CN108419134B - Channel recommendation method based on fusion of individual history and group current behaviors - Google Patents

Channel recommendation method based on fusion of individual history and group current behaviors Download PDF

Info

Publication number
CN108419134B
CN108419134B CN201810110304.1A CN201810110304A CN108419134B CN 108419134 B CN108419134 B CN 108419134B CN 201810110304 A CN201810110304 A CN 201810110304A CN 108419134 B CN108419134 B CN 108419134B
Authority
CN
China
Prior art keywords
channel
state matrix
channels
user
channel state
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
Application number
CN201810110304.1A
Other languages
Chinese (zh)
Other versions
CN108419134A (en
Inventor
杨灿
任思璇
徐映雪
盛栋铭
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN201810110304.1A priority Critical patent/CN108419134B/en
Publication of CN108419134A publication Critical patent/CN108419134A/en
Application granted granted Critical
Publication of CN108419134B publication Critical patent/CN108419134B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/466—Learning process for intelligent management, e.g. learning user preferences for recommending movies
    • H04N21/4668—Learning process for intelligent management, e.g. learning user preferences for recommending movies for recommending content, e.g. movies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
    • H04N21/442—Monitoring of processes or resources, e.g. detecting the failure of a recording device, monitoring the downstream bandwidth, the number of times a movie has been viewed, the storage space available from the internal hard disk
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
    • H04N21/442—Monitoring of processes or resources, e.g. detecting the failure of a recording device, monitoring the downstream bandwidth, the number of times a movie has been viewed, the storage space available from the internal hard disk
    • H04N21/44213—Monitoring of end-user related data
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/45—Management operations performed by the client for facilitating the reception of or the interaction with the content or administrating data related to the end-user or to the client device itself, e.g. learning user preferences for recommending movies, resolving scheduling conflicts
    • H04N21/4508—Management of client data or end-user data
    • H04N21/4532—Management of client data or end-user data involving end-user characteristics, e.g. viewer profile, preferences
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
    • H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
    • H04N21/47—End-user applications
    • H04N21/482—End-user interface for program selection
    • H04N21/4826—End-user interface for program selection using recommendation lists, e.g. of programs or channels sorted out according to their score

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Software Systems (AREA)
  • Human Computer Interaction (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Social Psychology (AREA)
  • Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a channel recommendation method based on the fusion of individual history and group current behaviors, which comprises the steps of firstly, constructing a group current watching channel state matrix for describing the characteristics of each current channel, wherein the characteristics comprise the current number of watching people of each channel and the online number of watching people of each channel; then, establishing a personal historical watching channel state matrix for the individual user, wherein the historical watching channel state matrix is used for describing the watching channel characteristics of the user in the past time period and comprises the times of watching each channel by the user and the weight of each watched channel; and finally, sending the group current channel state matrix and the individual historical watching channel state matrix into a recommendation fusion calculation module, calculating the possible watching channels of each user at the current moment by using an intersection selection method or a distance selection method, and pushing the N most possible watching channels to the user.

