WO2018223772A1 - 内容推荐方法和系统 - Google Patents

内容推荐方法和系统 Download PDF

Info

Publication number
WO2018223772A1
WO2018223772A1 PCT/CN2018/083108 CN2018083108W WO2018223772A1 WO 2018223772 A1 WO2018223772 A1 WO 2018223772A1 CN 2018083108 W CN2018083108 W CN 2018083108W WO 2018223772 A1 WO2018223772 A1 WO 2018223772A1
Authority
WO
WIPO (PCT)
Prior art keywords
user data
content
sample
feature
terminal device
Prior art date
Application number
PCT/CN2018/083108
Other languages
English (en)
French (fr)
Inventor
胡仲义
王义寅
王细勇
Original Assignee
华为技术有限公司
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 华为技术有限公司 filed Critical 华为技术有限公司
Publication of WO2018223772A1 publication Critical patent/WO2018223772A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/22Matching criteria, e.g. proximity measures
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • G06F18/241Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133Distances to prototypes
    • G06F18/24137Distances to cluster centroïds

Definitions

  • the present application relates to the field of communications, and in particular, to a content recommendation method and system.
  • terminal devices such as mobile phones and tablet computers and the development of mobile Internet technologies
  • terminal devices are gradually becoming an important platform for people to obtain information. For example, people can use the terminal device for news browsing, book reading, video viewing, music listening, and social communication.
  • the continuous expansion of network services and information content will cause "information overload" in the terminal device, which will seriously affect the user experience of the terminal device and the resource utilization of the mobile Internet.
  • the content recommendation system analyzes and understands the user's needs through machine learning technology, and filters and filters the information based on the user's needs, thereby recommending content that meets the user's needs to the user.
  • the content recommendation system has become one of the important means to solve the problem of information overload, and has been widely concerned and applied in the field of Internet and terminal.
  • a terminal device collects user data and then uploads user data to a server; the server performs analysis modeling and recommendation calculation based on user data to mine a binary relationship between the user and the project (user -item), and then discover items related to user requirements from a large amount of content data, for example, news, videos, online products, etc., and then send the recommendation results to the terminal device to meet the personalized needs of the users of the terminal devices.
  • the terminal device Since the terminal device usually needs to collect user data including information about the location, status, behavior, and the like of the user and upload it to the server, the user data may relate to the user's private information, for example, the user's action track included in the user's location information, Browsing records, etc., is likely to cause user privacy leaks.
  • the terminal device manufacturer it is necessary to store the user information in the user's own terminal device as much as possible, thereby fully protecting the user's privacy. If a terminal device cannot protect privacy, the user will select the terminal device of another manufacturer, which will have a huge impact on the terminal device manufacturer. Therefore, the existing content recommendation system is difficult to meet the needs of users and terminal equipment manufacturers for user privacy protection.
  • the embodiment of the present application provides a content recommendation method and system, which can recommend corresponding content to a user of the terminal device under the condition that the privacy of the user of the terminal device is protected.
  • the embodiment of the present application provides a content recommendation method, including: collecting first user data of a terminal device in a first time period, where the first user data includes personal data of a user who uses the terminal device in a first time period. At least one of behavioral information, and real-time context-related information; extracting, based on the first user data, at least one content feature from a content recommendation model saved by the terminal device, the at least one content feature being a user characterizing the terminal device a content feature of the recommended content, wherein the content recommendation model is constructed by the terminal device based on the historical user data; and transmitting at least one content feature to the server, and receiving recommended content for the user of the terminal device from the server, wherein The recommended content of the user of the device is obtained by the server according to at least one content feature query.
  • the content recommendation model is constructed by using the computing power of the terminal device and the recommendation calculation is performed, thereby fully utilizing the computing power of the terminal device and liberating the computing power of the server.
  • the foregoing content recommendation method further includes: collecting, in a second time period, second user data of the terminal device, where the second time period is earlier than the first time period, and the second user data is historical user data;
  • a content recommendation model is constructed based on historical user data and a plurality of content features corresponding to historical user data.
  • the foregoing content recommendation method further includes: collecting, in a third time period, third user data of the terminal device, where the third time period is later than the second time period;
  • the content recommendation model is updated by the three user data and the plurality of content features corresponding to the third user data.
  • the update of the content recommendation model may make the recommended content for the user of the terminal device more in line with the needs of the user of the terminal device.
  • the historical user data includes multiple historical user data samples, each historical user data sample includes multiple sample feature values, and the content recommendation model includes multiple historical user data. The correspondence between each sample category and content features to which the sample belongs.
  • Constructing the content recommendation model based on the plurality of historical user data samples and the content features corresponding to the plurality of historical user data samples includes: for each historical user data sample, utilizing the feature weight vector sum corresponding to the historical user data sample And acquiring, by the plurality of sample feature values included in the historical user data sample, a weighting value of the historical user data sample, where the feature weight vector corresponding to the historical user data sample includes and the plurality of historical user data samples respectively included a plurality of weight values corresponding to the sample feature values; dividing the plurality of historical user data samples into a plurality of sample categories based on the weighted values of the plurality of historical user data samples; for each sample category, belonging to the sample category Content features corresponding to respective historical user data samples are associated with the sample category.
  • dividing the plurality of historical user data samples into the plurality of sample categories includes: dividing the plurality of historical user data samples into the plurality of sample categories by using a clustering algorithm.
  • the content recommendation model further includes a cluster center point of each sample category, a number of user data samples belonging to the sample category, and various content features associated with the sample category in the sample category. The probability of occurrence.
  • the clustering algorithm is used to construct the content recommendation model, which avoids the large-scale matrix operation used in the conventional content recommendation system, thereby alleviating the power consumption problem of the terminal device caused by the excessive calculation of the content recommendation model.
  • the third user data includes multiple third user data samples, and each third user data sample of the plurality of third user data samples includes multiple samples.
  • the feature value, based on any one of the plurality of third user data samples, and the content feature corresponding to the third user data sample, the content recommendation model includes: utilizing features corresponding to the third user data sample And obtaining, by the weight vector and the plurality of sample feature values included in the third user data sample, a weight value of the third user data sample, where the feature weight vector corresponding to the third user data sample includes Calculating a plurality of weight values corresponding to the plurality of sample feature values included in the three user data samples; calculating a distance between the weighted value of the third user data sample and a cluster center point of each sample category; and the third user data The sample is added to a sample category that is closest to the distance; the sample category is updated with the weighted value of the third user data sample Cluster center, and wherein the third user using the content data samples
  • the first user data includes a first user data sample
  • the first user data sample includes a plurality of sample feature values
  • the content recommendation model is based on the first user data.
