaborative filtering algorithms | |
| US10572565B2 (en) | User behavior models based on source domain | |
| CN110704674B (en) | Video playing integrity prediction method and device | |
| CN110019943B (en) | Video recommendation method and device, electronic equipment and storage medium | |
| Košir et al. | Database for contextual personalization | |
| US20160171116A1 (en) | System and method for crowdsourced template based search | |
| CN109558535B (en) | Personalized article pushing method and system based on face recognition | |
| CN103559206B (en) | A kind of information recommendation method and system | |
| CN104991966B (en) | Media content ordering method and system | |
| US20200226133A1 (en) | Knowledge map building system and method | |
| WO2015188699A1 (en) | Item recommendation method and device | |
| CN107562939B (en) | Vertical domain news recommendation method and device and readable storage medium | |
| US20130036121A1 (en) | System and method for recommending blog | |
| WO2015034850A2 (en) | Feature selection for recommender systems | |
| CN109511015B (en) | Multimedia resource recommendation method, device, storage medium and equipment | |
| Jin et al. | Reinforced similarity integration in image-rich information networks | |
| CN106383887A (en) | Environment-friendly news data acquisition and recommendation display method and system | |
| CN112052387B (en) | Content recommendation method, device and computer readable storage medium | |
| US20170199930A1 (en) | Systems Methods Devices Circuits and Associated Computer Executable Code for Taste Profiling of Internet Users | |
| CN106599047B (en) | Information pushing method and device | |
| CN111506831A (en) | Collaborative filtering recommendation module and method, electronic device and storage medium |
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 |