Description

Channel recommendation method based on fusion of individual history and group current behaviors
Technical Field
The invention relates to the field of IPTV, in particular to a channel recommendation method based on fusion of individual history and group current behaviors.
Background
IPTV is an interactive network television, and television channel recommendation based on IPTV has attractive application prospect. With the development of IPTV and internet tv live broadcasting technologies, users can watch more and more tv channels, so the conventional tv program guide (EPG) cannot meet the use requirement of people to find interested channels in time. In recent years, personalized recommendation systems for IPTV have been studied, however, most recommendation systems are only for on-demand programs and not live recommendations, because the following features of live channels make their recommendation more complicated:
and (3) timeliness: the live television content can only be played in a specific time window, if a user wants to watch a certain program, the user needs to watch the live channel in a specific time zone, the program on demand is not limited to the time zone, and the user can watch the program at any time.
Complexity: an IPTV account is usually shared in a family, that is, multiple people share the same account, which makes it more difficult to capture the behavior of the user because the user's preferences are different in different time periods.
Noise: a viewing record of a live channel has more noisy data than on-demand. In the live broadcast situation, the user may have many fast switching behaviors and advertisement skipping, and the noise data may interfere with the recommendation result of the user.
Based on the characteristics of the IPTV live broadcast channels, it is complicated to find channels that users like from hundreds of live broadcast channels. It is not easy for a recommender system to need to help users find their favorite live channels among hundreds of channels. The recommendation system needs to perform deep analysis on the historical viewing behavior of the user and establish a proper model to perform corresponding recommendation on the user, and the current recommendation algorithm mainly uses collaborative filtering or some other related machine learning methods to perform recommendation. The collaborative filtering can recommend the target user according to the favorite channels of other users, and the problem of cold start is solved. Machine learning methods are also applied to program recommendation of IPTV, for example, clustering users by using methods such as k-means, and recommending users by combining methods of collaborative filtering. However, these methods either require additional user rating information or require a lot of machine computation time, and currently, effective recommendation algorithms directed to live channels are still rarely found.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a channel recommendation method based on the fusion of individual history and group current behaviors, which combines the historical watching behaviors of users and the group watching behaviors of the users to perform personalized channel recommendation for each user.
The purpose of the invention is realized by the following technical scheme: a channel recommendation method based on fusion of individual history and group current behaviors comprises the following steps:
designing a current channel state matrix generator (101) for the overall user, wherein the current channel state matrix generator is used for describing the current state of each channel when the user watches IPTV; aiming at individual users, a personalized historical state matrix generator (102) is designed and used for describing the state of each historical watching channel of each user in a certain historical time, the current channel state matrix constructed by 101 and the personalized historical state matrix constructed by 102 are sent to a user recommendation calculation module (103), and a recommendation algorithm used by the module can generate a recommended program list for each user and send the recommended program list to a push module (104) for independent push of the user.
The construction of a current channel state matrix generator (101) comprises the following steps:
1. a time window deltat is selected.
2. Counting the channel heat p of each channel in the current time window [ t-delta t, t) under the condition that all users watchi。piThe number of viewers in the time window for each channel.
3. Counting the growth rate r of the number of online people of each channel in the current time window [ t-delta t, t) under the condition that all users watchi，riThe calculation formula of (a) is as follows:
Figure GDA0002286857210000021
4. to piThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure GDA0002286857210000022
Piis to piNormalized value, for riThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure GDA0002286857210000023
Riis to riThe values after normalization. Constructing a group current watching channel state matrix at the current time t
Figure GDA0002286857210000024
Wherein the first column CiThe second column indicates the normalized channel heat of each channel, the third column indicates the normalized instantaneous growth rate of each channel, and n is the number of channels.
Secondly, the construction of the personalized historical state matrix generator (102) comprises the following steps:
1. a history sliding window Δ T is selected, where Δ T > Δ T.
2. For each user in IPTV, counting the time period T-delta T, T of the user]Number of views o per channel ini，
3. For each user in IPTV, calculating the time period T-delta T, T of the user]Weight value w of each channel in the channeli，wiIs calculated by the formulawiIs a time period [ T- Δ T, T]Channel C watched by the useriWith Nci being at [ T- Δ T, T]Channel C watched by the user U in the intervaliOf times τ ofkViewing channel C for user k time in user history windowiWeight of (d), τkIs represented by the formula:
Figure GDA0002286857210000032
tkfor the user at [ T- Δ T, T]Inner kth viewing channel CiAt the moment, ts is the starting moment of the sliding window, that is, ts is T- Δ T; t is the current time and Δ T is the historical sliding window size. In addition, τkCan also be expressed by another calculation formula as:
Figure GDA0002286857210000033
t is the current time, tkFor the user in the time window T-delta T, T]Inner kth viewing channel CiAt the instant Δ T is the sliding window size.
4. To o isiThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure GDA0002286857210000034
Oiis a pair oiNormalized value, for wiThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure GDA0002286857210000035
Wiis to wiThe values after normalization. Build each applicationThe channel state matrix H is viewed by the User's personal history, for example, the channel state matrix is viewed by the User's personal history
Figure GDA0002286857210000036
Each row of the matrix sequentially represents the channel number, the channel watching times and the weight of the channel, m is the total number of the channels watched by the user in the time window, and m is the number of the channels.