  • Extracting the at least one content feature includes: obtaining a weight value of the first user data sample by using the feature weight vector corresponding to the first user data sample and the plurality of sample feature values included in the first user data sample, where The feature weight vector corresponding to the user data sample includes a plurality of weight values respectively corresponding to the plurality of sample feature values included in the first user data sample; calculating a weight value of the first user data sample and clustering of each sample category The distance between the center points as the similarity between the first user data sample and each sample category; finding a sample category with the highest similarity with the first user data sample, and extracting the associated with the sample category At least one content feature as at least one content feature characterizing the recommended content for the user.
  • an embodiment of the present application provides a content recommendation method, including: receiving at least one content feature that represents recommended content for a user of the terminal device; and storing the content feature in advance based on the at least one content feature, The recommended content for the user of the terminal device is found in the memory of the content characterized by the content feature and the correspondence between them; and the recommended content for the user of the terminal device is returned to the terminal device.
  • one or both of the query function of the search engine and the query function of the database are used to find the recommended content of the user for the terminal device.
  • the server is only used for storage, query, and transmission of content without performing related processes such as constructing a content recommendation model and performing recommendation calculation, and thus utilizing server storage.
  • an embodiment of the present application provides a terminal device including at least one processor, a memory, and computer executable instructions stored on the memory and executable by the at least one processor, wherein the at least one processor executes The computer executable instructions to implement the content recommendation method performed by the terminal device described above.
  • an embodiment of the present application provides a server including at least one processor, a memory, and computer executable instructions stored on the memory and executable by the at least one processor, wherein the at least one processor executes the computer
  • the instructions are executable to implement the content recommendation method performed by the server described above.
  • an embodiment of the present application provides a computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements the content recommendation method performed by the terminal device.
  • an embodiment of the present application provides a content recommendation system, including the foregoing terminal device and a server.
  • FIG. 1 is a schematic structural diagram of a content recommendation system according to an embodiment of the present application
  • FIG. 2 is a schematic structural diagram of a terminal device and a server for implementing a content recommendation method according to an embodiment of the present application;
  • FIG. 3 is a flowchart showing a content recommendation process performed by the terminal device shown in FIG. 2;
  • FIG. 4 is a flow chart showing a content recommendation process performed by the server shown in FIG. 2;
  • FIG. 5 shows a schematic diagram of an example hardware architecture of a terminal device for implementing a content recommendation method according to an embodiment of the present application
  • FIG. 6 shows a schematic diagram of an example hardware architecture of a server for implementing a content recommendation method in accordance with an embodiment of the present application
  • FIG. 7 shows a schematic diagram of another example architecture of a server for implementing a content recommendation method in accordance with an embodiment of the present application.
  • FIG. 1 shows a schematic diagram of the architecture of a content recommendation system according to an embodiment of the present application.
  • the content recommendation system 100 includes two parts, an end side and a cloud side, wherein the end side includes at least one terminal device 102 (for example, the terminal devices 102-1, 102-1, etc. shown in FIG. 1).
  • the cloud side is a computing/storage system consisting of at least one server 106 (e.g., servers 106-1, 106-2, and 106-3 shown in Figure 1).
  • the terminal device 102 communicates with the server 106 via the wireless access point 104 or the Internet (Internet) to obtain the desired service or content.
  • the present application provides a content recommendation method, in which user data of a terminal device is collected by a terminal device, and the user data includes a picture, text, and video stored locally by a user of the terminal device.
  • Information such as the application, the user's operation behavior on the terminal device, and the context in which the user of the terminal device is located, including the physical environment, the network environment, and the terminal environment.
  • FIG. 2 shows an architectural diagram of a terminal device and a server for implementing a content recommendation method according to an embodiment of the present application.
  • FIG. 3 shows a flow chart of a content recommendation process implemented by the terminal device shown in FIG. 2. The implementation details of the specific processing performed by the terminal device in the content recommendation method described above will be described in detail below with reference to FIGS. 2 and 3.
  • the terminal device 102 includes a data collection unit 200, a model construction unit 202, a feature recommendation unit 204, an end side communication unit 206, and a content output unit 208, wherein: the data collection unit 200 is configured to collect the terminal device.
  • the model construction unit 202 is configured to construct a content recommendation model (ie, a content recommendation model for a user of the terminal device) based on historical user data and a plurality of content features corresponding to the historical user data, based on the latest user data and the latest user The plurality of content features corresponding to the data update the content recommendation model, and the save content recommendation model;
  • the feature recommendation unit 204 is configured to extract at least one content feature from the content recommendation model as the feature recommendation content based on the current user data (ie, for The content feature of the recommended content of the user of the terminal device;
  • the end-side communication unit 206 is configured to transmit the content feature extracted by the feature recommendation unit 204 that represents the recommended content to the server 106, and receive the recommended content from the server 106;
  • the content output unit 208 is configured to be from the end side communication unit 20 6 Receive the recommended content and present the recommended content to the user of the terminal device.
  • the data collection unit 200 can collect user data from the terminal device periodically or in real time; the model construction unit 202 can be based on the historical user when the data collection unit 200 collects historical user data sufficient to construct the content recommendation model.
  • the data and its corresponding plurality of content features construct a content recommendation model, or construct a content recommendation model based on historical user data generated at a predetermined time period prior to the current time and its corresponding plurality of content features.
  • the model construction unit 202 may be based on the latest user data. And corresponding content features to update the content recommendation model.
  • the feature recommendation unit 204 can base the current user data collected based on a short period of time before the current time, for example, within 2 seconds.
  • the content recommendation model extracts at least one content feature as a content feature characterizing the recommended content for the user of the terminal device, and transmits the extracted content feature to the server via the end-side communication unit 206.
  • the current user data refers to the first user data generated in the first time period before the current time
  • the historical user data refers to the second user data generated in the second time period before the current time, wherein the second time period is early In the first time period and generally longer than the duration of the first time period
  • the latest user data refers to the third user data generated in the third time period before the current time, wherein the third time period is later than the second time period and early Or include the first time period.
  • the current time is 12:00 on April 26, 2015
  • the first time period can be 11:55-11:58 on April 26, 2015
  • the second time period can be 00:00 on January 1, 2015.
  • the third period can be from 00:00 on April 1, 2015 to 23:59 on April 15, 2015.
  • the terminal device 102 shown in FIG. 2 can implement content recommendation for its user by the following process: S302, collecting first user data in a first time period; S304, based on the first user data, from Extracting, by the content recommendation model saved by the terminal device, at least one content feature as a content feature that represents the recommended content of the user for the terminal device; and S306, transmitting the extracted at least one content feature to the server, and receiving, from the server, The recommended content of the user of the terminal device, wherein the recommended content is obtained by the server according to at least one content feature query sent by the terminal device.