Thirdly, the user recommendation calculation module (103) selects a proper algorithm for recommendation according to the current viewing channel state matrix C generated by the 101 and the personal history viewing channel state matrix H constructed by the 102, and the construction of the user recommendation calculation module (103) comprises the following methods:
1. the method adopts an intersection selection method and comprises the following steps:
1.1 constructing 101 the current viewing channel state matrix C generated in step 4, and assigning the channel number C to the channel number CiAccording to CiFor example, according to the descending order of the channel watching heat Pi, the TOP x channels that are hot are selected to form the current candidate set: v1＝(C1，C2......Cx)；
1.2 channel number C is assigned according to the personalized historical state matrix generated for each user in step 4 of the construction 102iAccording to CiIs ordered, e.g. by the number of views O per channeliThe popular y channels are selected to form a history candidate set: v2＝(C1，C2......Cy)；
1.3 get V12＝V1∩V2As a recommended channel of the user, if the maximum number K of recommended channels is greater than or equal to the number of the intersection elements, recommending all channels in the intersection to the user; and if the maximum number K of the recommended channels is less than the number of the intersection elements, arranging the Top-K recommending the intersection elements in a descending order according to the channel heat degree.
2. The distance selecting method comprises the following steps:
2.1 Current viewing channel State matrix C and channel number from 101 and 102In the channel state matrix H viewed by people in history, the characteristic value of each channel is used to calculate each channel C in 101iAnd C in 102jForm a distance matrix (201)Wherein DijRepresenting 101 channel CiChannel C in distance 102jThe distance of (d); (each row index value in matrix D corresponds to each channel number in current viewing channel state matrix C, and each column index value in matrix D corresponds to each channel number in personal history viewing channel state matrix H); DijThe calculation is performed using one of the following formulas:
euclidean distance:wherein, C (F)i)nRepresenting channel C in matrix 101iH (F) of the nth characteristic column valuei)nRepresenting channel C in matrix 102jN is the total number of eigencolumns of the matrix.
Manhattan distance:
Figure GDA0002286857210000043
cosine distance:
Figure GDA0002286857210000044
2.2 according to the matrix D, finding out the index value of the column to which the minimum value in each row of the matrix belongs, namely the channel number in the H matrix of 102, and obtaining the channel vector V3＝(C1，C2.....Cn) The vector represents the nearest channel of the personal historical viewing channel state matrix H to each channel in the current viewing channel state matrix C;
2.3 vector V3And (4) sorting the channels in descending order according to the increasing rate of 101 of the medium channel, and taking the first n channels with the fastest increasing rate as recommendations.
Compared with the prior art, the invention has the following advantages and beneficial effects:
the invention takes the group watching behavior of the current user as one of the recommended bases, and then the group watching behavior is recommended by combining the historical watching behavior of each user. The current channel state matrix and the personal historical watching channel state matrix of each user are used for recommending each user in real time, and the method provided by the invention can better capture the watching behavior of each user.
The recommendation algorithm provided by the invention can use less user behavior information to recommend the user. The traditional television program recommendation needs to obtain related information of programs or a rating matrix of a user, but an algorithm provided by the invention can use less user behavior characteristics for recommendation, a data tuple structure used by the invention is { user ID, channel watching starting time, channel number and channel watching duration, and the data structure is easier to obtain relative to program information.
The intersection selection method and the distance selection method provided by the invention can save more computing resources during computation. Unlike training using machine learning or deep learning, the proposed recommendation method does not have much overhead of training time, so that the current channel state matrix and the personal viewing channel history state matrix can be generated in a short time, and real-time recommendation can be performed for the user.
Drawings
FIG. 1 is a schematic diagram of a basic architecture of a recommendation system.
Fig. 2 is a flowchart of a channel recommendation method based on intersection provided in embodiment 2.
FIG. 3 is a diagram illustrating weight calculation in embodiment 2.
Fig. 4 is a flowchart of a distance-based channel recommendation method provided in embodiment 3.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited thereto.
Example 1
A channel recommendation method based on fusion of individual history and group current behavior, as shown in fig. 1, includes:
designing a current channel state matrix generator (101) for the overall user, wherein the current channel state matrix generator is used for describing the current state of each channel when the user watches IPTV; aiming at individual users, a personalized historical state matrix generator (102) is designed and used for describing the state of each historical watching channel of each user in a certain historical time, the current channel state matrix constructed by 101 and the personalized historical state matrix constructed by 102 are sent to a user recommendation calculation module (103), and a recommendation algorithm used by the module can generate a recommended program list for each user and send the recommended program list to a push module (104) for independent push of the user.
The construction of a current channel state matrix generator (101) comprises the following steps:
1. a time window deltat is selected.
2. Counting the channel heat p of each channel in the current time window [ t-delta t, t) under the condition that all users watchi。piThe number of viewers in the time window for each channel.
3. Counting the growth rate r of the number of online people of each channel in the current time window [ t-delta t, t) under the condition that all users watchi，riThe calculation formula of (a) is as follows:
Figure GDA0002286857210000061
4. to piThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure GDA0002286857210000062
Piis to piNormalized value, for riThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure GDA0002286857210000063
Riis to riThe values after normalization. Constructing a group current watching channel state matrix at the current time t
Figure GDA0002286857210000064
Wherein the first column CiChannel number, second columnThe normalized channel heat of each channel is shown, the third column shows the normalized instantaneous growth rate of each channel, and n is the number of channels.
Secondly, the construction of the personalized historical state matrix generator (102) comprises the following steps:
1. a history sliding window Δ T is selected, where Δ T > Δ T.
2. For each user in IPTV, counting the time period T-delta T, T of the user]Number of views o per channel ini，
3. For each user in IPTV, calculating the time period T-delta T, T of the user]Weight value w of each channel in the channeli，wiIs calculated by the formula
Figure GDA0002286857210000065
wiIs a time period [ T- Δ T, T]Channel C watched by the useriWith Nci being at [ T- Δ T, T]Channel C watched by the user U in the intervaliOf times τ ofkViewing channel C for user k time in user history windowiWeight of (d), τkIs represented by the formula:
Figure GDA0002286857210000066
tkfor the user at [ T- Δ T, T]Inner kth viewing channel CiAt the moment, ts is the starting moment of the sliding window, that is, ts is T- Δ T; t is the current time and Δ T is the historical sliding window size. In addition, τkCan also be expressed by another calculation formula as:
Figure GDA0002286857210000067
t is the current time, tkFor the user in the time window T-delta T, T]Inner kth viewing channel CiAt the instant Δ T is the sliding window size.
4. To o isiThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure GDA0002286857210000068
Oiis a pair oiAfter normalizationValue, pair wiThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure GDA0002286857210000069
Wiis to wiThe values after normalization. Constructing a personal historical viewing channel state matrix H of each User, for example, the personal historical viewing channel state matrix of a User of a certain User is
Figure GDA0002286857210000071
Each row of the matrix sequentially represents the channel number, the channel watching times and the weight of the channel, m is the total number of the channels watched by the user in the time window, and m is the number of the channels.
Thirdly, the user recommendation calculation module (103) selects a proper algorithm for recommendation according to the current viewing channel state matrix C generated by the 101 and the personalized history state matrix H constructed by the 102, and the construction of the user recommendation calculation module (103) comprises the following methods:
1. the method adopts an intersection selection method, and comprises the following steps:
1.1 constructing 101 the current viewing channel state matrix C generated in step 4, and assigning the channel number C to the channel number CiAccording to CiIs ordered, e.g. according to channel viewing heat PiThe TOP x channels are selected to form the current candidate set: v1＝(C1，C2......Cx)；
1.2 channel number C is assigned according to the personalized historical state matrix generated for each user in step 4 of the construction 102iAccording to CiIs ordered, e.g. by the number of views O per channeliThe popular y channels are selected to form a history candidate set: v2＝(C1，C2......Cy)；
1.3 get V12＝V1∩V2As a recommended channel of the user, if the maximum number K of recommended channels is greater than or equal to the number of the intersection elements, recommending all channels in the intersection to the user; if the maximum number K of recommended channels is less than the number of the intersecting elements, thenAnd sorting the Top-K recommending the intersection elements in descending order of the heat of the channel.
And (3) other intersection selection method rules:
rule 1 of intersection selection method: according to the current watching channel state matrix, arranging the channel numbers according to the descending order of the channel growth rate, selecting the x channels with the highest growth rate in the period of time, and forming a current candidate set V1＝(C1，C2......Cx) (ii) a According to the personal history viewing channel state matrix, the channel numbers are arranged according to the descending order of the viewing times of the channels, the y channels with the maximum viewing times of the user U are selected, and a history candidate set V is formed2＝(C1，C2......Cy) (ii) a Get V1∩V2As a recommended channel for the user.
Rule 2 of intersection selection method: according to the current watching channel state matrix, arranging the channel numbers according to the descending order of the channel growth rate, selecting the x channels with the highest growth rate in the period of time, and forming a current candidate set V1＝(C1，C2......Cx) (ii) a According to the personal history viewing channel state matrix, the channel numbers are arranged according to the weight value descending order of the channels, y channels with larger weight values are selected, and a history candidate set V is formed2＝(C1，C2......Cy) (ii) a Get V1∩V2As a recommended channel for the user.
Rule 3 of intersection selection method: according to the current channel state matrix, the channel numbers are arranged according to the descending order of the number of the channel viewers, x channels with the largest number of the channel viewers are selected, and a current candidate set V is formed1＝(C1，C2......Cx) (ii) a According to the personal history viewing channel state matrix, the channel numbers are arranged according to the weight of the channels in a descending order, the y channels with the maximum weight corresponding to the user are selected, and a history candidate set V is formed2＝(C1，C2......Cy) (ii) a Get V1∩V2As a recommended channel for the user.
2. The distance selecting method comprises the following steps:
2.1 the vectors generated in accordance with 101 and 102In the pre-viewing channel state matrix C and the personal history viewing channel state matrix H, the feature value of each channel is used to calculate each channel C in 101iAnd C in 102jForm a distance matrix (201)
Figure GDA0002286857210000081
Wherein DijRepresenting 101 channel CiChannel C in distance 102jThe distance of (d); (each row index value in matrix D corresponds to each channel number in current viewing channel state matrix C, and each column index value in matrix D corresponds to each channel number in personal history viewing channel state matrix H); DijThe calculation is performed using one of the following formulas:
euclidean distance:
Figure GDA0002286857210000082
wherein, C (F)i)nRepresenting channel C in matrix 101iH (F) of the nth characteristic column valuei)nRepresenting channel C in matrix 102jN is the total number of eigencolumns of the matrix.
Manhattan distance:
Figure GDA0002286857210000083
cosine distance:
Figure GDA0002286857210000084
2.2 according to the matrix D, finding out the index value of the column to which the minimum value in each row of the matrix belongs, namely the channel number in the H matrix of 102, and obtaining the channel vector V3＝(C1，C2.....Cn) The vector represents the nearest channel of the personal historical viewing channel state matrix H to each channel in the current viewing channel state matrix C;