  • the user data includes one or more of user behavior data, user context data, and user personal data
  • the user behavior data includes information about an operation behavior of a user of the terminal device on the terminal device
  • the user context data includes Information about a context in which the user of the terminal device is located, including a physical environment, a network environment, a terminal environment, etc., where the user personal data includes a picture, text, video, application, etc. stored locally by the user of the terminal device. Information about each item inside.
  • User behavior data may contain various related information of the user's operation behavior on the terminal device, and different for different recommended services (eg, recommended news, recommended applications, etc.), but the overall model can be characterized as shown in Table 1. Shown as follows:
  • the user profile data may include information about the context of the user of the terminal device, including physical environment data, terminal state data, user state data, etc., and the overall model can be characterized as shown in Table 2:
  • the user personal data may contain related information about various contents stored locally by the user of the terminal device, and may be selectively used according to different recommended services, and the overall model can be characterized as shown in Table 3:
  • the content feature corresponding to the user data may be acquired by the terminal device from the content source while acquiring the content data corresponding to the user data from the corresponding content source, or may be directly extracted from the content data corresponding to the user data by the terminal device.
  • the set of content features stored by the terminal device is a subset of the set of content features stored by the server, the overall model of which can be characterized as shown in Table 4:
  • user data is a collection of one or more user data samples, each of which includes one or more of user behavior data, user context data, and user personal data collected at the same time.
  • user data samples can be characterized in the manner shown in expression (1):
  • ⁇ X ⁇ ⁇ user behavior data, user context data, user personal data ⁇ expression (1)
  • X represents a user data sample
  • user behavior data user context data
  • user personal data are sample feature values of user data samples.
  • the user data samples may include more or fewer sample feature values
  • expression (1) is just one example representation of the user data samples. More generally, user data samples can be characterized as expression (2):
  • X i represents the i-th user data sample
  • users of terminal devices have different activity patterns within 24 hours a day and/or at different times and/or seasons throughout the year, and thus may have different preferences.
  • different feature weight vectors may be constructed for different time periods and/or seasons to pass weight values corresponding to sample feature values such as user behavior data, user context data, and user personal data. The difference is at least partially reflected in the difference in preferences of the user of the terminal device. Therefore, the historical user data, the latest user data, and the current user data may be applied with feature weight vectors respectively corresponding to their temporal relationships to at least partially reflect different preferences of the user of the terminal device at different times.
  • the feature weight vector having a temporal correspondence with the content data sample X i can be characterized as expression (3):
  • W i represents a feature weight vector (ie, an i-th feature weight vector) having a temporal correspondence with the i-th user data sample X i
  • a feature weight ie, a weight value
  • corresponding to the nth sample feature value in the i-th user data sample X i is represented in the i-th feature weight vector.
  • each user data sample may correspond to one content feature, and multiple user data samples may correspond to the same content feature, and user data including multiple user data samples may correspond to at least one content feature, and each content feature may be Has one or more feature values.
  • the content features can be characterized in the manner shown in expression (4):
  • Y i represents the content feature of the i-th item, Indicates the mth feature value in the i-th content feature.
  • the model building unit content recommendation model 202 may be constructed by the following process: for each user historical data sample, e.g., historical user data samples X i (1 ⁇ i ⁇ Q), users using historical data samples X i corresponding feature weight vectors W i And sample feature values included in the historical user data sample X i S i weighted value acquiring historical user data samples of X i; q historical user based on the weighted data samples X 1 to X q value S 1 to S q, q the user historical data samples X 1 to X q divided into a plurality of sample type, e.g., sample type a C 1 to C r; and for each sample type, e.g., sample type C j (1 ⁇ j ⁇ r), will be assigned to each sample type C j historical
  • the content recommendation model may include a correspondence between each sample category and content features to which the plurality of historical user data samples belong, and the process of creating the content recommendation model is to divide the plurality of historical user data samples into multiple samples.
  • a category and the process of associating each sample category with a content feature corresponding to a historical user data sample attributed to it.
  • the model construction unit 202 may divide the historical user data samples X 1 to X q into a plurality of sample categories C 1 to C r by a clustering algorithm, in which case the content recommendation model may also include each The cluster center point of the sample category, the number of user data samples attributed to the sample category, and the probability that each content feature associated with the sample category appears in the sample category.
  • the content recommendation model can be characterized in the manner shown in Expression (5):
  • the form of the key-value pair is used to indicate the correspondence relationship between the sample category C j and the content feature corresponding to the user data sample belonging to the sample category C j
  • C j represents the category label of the j-th sample category. Representing the number of user data samples attributed to the sample category C j , Representing the cluster center point of the sample category C j , ⁇ Y k :p k ⁇ indicates that the kth content feature Y k corresponding to the user data sample belonging to the sample category C j and the content feature appear in the sample category C j
  • the probability p k , the probability p k can be derived from the expression (6):
  • model construction unit 202 uses the clustering algorithm to construct the content recommendation model, the large-scale matrix operation used in the conventional content recommendation system is avoided, thereby alleviating the terminal device caused by the excessive calculation of the content recommendation model. Power consumption problem.
  • the model construction unit 202 has finished its construction content recommendation model based on historical user data samples X 1 to X Q, if the data acquisition unit 200 further collect enough to update the content recommendation models including the latest user data samples X
  • the latest user data of q+i to Xq+j the model construction unit 202 can update the content recommendation model by using the latest user data for each latest user data sample, for example, the latest user data sample Xq+i sample X q + i corresponding feature weight vector W q + i and the latest user data samples X q + i wherein sample values included Get latest user data X q + i of the sample value S q + i weighting; calculating the latest user data X q + sample weight value of i + Q i with the respective S sample R & lt category of a C 1 cluster center C to The distance between, as the distance between the latest user data sample X q+i and each sample category C 1 to C r ; the latest user data sample X q
  • the feature recommendation unit 204 may extract at least the recommended content characterizing the user for the terminal device from the content recommendation model by the following process A content feature: utilizing a feature weight vector W t corresponding to the current user data sample X t (ie, the current user data includes only the user data sample) and sample feature values included in the current user data sample X t Get current user data sample X t) weighted value of S t; calculating a current user data sample X t S t a weighted value to each cluster center sample type of a C 1 to C r to The distance between the current user data sample X t and the respective sample categories C 1 to C r ; find a sample category with the highest similarity to the current user data sample X t , for example, a sample The category Cj , and extracting at least one content feature associated with the sample category Cj as at least one content feature characterizing the
  • the K item content features associated with the sample category C j may be extracted according to the TOP-K recommendation rule.