2.3 vector V3And (4) sorting the channels in descending order according to the increasing rate of 101 of the medium channel, and taking the first n channels with the fastest increasing rate as recommendations.
Example 2
Fig. 2 illustrates an example of performing real-time channel recommendation for an IPTV user using an intersection selection method, where performing recommendation by the intersection selection method includes the following steps:
1) selecting a time window delta t, and counting the number p of people watching each channel in the time window [ t-delta t, t ]i。
2) Calculating the increasing rate of the online number of people of each channel in the time window [ t-delta t, t ]
Figure GDA0002286857210000091
Figure GDA0002286857210000092
3) To piAnd riThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure GDA0002286857210000093
Piis piThe value after the normalization is carried out,
Figure GDA0002286857210000094
Riis to riThe values after normalization. Constructing a current viewing channel state matrix at the time t
Figure GDA0002286857210000095
Each row of the matrix represents a channel C within a time window t- Δ t, t)iThe number of the watching people, the growth rate of the number of the watching people on line;
4) aiming at a user U, constructing a personal historical watching channel state matrix H of the user, selecting a historical sliding window delta T, wherein delta T is greater than delta T, and counting the historical time [ T-delta T, T ] of the user U]Within, the number of views per channel is noted as oi；
5) Calculating [ T- Δ T, T]The weight value of each channel watched by the user U in the time window is described as follows:
Figure GDA0002286857210000096
wi indicates that the user U watches the channel C within the time windowiWeight of (d), τkIs [ T- Δ T, T]In, user U watches channel C the kth timeiN is channel CiAt [ T- Δ T, T]Channel C watched by the user U in the intervaliThe number of times. The formula is described as
Figure GDA0002286857210000097
tkWatching channel C for user U the kth timeiTs is the start time of the historical time window, i.e. T- Δ T, Δ T is the width of the historical sliding window, for oiAnd wiThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure GDA0002286857210000098
Oiis a pair oiThe value after the normalization is carried out,Wiis to wiThe values after normalization. Obtaining a personal historical viewing channel state matrix of a user U
Figure GDA00022868572100000910
FIG. 3 depicts a user U at [ T- Δ T, T [ ]]Within an inner time window, channel CiWeight calculation procedure, in which user U watches channel CiTwice in total, so the weight is wi＝τ1+τ2，
Figure GDA00022868572100000911
ts is the starting time of the historical time window, i.e. T- Δ T;
6) the channel numbers in the current viewing channel state matrix C are sorted according to the characteristics of the channels, and the number P of the viewing people is normalized in this embodimentiThe channels are arranged in descending order, and the hot x channels are selected to obtain the current candidate set V1＝(C1，C2......Cx) Wherein the number x of channels is less than or equal to the number of channels of the current channel state matrix, and the matrix for the user UH, sorting according to the characteristics of the channels, sorting the channels according to the descending order of the normalized channel watching times in the embodiment, selecting y channels popular in the history period of the user U, wherein y is less than or equal to the number of the channels of the personal history watching channel state matrix, and forming a history candidate set V of the user U2＝(C1，C2......Cy)。
7) Selection of V1∩V2And as the recommended channels to the user U, sorting the intersection results according to the descending order of the channel growth rate, and selecting the first N channels in the intersection to recommend to the user.
Example 3
A real-time channel recommendation strategy based on a distance selection method, fig. 4 illustrates an example of real-time channel recommendation for an IPTV user using the distance selection method, and the recommendation using the distance selection method includes the following steps:
the method comprises the following steps: calculating the distance between each row vector in the current viewing channel state matrix C and each row vector in the personal historical viewing channel state matrix H, wherein the row vectors do not include channel numbers, and the distance can be calculated in various calculation modes, such as Euclidean distance, for example, the row vector vec in the current viewing channel state matrix C1＝(P1，R1) And the row vector vec in the personal history viewing channel state matrix H2＝(O1，W1) The calculation method is as follows:
Figure GDA0002286857210000101
where P1 denotes channel C in the currently viewed channel state matrix C1The number of viewers, O1Representing channel C in personal historical viewing channel state matrix H1Number of views of R1Indicating channel C in the currently viewed channel state matrix C1Rate of increase of W1Representing channel C in personal historical viewing channel state matrix H1The weight of (c). In a way of being generalized, the method can be widely used,
Figure GDA0002286857210000102
where P isiIs shown asChannel C in front view channel state matrix CiThe number of viewers, OjRepresenting channel C in personal historical viewing channel state matrix HiNumber of views of RiIndicating channel C in the currently viewed channel state matrix CiOn-line population growth rate, WjRepresenting channel C in personal historical viewing channel state matrix HjThe weight of (c).
After the distance between each row vector in the current viewing channel state matrix C and each row vector in the personal historical viewing channel state matrix H is calculated, an n multiplied by m distance matrix D is obtained, n is the number of channels in the current viewing channel state matrix C, m is the number of channels in the personal historical viewing channel state matrix H, the index value of each row in the matrix D corresponds to the number of the channels in the current viewing channel state matrix C, and the index value of each column in the matrix D corresponds to the number of the channels in the personal historical viewing channel state matrix H.
Step two: selecting the index value of the column to which the minimum value of each row in the matrix D belongs, namely the channel number in the personal history viewing channel state matrix H, and finally obtaining a candidate set S ═ of n channels (CH ═1，CH2，......，CHn) Indicating the channel of the personal historical watching channel state matrix H which is closest to the channel in the current watching channel state matrix C;
step three: the candidate set S is ranked according to the characteristics of the channels in the personal history viewing channel state matrix H, in this example by the number R of viewers per channeliAnd performing descending order arrangement, and selecting the first K channels as recommendations for the user U.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims (6)

1. The channel recommendation method based on the fusion of the individual history and the group current behaviors is characterized by comprising the following steps of:
1.1, constructing a current watching channel state matrix of a group; confirming the time window delta t and counting the number p of the watched people of each channel under the watching conditions of all the users in the time window [ t-delta t, t ]iAnd to piThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure FDA0002293086750000011
Piis piNormalizing the processed values, and calculating a growth rate r of each channeliThe channel growth rate is:
and to riThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure FDA0002293086750000013
constructing a group current viewing channel state matrix according to the normalized number of the current viewing people of the channel and the growth rate
Figure FDA0002293086750000014
CiFor the channel number, each row of the matrix represents the channel number of each channel, the current number of people watching the channel and the current instantaneous growth rate of the channel;
1.2, constructing a personal historical watching channel state matrix; selecting a historical sliding window delta T, and counting the time window T-delta T, T of each user for the user]Number of views o per channel in a segmentiAnd to oiThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure FDA0002293086750000015
Oiis oiA normalized value; calculating the time window T-delta T, T-delta T of each channel of the user]Inner weight wiAnd to wiThe normalization treatment is carried out, and the normalization treatment is carried out,
Figure FDA0002293086750000016
Wiis wiThe normalized value is used for constructing a personal historical viewing channel state matrix of each user according to the number of the viewers of each channel and the weight of each channel
Figure FDA0002293086750000017
Each row of the matrix represents the watching times of each channel of the user and the weight of each channel;
1.3, sending the 1.1 current viewing channel state matrix C and the 1.2 personal historical viewing channel state matrix H into a recommendation calculation module and selecting a recommended channel; the recommendation calculation module adopts the following method: an intersection selection method or a distance selection method; the distance selection method comprises the following steps: euclidean distance selection method, Manhattan distance selection method or cosine distance selection method.
2. The channel recommendation method according to claim 1, wherein in step 1.2, each channel is in a time window [ T- Δ T, T- Δ T [ ]]Inner weight wiThe calculation method adopts one of the following steps:
2.1 method one: the weight of each channel is determined by the user in the time window T-delta T, T]The sum of the weights of each time the channel is viewed is formed, and the formula is described as:
Figure FDA0002293086750000018
wiis channel CiThe weight value Nci is channel CiAt [ T- Δ T, T]The user watches channel C in the intervaliOf times τ ofkViewing channel C for the user's k timeiWeight of (d), τkThe formula of (c) is described as:
Figure FDA0002293086750000021
ts is the start time of the time window, i.e., T- Δ T, TkFor the user in the time window T-delta T, T]Inner kth viewing channel CiAt the moment, the delta T is the size of the sliding window;
2.2 method two: the weight of each channel is determined by the user in the time window T-delta T, T]In each time the channel is watchedThe formula is described as:
Figure FDA0002293086750000022
wiis channel CiThe weight value Nci is channel CiAt [ T- Δ T, T]The user watches channel C in the intervaliOf times τ ofkViewing channel C for the user's k timeiWeight of (d), τkThe formula of (c) is described as:
Figure FDA0002293086750000023
t is the current time, tkFor the user in the time window T-delta T, T]Inner kth viewing channel CiAt the instant Δ T is the sliding window size.
3. The channel recommendation method according to claim 1, wherein in step 1.3, the recommended channel selection is performed according to the current viewing channel state matrix C and the personal history viewing channel state matrix H, and an intersection selection method is adopted, using one of the following rules:
3.1 rule one is selected in the intersection: according to the current channel state matrix, the channel numbers are arranged according to the descending order of the number of the channel viewers, x channels with the largest number of the channel viewers are selected, and a current candidate set V is formed1＝(C1，C2......Cx) (ii) a According to the personal history viewing channel state matrix, the channel numbers are arranged according to the descending order of the viewing times of the channels, the y channels with the maximum viewing times of the user U are selected, and a history candidate set V is formed2＝(C1，C2......Cy) (ii) a Get V1∩V2As a recommended channel for the user U;
3.2 intersection selection rule two: according to the current watching channel state matrix, arranging the channel numbers according to the descending order of the channel growth rate, selecting the x channels with the highest growth rate in the period of time, and forming a current candidate set V1＝(C1，C2......Cx) (ii) a Viewing the channel state matrix according to personal history, arranging the channel numbers in descending order of the viewing times of the channels, and selectingOutputting the y channels with the maximum number of times of U watching by the user to form a history candidate set V2＝(C1，C2......Cy) (ii) a Get V1∩V2As a recommended channel for the user;
3.3 rule three of intersection selection: according to the current watching channel state matrix, arranging the channel numbers according to the descending order of the channel growth rate, selecting the x channels with the highest growth rate in the period of time, and forming a current candidate set V1＝(C1，C2......Cx) (ii) a According to the personal history viewing channel state matrix, the channel numbers are arranged according to the weight value descending order of the channels, y channels with larger weight values are selected, and a history candidate set V is formed2＝(C1，C2......Cy) (ii) a Get V1∩V2As a recommended channel for the user;