  • the feature recommendation unit 204 may extract two content features having the highest frequency of occurrence in the sample category C j as content features characterizing the recommended content for the user of the terminal device.
  • FIG. 4 shows a flow chart of a content recommendation process performed by the server shown in FIG. 2.
  • the server 106 includes a cloud side communication unit 402 and a content inquiry unit 404, wherein: the cloud side communication unit 402 is configured to receive at least one of the recommended content characterizing the user for the terminal device from the end side communication unit 206.
  • the content feature ie, performing step S402 and transmitting the recommended content for the user of the terminal device to the end-side communication unit 206 (ie, performing step S406);
  • the content query unit 404 is configured to be based on the recommended content characterizing the user for the terminal device
  • At least one content feature of the user searches for a recommended content for the user of the terminal device in a memory in which the content feature, the content characterized by the content feature, and the correspondence between the content are stored in advance (ie, step S404 is performed).
  • the content query unit 404 can search for the terminal device based on one or both of the query function of the search engine and the query function of the database based on at least one content feature that characterizes the recommended content for the user of the terminal device. Recommended content for users.
  • the content query unit 404 may search for the recommended content of the user for the terminal device based on the feature keyword using the query function of the search engine;
  • the content query unit 404 may first store a table in which the keyword ID, the feature keyword, and the correspondence relationship therebetween are stored (ie, the feature table) Finding a corresponding feature keyword, and then searching for the recommended content of the user for the terminal device based on the found feature keyword using the query function of the search engine; or when the content feature characterizing the recommended content for the terminal device is represented as a keyword
  • the ID query unit 404 may first search for a corresponding feature keyword in a table storing a keyword ID, a feature keyword, and a correspondence relationship therebetween, and then use the query function of the database to store the feature keyword, a table of contents, and correspondence between them (ie, a table of contents)
  • the terminal device 102 performs the following processes: collecting user data and updating the content recommendation model with the collected user data; extracting, from the content recommendation model, a keyword ID characterizing the recommended content for the user of the terminal device based on the user data, that is, the content feature is composed of
  • the keyword ID is characterized, assuming that the content feature is "game, chess" and the corresponding keyword ID is "ID: 2"; the extracted keyword ID, that is, "ID: 2" is transmitted to the server 106.
  • the server 106 performs the following process: receiving the keyword ID, ie, "ID: 2", and querying the content feature and the content data to obtain two results of "chess, game, chess" and “go, game, chess".
  • “Game, chess” is the content represented by the content feature; the contents of "Chess, Game, Chess” and “Go, Game, Chess” are sent back to the terminal device 102.
  • the terminal device 102 recommends the "Chess, Go” application to the user of the terminal device to complete the entire recommendation process.
  • the collection of user data, the modeling of the content recommendation model, and the recommendation calculation of the content feature are completed by the terminal device without uploading the user data from the terminal device to the server, so The personal privacy of the user of the terminal device is ensured; the terminal device cooperates with the server, and the content feature is extracted by the terminal device and the content query of the server based on the content feature solves the problem that the recommended content of the terminal device is lacking;
  • the class algorithm performs the modeling of the content recommendation model, avoids the matrix operation, and has a small amount of calculation. It not only utilizes the computing power of the terminal device, but also avoids the problem of excessive energy consumption caused by the large calculation amount of the terminal device.
  • the content recommendation scheme according to the embodiment of the present application is suitable for all terminal device recommendation scenarios, and is particularly suitable for use by terminal device manufacturers.
  • FIG. 5 shows a schematic diagram of an exemplary hardware architecture of a terminal device for implementing a content recommendation method according to an embodiment of the present application.
  • the hardware architecture of the terminal device 102 can include an application processor 510, a memory 520, a communication subsystem 530, and a power management subsystem 540, wherein: the memory 520 stores a computer executable program, the computer executable program The operating system, the protocol stack program, and the application are included; the power management subsystem 540 is configured to supply power to the entire terminal device 102, and specifically may be a power management chip; the communication subsystem 530 is a basic communication unit of the terminal device 102.
  • communication subsystem 530 is a wireless modem (Modem) that is primarily used to implement baseband processing, modem, signal amplification and filtering, equalization, and the like.
  • the communication subsystem 530 can include a baseband processor 531, a radio frequency module 532, and an antenna 533, wherein the baseband processor 531 and the application processor 533 can be integrated in the same chip.
  • the baseband processor 531 and the application processor 510 can also be deployed separately, for example, the baseband processor 531 and the application processor 510 interact as information between two independent chips through inter-core communication. In the case of a split deployment, the baseband processor 531 is equivalent to one peripheral of the application processor 510, and the two processors require separate external memories and software upgrade interfaces.
  • the radio frequency module 532 is primarily responsible for signal transmission and reception; the baseband processor 531 is primarily responsible for signal processing, such as A/D, D/A conversion of signals, codec decoding, channel coding and decoding.
  • the baseband processor 531 supports one or more of the wireless communication standards including, but not limited to, GSM, CDMA 1x, CDMA2000, WCDMA, HSPA, LTE, and the like.
  • the radio frequency module 532 includes a radio frequency circuit that implements functions of radio frequency transceiver, frequency synthesis, power amplification, etc., and the radio frequency circuit can be packaged in a radio frequency chip. In other embodiments, some or all of the RF circuitry included in the RF module 532 is integrated with the baseband processor 531 in a baseband chip.
  • the memory 520 generally includes a memory and an external memory, wherein the memory may be a random access memory (RAM), a read only memory (ROM), a cache (CACHE), etc., and the external storage may be a hard disk, an optical disk, a USB disk, a floppy disk or a tape drive. Wait.
  • the computer executable program is usually stored on the external memory, and the application processor 510 loads the computer executable program from the external memory into the memory before executing.
  • the communication subsystem 530 is used to receive data from the outside or to transmit data of the terminal device 102 to an external device.
  • the terminal device 102 will typically include both the communication subsystem 530 and the Wi-Fi module 550 to support both cellular network access and WLAN access.
  • the terminal device 500 may also include only one of the communication subsystem 530 and the Wi-Fi module 550 for cost or other considerations.
  • the terminal device 102 further includes a display 560 for displaying information input by the user or information provided to the user, various menu interfaces of the terminal device 102, and the like.
  • the display 560 can be a liquid crystal display (LED) or an organic light-emitting diode (OLED).
  • the touch panel can be overlaid on the display 560 to form a touch display.