3.4 intersection selection rule four: according to the current channel state matrix, the channel numbers are arranged according to the descending order of the number of the channel viewers, x channels with the largest number of the channel viewers are selected, and a current candidate set V is formed1＝(C1，C2......Cx) (ii) a According to the personal history viewing channel state matrix, the channel numbers are arranged according to the weight of the channels in a descending order, the y channels with the maximum weight corresponding to the user are selected, and a history candidate set V is formed2＝(C1，C2......Cy) (ii) a Get V1∩V2As a recommended channel for the user.
4. The channel recommendation method according to claim 1, wherein in step 1.3, the recommended channel selection is performed according to the current viewing channel state matrix and the personal history viewing channel state matrix, and the euclidean distance selection method is adopted:
4.1, calculating the distance between each row vector in the current viewing channel state matrix C and each row vector in the personal historical viewing channel state matrix H by adopting an Euclidean distance calculation method, wherein the channel number of the row vector is not put into a calculation formula, and constructing a distance matrix D with dimension of n multiplied by m, wherein n in the matrix D is the channel number of the current viewing channel state matrix C, namely the row vector number of C; m is the number of channels in the personal history viewing channel state matrix H, namely the number of row vectors of H; the index value of each row in the matrix D corresponds to each channel number of the current viewing channel state matrix C, namely a first column in C; the index value of each column in the matrix D corresponds to each channel number of the user personal history viewing channel state matrix H, namely the first column in H;
4.2, finding out an index value of a column to which the minimum value of each row in the matrix D belongs, namely the channel number in the personal historical viewing channel state matrix H, wherein n is n candidate channels, and n is the number of channels in the current viewing channel state matrix C, namely the row number of C;
and 4.3, arranging the n channels obtained in the step 4.2 according to the characteristic columns in the C or H, sorting according to the historical watching times of the channels or the weight of each channel, and selecting the first K channels as corresponding user recommendations.
5. The channel recommendation method according to claim 1, wherein in step 1.3, the recommended channel selection is performed according to the current channel state matrix and the personalized historical state matrix, and a manhattan distance selection method is adopted:
5.1, calculating the distance between each row vector in the current viewing channel state matrix C and each row vector in the personal historical viewing channel state matrix H by adopting a Manhattan distance calculation method, wherein the channel number of the row vector is not put into a calculation formula, and constructing a distance matrix D with dimension of n multiplied by m, wherein n in the matrix D is the channel number of the current viewing channel state matrix C, namely the row vector number of C; m is the number of channels in the personal history viewing channel state matrix H, namely the number of row vectors of H; the index value of each row in the matrix D corresponds to each channel number of the current viewing channel state matrix C, namely a first column in C; the index value of each column in the matrix D corresponds to each channel number of the user personal history viewing channel state matrix H, namely the first column in H;
5.2 finding out the index value of the column to which the minimum value of each row in the matrix D belongs, namely the channel number in the personal historical viewing channel state matrix H, wherein n is n candidate channels, and n is the number of the channels in the current viewing channel state matrix C, namely the row number of C;
and 5.3, arranging the n channels obtained in the step 5.2 according to the characteristic columns in the C or H, sorting according to the historical watching times of the channels or the weight of each channel, and selecting the first K channels as corresponding user recommendations.
6. The channel recommendation method according to claim 1, wherein in step 1.3, the recommended channel is selected according to the current channel state matrix and the personalized history state matrix, and a cosine distance selection method is adopted:
6.1 calculating the distance between each row vector in the current viewing channel state matrix C and each row vector in the personal historical viewing channel state matrix H by adopting a cosine distance calculation method, wherein the channel number of the row vector is not put into a calculation formula, and constructing a distance matrix D with dimension of n multiplied by m, wherein n in the matrix D is the number of channels of the current viewing channel state matrix C, namely the number of the row vectors of C; m is the number of channels in the personal history viewing channel state matrix H, namely the number of row vectors of H; the index value of each row in the matrix D corresponds to each channel number of the current viewing channel state matrix C, namely a first column in C; the index value of each column in the matrix D corresponds to each channel number of the user personal history viewing channel state matrix H, namely the first column in H;
6.2 finding out the index value of the column to which the minimum value of each row in the matrix D belongs, namely the channel number in the personal historical viewing channel state matrix H, wherein n is n candidate channels, and n is the number of the channels in the current viewing channel state matrix C, namely the row number of C;
and 6.3, arranging the n channels obtained in the step 6.2 according to the characteristic columns in the C or H, sorting according to the historical watching times of the channels or the weight of each channel, and selecting the first K channels as corresponding user recommendations.
CN201810110304.1A 2018-02-05 2018-02-05 Channel recommendation method based on fusion of individual history and group current behaviors Active CN108419134B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810110304.1A CN108419134B (en) 2018-02-05 2018-02-05 Channel recommendation method based on fusion of individual history and group current behaviors