  • the terminal device 102 may further include a camera 580 for taking a photo or video, one or more sensors 570 such as a gravity sensor, an acceleration sensor, a light sensor, and the like.
  • sensors 570 such as a gravity sensor, an acceleration sensor, a light sensor, and the like.
  • terminal device 102 may include fewer or more components than those shown in FIG. 5, and the terminal device illustrated in FIG. 5 is only shown as disclosed in the embodiments of the present application. Multiple implementations of more relevant components.
  • the computer executable program stored by the memory 520 includes an operating system and an application.
  • the protocol stack program is a separate computer executable program, and the operating system calls the protocol stack program through the interface for message processing.
  • protocol stack programs can also be included in the operating system as part of the operating system kernel.
  • the protocol stack program can be divided into multiple modules according to the protocol level or function, and each module implements a layer protocol function, for example, the network layer module is used to implement a network layer protocol (for example, IP protocol), and the transport layer module is used. Implement a transport layer protocol (for example, TCP or UDP protocol), and so on.
  • the hardware driver of the terminal device 102 adds the message to the cache queue and notifies the operating system, and the operating system uses the system call interface to schedule each module of the protocol stack to The workflow described in the embodiment of FIG. 3 is performed.
  • computer executable program as used in the embodiments of the present invention should be interpreted broadly to include, but not limited to, instructions, instruction sets, codes, code segments, subroutines, software modules, applications, software packages. , threads, processes, functions, firmware, middleware, and more.
  • FIG. 6 shows a structural diagram of an exemplary hardware architecture of a server for implementing a content recommendation method according to an embodiment of the present application.
  • server 106 can include input device 601, input interface 602, central processor 603, memory 604, output interface 605, and output device 606.
  • the input interface 602, the central processing unit 603, the memory 604, and the output interface 605 are connected to each other through a bus 610.
  • the input device 601 and the output device 606 are connected to the bus 610 through the input interface 602 and the output interface 605, respectively, and the computing device 600.
  • the other components are connected.
  • the terminal device 102 and the server 106 can also be implemented to include: a memory storing computer executable instructions; and at least one processor that can implement the combination of FIG. 3 or FIG. 4 when executing the computer executable instructions.
  • the server 106 for implementing the content recommendation function according to an embodiment of the present application may be a virtual functional unit, for example, a virtual network function (VNF built on a general hardware resource by virtualization technology). ) or container.
  • FIG. 7 shows a schematic diagram of another example architecture of a server in accordance with an embodiment of the present application.
  • a hardware layer having multiple physical hosts is composed of a hardware layer 710 (also referred to as an infrastructure layer), a virtual machine monitor (VMM) 720, and a plurality of virtual machines (Virtual Machines, The VM) runs on the hardware layer 710.