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810110304.1A CN108419134B (en) 2018-02-05 2018-02-05 Channel recommendation method based on fusion of individual history and group current behaviors

Publications (2)

Publication Number Publication Date
CN108419134A CN108419134A (en) 2018-08-17
CN108419134B true CN108419134B (en) 2020-02-18

Family

ID=63127699

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810110304.1A Active CN108419134B (en) 2018-02-05 2018-02-05 Channel recommendation method based on fusion of individual history and group current behaviors

Country Status (1)

Country Link
CN (1) CN108419134B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109600641B (en) * 2018-12-18 2021-08-10 北京字节跳动网络技术有限公司 Method and apparatus for processing information
CN109889864B (en) * 2019-01-03 2021-08-31 百度在线网络技术（北京）有限公司 Video pushing method, device, equipment and storage medium
CN110087119B (en) * 2019-04-26 2022-02-22 广州酷狗计算机科技有限公司 Live broadcast home page display method and device and computer readable storage medium
CN110351580B (en) * 2019-07-12 2021-07-13 四川长虹电器股份有限公司 Television program topic recommendation method and system based on non-negative matrix factorization
CN111461516B (en) * 2020-03-26 2021-04-13 深圳市百纳九洲科技有限公司 Comprehensive grid management system platform for smart city
CN116320511A (en) * 2023-02-03 2023-06-23 华南理工大学 Cross-domain fusion recommendation method based on graph convolution network

Family Cites Families (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1756336A (en) * 2004-09-29 2006-04-05 松下电器产业株式会社 Television channel commending system and commending method
US20120151367A1 (en) * 2010-12-10 2012-06-14 Nokia Corporation Method and apparatus for registering a content provider channel for recommendation of content segments
WO2013118232A1 (en) * 2012-02-09 2013-08-15 パナソニック株式会社 Information recommendation device, information recommendation system, information recommendation method and program
CN102769781B (en) * 2012-07-17 2015-01-21 青岛海信传媒网络技术有限公司 Method and device for recommending television program
CN102999634B (en) * 2012-12-18 2016-07-06 百度在线网络技术（北京）有限公司 Navigate recommendation method, system and cloud server based on the user of browser data
CN104123284B (en) * 2013-04-24 2018-01-23 华为技术有限公司 The method and server of a kind of recommendation
CN103546773B (en) * 2013-08-15 2017-07-11 Tcl集团股份有限公司 The recommendation method and its system of TV programme
CN103491441B (en) * 2013-09-09 2017-02-01 东软集团股份有限公司 Recommendation method and system of live television programs
CN103888852B (en) * 2014-03-24 2017-05-31 清华大学 For the video recommendation method and device of social television
KR101552952B1 (en) * 2014-12-02 2015-09-14 성균관대학교산학협력단 Tv channel recomendation method, apparatus and system by using tv watching time record
CN104486339B (en) * 2014-12-16 2018-07-10 广州华多网络科技有限公司 The method and apparatus that recommending data is shown in social networking application
CN105915949A (en) * 2015-12-23 2016-08-31 乐视网信息技术（北京）股份有限公司 Video content recommending method, device and system
KR20170078464A (en) * 2015-12-29 2017-07-07 성균관대학교산학협력단 Personalized tv channel recommendation considering time dependent propensity of viewer using constrained optimization technique
CN106791901B (en) * 2016-12-15 2020-09-22 Tcl科技集团股份有限公司 Live broadcast channel ordering method and device

Also Published As

Publication number Publication date
CN108419134A (en) 2018-08-17

Similar Documents

Publication Publication Date Title
CN108419134B (en) Channel recommendation method based on fusion of individual history and group current behaviors
US20200341975A1 (en) Methods and systems for identifying an information resource for answering natural language queries
US8943537B2 (en) Method and system for presenting personalized television program recommendation to viewers
JP6235556B2 (en) Content presentation method, content presentation apparatus, and program
Bjelica Towards TV recommender system: experiments with user modeling
US20200193288A1 (en) System and Method for Content Discovery
US20030131355A1 (en) Program guide system
US10575057B2 (en) Systems and methods for improving accuracy in media asset recommendation models
US20120317123A1 (en) Systems and methods for providing media recommendations
KR100839306B1 (en) System and method for recommendation of digital television program using reflection techniques of program veiwing pattern
US20130263168A1 (en) Cooperative Filtering Algorithm-Based Personal Preference Program Recommendation System for IPTV
Sanchez et al. Recommender system for sport videos based on user audiovisual consumption
JP2004505562A (en) 3-way media recommendation method and system designation
CN103546778A (en) Television program recommendation method and system, and implementation method of system
JP4370850B2 (en) Information processing apparatus and method, program, and recording medium
DE102011056870A1 (en) Blaster/digital receiver type remote control device i.e. smartphone, for controlling set of e.g. TVs, has computer system analyzing use information and producing set of references including information, which identify set of contents
CN103546773A (en) Television program recommendation method and system
JP2005056361A (en) Information processor and method, program, and storage medium
EP1337914A1 (en) Method and apparatus for obtaining auditory and gestural feedback in a recommendation system
EP3480767A1 (en) Systems and methods for improving accuracy in media asset recommendation models
US10003836B2 (en) Systems and methods for improving accuracy in media asset recommendation models based on users' levels of enjoyment with respect to media assets
US20160314404A1 (en) Systems and methods for improving accuracy in media asset recommendations based on data from multiple data spaces
KR20130090042A (en) Apparatus, system, method and computer readable recording media storing the program for automatic recommendation of tv program contents based on user-preferred topic
Li et al. IPTV channel zapping recommendation with attention mechanism
CN112784069B (en) IPTV content intelligent recommendation system and method

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