  • Each virtual machine can be viewed as a separate computer that can invoke resources at the hardware layer to implement specific functions.
  • the hardware layer includes, but is not limited to, an I/O device, a CPU, and a memory.
  • the server 106 may specifically be a virtual machine such as a VM 740.
  • the VM 740 also runs a computer executable program, and the VM 740 runs the computer executable program by calling resources such as CPU, memory, etc. in the hardware layer 710, thereby implementing the content query and communication functions described in connection with FIGS. 2 and 4.
  • the disclosed systems, devices, and methods may be implemented in other manners.
  • the device embodiments described above are merely illustrative.
  • the division of the unit is only a logical function division.
  • there may be another division manner for example, multiple units or components may be combined or Can be integrated into another system, or some features can be ignored or not executed.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or unit, or an electrical, mechanical or other form of connection.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the embodiments of the present invention.
  • each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

公开了一种内容推荐方法和系统。该内容推荐方法包括:在第一时段内采集终端设备的第一用户数据,第一用户数据包括在第一时段内使用终端设备的用户的个人数据、行为信息、以及实时情境相关信息中的至少一种;基于第一用户数据,从终端设备保存的内容推荐模型中提取至少一项内容特征,至少一项内容特征为表征针对终端设备的用户的推荐内容的内容特征,其中,内容推荐模型是终端设备根据历史用户数据构建的;以及将至少一项内容特征发送给服务器,并从服务器接收针对终端设备的用户的推荐内容,其中,针对终端设备的用户的推荐内容是服务器根据至少一项内容特征查询得到的。

Description

内容推荐方法和系统 技术领域
本申请涉及通信领域,尤其涉及一种内容推荐方法和系统。
背景技术
随着诸如,手机、平板电脑之类的终端设备的普及以及移动互联网技术的发展,终端设备正逐渐成为人们获取信息的重要平台。例如,人们可以通过终端设备进行新闻浏览、书籍阅读、视频收看、音乐聆听、以及社会交流等活动。但是,由于终端设备的信息处理能力非常有限,网络服务和信息内容的不断膨胀将在终端设备中引发“信息过载”,这会严重影响终端设备的用户体验和移动互联网的资源利用率。
内容推荐系统通过机器学习技术分析和理解用户的需求,并基于用户的需求对信息进行筛选和过滤,从而将符合用户的需求的内容推荐给用户。内容推荐系统成为解决信息过载问题的重要手段之一,在互联网和终端领域得到了广泛关注和应用。
通常,在现有的内容推荐系统中,终端设备收集用户数据,然后将用户数据上传到服务器;服务器基于用户数据进行分析建模和推荐计算,以挖掘用户与项目之间的二元关系(user-item),进而从大量内容数据中发现用户需求相关的项目,例如,新闻、视频、在线商品等,再将推荐结果发送到终端设备,以满足终端设备的用户的个性化需求。由于终端设备通常需要收集包含有用户的位置、状态、行为等信息的用户数据并上传至服务器,而这些用户数据可能涉及用户的隐私信息,例如,用户的位置信息里包含的用户的行动轨迹、浏览记录等,因此很可能造成用户隐私泄漏。另一方面,从终端设备制造商的角度,需要尽可能把用户信息保存在用户自己的终端设备中,从而充分保护用户隐私。如果某款终端设备不能保护隐私,则用户会选择其他厂商的终端设备,这对终端设备制造商的影响会非常巨大。因此,现有的内容推荐系统难以满足用户和终端设备制造商对于用户隐私保护的需求。
发明内容
本申请的实施例提供了一种内容推荐方法和系统,能够在保护终端设备的用户的个人隐私安全的条件下向终端设备的用户推荐相应的内容。
第一方面,本申请实施例提供了一种内容推荐方法,包括:在第一时段内采集终端设备的第一用户数据,第一用户数据包括在第一时段内使用终端设备的用户的个人数据、行为信息、以及实时情境相关信息中的至少一种;基于第一用户数据,从终端设备保存的内容推荐模型中提取至少一项内容特征,该至少一项内容特征为表征针对终端设备的用户的推荐内容的内容特征,其中,内容推荐模型是终端设备根据历史用户数据构建的;以及将至少一项内容特征发送给服务器,并从服务器接收针对终端设备的用户的推荐内容,其中,针对终端设备的用户的推荐内容是服务器根据至少一项内容特征查询得到的。
在根据本申请实施例的第一方面的内容推荐方法中,无需将用户数据从终端设备上传到服务器,因此充分保证了终端设备的用户的个人隐私安全。另外,利用终端设备的计算 能力来构建内容推荐模型并进行推荐计算,因此充分利用了终端设备的计算能力,同时解放了服务器的计算能力。
在第一种可能的实现方式中,上述内容推荐方法还包括:在第二时段内采集终端设备的第二用户数据,第二时段早于第一时段,第二用户数据为历史用户数据;以及基于历史用户数据、以及历史用户数据所对应的多项内容特征,构建内容推荐模型。
结合上述可能的实现方式,在第二种可能的实现方式中,上述内容推荐方法还包括:在第三时段内采集终端设备的第三用户数据,第三时段晚于第二时段;以及基于第三用户数据、以及第三用户数据所对应的多项内容特征,更新内容推荐模型。这里,对于内容推荐模型的更新可以使得针对终端设备的用户的推荐内容更加符合终端设备的用户的需求。
结合上述可能的实现方式,在第三种可能的实现方式中,历史用户数据包括多个历史用户数据样本,每个历史用户数据样本包括多个样本特征值,内容推荐模型包括多个历史用户数据样本所归属的各个样本类别与内容特征之间的对应关系。基于多个历史用户数据样本、以及多个历史用户数据样本所对应的各项内容特征构建内容推荐模型包括:对于每个历史用户数据样本,利用该历史用户数据样本所对应的特征权值向量和该历史用户数据样本中包括的多个样本特征值,获取该历史用户数据样本的加权值,其中,该历史用户数据样本所对应的特征权值向量包括分别与该历史用户数据样本中包括的多个样本特征值相对应的多个权重值;基于多个历史用户数据样本的加权值,将多个历史用户数据样本划分为多个样本类别;对于每个样本类别,将归属于该样本类别的各个历史用户数据样本所对应的内容特征与该样本类别相关联。
结合上述可能的实现方式,在第四种可能的实现方式中,将多个历史用户数据样本划分为多个样本类别包括:通过聚类算法将多个历史用户数据样本划分为多个样本类别。在这种情况下,内容推荐模型还包括每个样本类别的聚类中心点、归属于该样本类别的用户数据样本的数目、以及与该样本类别相关联的各项内容特征在该样本类别中出现的概率。这里,使用聚类算法来构建内容推荐模型,避免了传统的内容推荐系统中使用的大规模矩阵运算,因此减轻了内容推荐模型的构建的计算量过大所引起的终端设备的功耗问题。
结合上述可能的实现方式,在第五种可能的实现方式中,第三用户数据包括多个第三用户数据样本,多个第三用户数据样本中的每个第三用户数据样本包括多个样本特征值,基于多个第三用户数据样本中的任意一个第三用户数据样本、以及该第三用户数据样本所对应的内容特征更新内容推荐模型包括:利用该第三用户数据样本所对应的特征权值向量和该第三用户数据样本中包括的多个样本特征值,获取该第三用户数据样本的加权值,其中,该第三用户数据样本所对应的特征权值向量包括分别与该第三用户数据样本中包括的多个样本特征值相对应的多个权重值;计算该第三用户数据样本的加权值与各个样本类别的聚类中心点之间的距离;将该第三用户数据样本添加到与其距离最近的一个样本类别中;利用该第三用户数据样本的加权值更新该样本类别的聚类中心点,并利用该第三用户数据样本所对应的内容特征更新与该样本类别相关联的各项内容特征在该样本类别中出现的概率。
结合上述可能的实现方式,在第六种可能的实现方式中,第一用户数据包括一个第一用户数据样本,第一用户数据样本包括多个样本特征值,基于第一用户数据从内容推荐模 型提取至少一项内容特征包括:利用第一用户数据样本所对应的特征权值向量和第一用户数据样本中包括的多个样本特征值,获取第一用户数据样本的加权值,其中,第一用户数据样本所对应的特征权值向量包括分别与第一用户数据样本中包括的多个样本特征值相对应的多个权重值;计算第一用户数据样本的加权值与各个样本类别的聚类中心点之间的距离,作为第一用户数据样本与各个样本类别之间的相似度;找出与第一用户数据样本之间的相似度最高的一个样本类别,并提取与该样本类别相关联的至少一项内容特征,作为表征针对用户的推荐内容的至少一项内容特征。
第二方面,本申请的实施例提供了一种内容推荐方法,包括:接收表征针对终端设备的用户的推荐内容的至少一项内容特征;基于至少一项内容特征,在预先存储有内容特征、内容特征所表征的内容、以及它们之间的对应关系的存储器中查找针对终端设备的用户的推荐内容;以及将针对终端设备的用户的推荐内容返回给终端设备。
在第一种可能的实现方式中,使用搜索引擎的查询功能和数据库的查询功能中的一种或两种,查找针对终端设备的用户的推荐内容。
在根据本申请实施例的第二方面的内容推荐方法中,服务器仅用于内容的存储、查询和发送,而无需执行构建内容推荐模型以及进行推荐计算等相关处理,因此在充分利用服务器的存储能力的同时减轻了服务器的计算处理的负担。
第三方面,本申请的实施例提供了一种终端设备,包括至少一个处理器、存储器、以及存储在存储器上并可被至少一个处理器执行的计算机可执行指令,其中,至少一个处理器执行计算机可执行指令,以实现上述由终端设备执行的内容推荐方法。
第四方面,本申请的实施例提供了一种服务器,包括至少一个处理器、存储器、以及存储在存储器上并可被至少一个处理器执行的计算机可执行指令,其中,至少一个处理器执行计算机可执行指令,以实现上述由服务器执行的内容推荐方法。
第五方面,本申请的实施例提供了一种计算机可读存储介质,其上存储有计算机程序,其中,该计算机程序在被处理器执行时实现上述由终端设备执行的内容推荐方法。
第六方面,本申请的实施例提供一种内容推荐系统,包括如上述终端设备和服务器。
附图说明
通过阅读以下参照附图对本申请的非限制性实施例所作的详细描述,本申请的其它特征、目的和优点将会变得更明显,其中,相同或相似的附图标记表示相同或相似的特征。
图1示出了根据本申请实施例的内容推荐系统的架构示意图;
图2示出了用于实现根据本申请实施例的内容推荐方法的终端设备和服务器的架构示意图;
图3示出了由图2所示的终端设备执行的内容推荐过程的流程图;
图4示出了由图2所示的服务器执行的内容推荐过程的流程图;
图5示出了用于实现根据本申请实施例的内容推荐方法的终端设备的示例硬件架构的示意图;
图6示出了用于实现根据本申请实施例的内容推荐方法的服务器的示例硬件架构的示意图;
图7示出了用于实现根据本申请实施例的内容推荐方法的服务器的另一示例架构的示意图。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚,下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行描述。
本申请一个实施方式中所描述的特征、结构或特性可以以任何合适的方式结合在一个或更多实施例中。在下面的描述中,提供许多具体细节从而给出对本发明的实施例的充分理解。然而,本领域技术人员将意识到,可以通过减少所述特定细节中的一个或多个实践本发明的技术方案。
需要说明的是,在不冲突的情况下,本申请中的实施例及实施例中的特征可以相互组合。
图1示出了根据本申请实施例的内容推荐系统的架构示意图。如图1所示,内容推荐系统100包括端侧和云侧两部分,其中,端侧包括至少一个终端设备102(例如,图1中所示的终端设备102-1、102-1等),云侧是由至少一台服务器106(例如,图1中所示的服务器106-1、106-2、和106-3等)组成的计算/存储系统。终端设备102通过无线接入点104或者因特网(Internet)与服务器106通信,以获取所需的服务或内容。
基于图1所示的系统架构,本申请提出了一种内容推荐方法,其中:由终端设备采集终端设备的用户数据,该用户数据包含有关终端设备的用户在本地存储的包括图片、文本、视频、应用程序等在内的各项内容、终端设备的用户在终端设备上的操作行为、以及终端设备的用户所处的包括物理环境、网络环境、终端环境等在内的上下文情境等方面的信息;由终端设备基于用户数据及其对应的各项内容特征构建或更新针对终端设备的用户的内容推荐模型,从所构建的内容推荐模型提取表征针对终端设备的用户的推荐内容的至少一项内容特征,并将所提取的内容特征发送给服务器;由服务器使用查询技术查找来自终端设备的内容特征所表征的内容,并将查找到的内容返回给终端设备。
图2示出了用于实现根据本申请实施例的内容推荐方法的终端设备和服务器的架构示意图。图3示出了由图2所示的终端设备实现的内容推荐过程的流程图。下面结合图2和图3,详细描述上述内容推荐方法中由终端设备执行的具体处理的实现细节。
如图2所示,终端设备102包括数据采集单元200、模型构建单元202、特征推荐单元204、端侧通信单元206、以及内容输出单元208,其中:数据采集单元200被配置为采集终端设备的用户数据;模型构建单元202被配置为基于历史用户数据以及历史用户数据所对应的多项内容特征构建内容推荐模型(即,针对终端设备的用户的内容推荐模型)、基于最新用户数据以及最新用户数据所对应的多项内容特征更新内容推荐模型、以及保存内容推荐模型;特征推荐单元204被配置为基于当前用户数据,从内容推荐模型提取至少一项内容特征,作为表征推荐内容(即,针对终端设备的用户的推荐内容)的内容特征;端侧通信单元206被配置为将特征推荐单元204提取的、表征推荐内容的内容特征发送给服务器106,并从服务器106接收推荐内容;内容输出单元208被配置为从端侧通信单元206接收推荐内容,并将推荐内容呈现给终端设备的用户。
在一些实施例中,数据采集单元200可以周期性地或者实时地从终端设备采集用户数据;模型构建单元202可以在数据采集单元200采集到足以构建内容推荐模型的历史用户数据时,基于历史用户数据及其对应的多项内容特征构建内容推荐模型,或者基于在当前时刻之前的预定时段生成的历史用户数据及其对应的多项内容特征构建内容推荐模型。
在一些实施例中,在模型构建单元202完成内容推荐模型的构建之后,如果数据采样单元200进一步从终端设备采集到足以更新内容推荐模型的最新用户数据,则模型构建单元202可以基于最新用户数据及其对应的各项内容特征对内容推荐模型进行更新。
在一些实施例中,在终端设备的用户通过一种或多种预定方式触发内容推荐时,特征推荐单元204可以基于在当前时刻之前很短的时段,例如,2秒内采集的当前用户数据从内容推荐模型提取至少一项内容特征,作为表征针对终端设备的用户的推荐内容的内容特征,并经由端侧通信单元206将所提取的内容特征发送给服务器。
这里,当前用户数据是指在当前时刻之前的第一时段内生成的第一用户数据;历史用户数据是指在当前时刻之前的第二时段内生成的第二用户数据,其中,第二时段早于第一时段并且通常比第一时段的持续时间长得多;最新用户数据是指在当前时刻之前的第三时段内生成的第三用户数据,其中,第三时段晚于第二时段并且早于或者包含第一时段。例如,假设当前时刻是2015年4月26日12:00,则第一时段可以是2015年4月26日11:55-11:58,第二时段可以是2015年1月1日00:00至2015年3月31日23:59,第三时段可以是2015年4月1日00:00至2015年4月15日23:59。
如图3所示,图2中所示的终端设备102可以通过以下处理来实现针对其用户的内容推荐:S302,在第一时段内采集第一用户数据;S304,基于第一用户数据,从终端设备保存的内容推荐模型中提取至少一项内容特征,作为表征针对终端设备的用户的推荐内容的内容特征;以及S306,将所提取的至少一项内容特征发送给服务器,并从服务器接收针对终端设备的用户的推荐内容,其中,该推荐内容是服务器根据由终端设备发送的至少一项内容特征查询得到的。
这里,用户数据包括用户行为数据、用户情境数据、以及用户个人数据中的一种或多种,其中:用户行为数据包含有关终端设备的用户在终端设备上的操作行为的信息,用户情境数据包含有关终端设备的用户所处的包括物理环境、网络环境、终端环境等在内的上下文情境的信息,用户个人数据包含有关终端设备的用户在本地存储的包括图片、文本、视频、应用程序等在内的各项内容的信息。
下面,分别介绍各种用户数据的示例表征方式。
(1)用户行为数据可以包含用户在终端设备上的操作行为的各种相关信息,针对不同的推荐业务(例如,推荐新闻、推荐应用程序等)各不相同,但整体模型可以表征如表1所示:
表1用户行为数据
Figure PCTCN2018083108-appb-000001
Figure PCTCN2018083108-appb-000002
(2)用户情景数据可以包含终端设备的用户所处的上下文情境的相关信息,包括物理环境数据、终端状态数据、用户状态数据等,其整体模型可以表征如表2所示:
表2用户情境数据
Figure PCTCN2018083108-appb-000003
Figure PCTCN2018083108-appb-000004
(3)用户个人数据可以包含有关终端设备的用户在本地存储的各项内容的相关信息,可以根据推荐业务的不同选择性地使用,其整体模型可以表征如表3所示:
表3用户个人数据
Figure PCTCN2018083108-appb-000005
这里,用户数据所对应的内容特征可以由终端设备在从相应的内容源获取用户数据所 对应的内容数据的同时从该内容源获取,也可以由终端设备直接从用户数据所对应的内容数据提取。通常,终端设备所存储的内容特征的集合是服务器所存储的内容特征的集合的子集,其整体模型可以表征如表4所示:
表4内容特征
Figure PCTCN2018083108-appb-000006
通常,用户数据是一个或多个用户数据样本的集合,每个用户数据样本包括在同一时间采集的用户行为数据、用户情景数据、以及用户个人数据中的一者或多者。因此,在一些实施例中,可以按照表达式(1)所示的方式对用户数据样本进行表征:
{X}={用户行为数据,用户情境数据,用户个人数据}表达式(1)
这里,X表示用户数据样本,用户行为数据、用户情境数据、以及用户个人数据是用户数据样本的样本特征值。应该理解的是,用户数据样本可以包括更多或更少的样本特征值,表达式(1)只是用户数据样本的一个示例表征。更一般地,可以将用户数据样本表征为表达式(2):
Figure PCTCN2018083108-appb-000007
其中,X i表示第i个用户数据样本,
Figure PCTCN2018083108-appb-000008
表示第i个用户数据样本中的第n个样本特征值。
通常,终端设备的用户在一天24小时之内和/或一年四季的不同时段和/或季节具有不同的活动规律,因而可能具有不同的偏好。在一些实施例中,可以针对不同的时段和/或季节构建不同的特征权值向量,以通过诸如,用户行为数据、用户情景数据、以及用户个人数据之类的样本特征值所对应的权重值的不同来至少部分地体现出终端设备的用户的偏好的不同。因此,可以对历史用户数据、最新用户数据、以及当前用户数据应用分别与它们存在时间上的对应关系的特征权值向量,来至少部分地体现出终端设备的用户在不同时间的不同偏好。
一般,与内容数据样本X i存在时间上的对应关系的特征权值向量可以表征为表达式(3):
Figure PCTCN2018083108-appb-000009
其中,W i表示与第i个用户数据样本X i存在时间上的对应关系的特征权值向量(即,第i个特征权值向量),
Figure PCTCN2018083108-appb-000010
表示第i个特征权值向量中与第i个用户数据样本X i中的第n个样本特征值相对应的特征权值(即,权重值)。
通常,每个用户数据样本可以对应一项内容特征,多个用户数据样本可以对应于同一项内容特征,包括多个用户数据样本的用户数据可以对应至少一项内容特征,并且每项内 容特征可以具有一个或多个特征值。在一些实施例中,可以按照表达式(4)所示的方式对内容特征进行表征:
Figure PCTCN2018083108-appb-000011
其中,Y i表示第i项内容特征,
Figure PCTCN2018083108-appb-000012
表示第i项内容特征中的第m个特征值。
在一些实施例中,假设数据采集单元200采集到q(q是大于1的整数)个历史用户数据样本X 1至X q,每个历史用户数据样本包括n个样本特征值,则模型构建单元202可以通过以下处理构建内容推荐模型:对于每个历史用户数据样本,例如,历史用户数据样本X i(1≤i≤q),利用历史用户数据样本X i所对应的特征权值向量W i和历史用户数据样本X i中包括的样本特征值
Figure PCTCN2018083108-appb-000013
获取历史用户数据样本X i的加权值S i;基于q个历史用户数据样本X 1至X q的加权值S 1至S q,将q个历史用户数据样本X 1至X q划分为多个样本类别,例如,样本类别C 1至C r;以及对于每个样本类别,例如,样本类别C j(1≤j≤r),将归属于样本类别C j的各个历史用户数据样本所对应的内容特征,例如,内容特征Y 1至Y k与样本类别C j相关联。
也就是说,内容推荐模型可以包括多个历史用户数据样本所归属的各个样本类别与内容特征之间的对应关系,创建内容推荐模型的过程即是将多个历史用户数据样本划分为多个样本类别,并将每个样本类别与归属于其的历史用户数据样本所对应的内容特征进行关联的过程。
在一些实施例中,模型构建单元202可以通过聚类算法将历史用户数据样本X 1至X q划分为多个样本类别C 1至C r,在这种情况下内容推荐模型还可以包括每个样本类别的聚类中心点、归属于该样本类别的用户数据样本的数目、以及与该样本类别相关联的各项内容特征在该样本类别中出现的概率。例如,可以按照表达式(5)所示的方式对内容推荐模型进行表征:
Figure PCTCN2018083108-appb-000014
其中,使用键值对的形式表示样本类别C j与归属于样本类别C j的用户数据样本所对应的内容特征之间的对应关系,C j表示第j个样本类别的类别标号,
Figure PCTCN2018083108-appb-000015
表示归属于样本类别C j的用户数据样本的数目,
Figure PCTCN2018083108-appb-000016
表示样本类别C j的聚类中心点,{Y k:p k}表示归属于样本类别C j的用户数据样本所对应的第k项内容特征Y k与该内容特征在样本类别C j中出现的概率p k,概率p k可由表达式(6)得出:
Figure PCTCN2018083108-appb-000017
这里,由于模型构建单元202使用聚类算法来构建内容推荐模型,避免了传统的内容推荐系统中使用的大规模矩阵运算,因此减轻了内容推荐模型的构建的计算量过大所引起的终端设备的功耗问题。
在一些实施例中,在模型构建单元202已经基于历史用户数据样本X 1至X q完成内容推荐模型的构建之后,如果数据采集单元200进一步采集到足以更新内容推荐模型的包括最新用户数据样本X q+i至X q+j的最新用户数据,则模型构建单元202可以通过以下处理更新内容推荐模型:对于每个最新用户数据样本,例如,最新用户数据样本X q+i,利用最新用户数据样本X q+i所对应的特征权值向量W q+i和最新用户数据样本X q+i中包括的样本特征值
Figure PCTCN2018083108-appb-000018
获取最新用户数据样本X q+i的加权值S q+i;计算最新用户数据样本X q+i的加权值S q+i与各个样本类别C 1至C r的聚类中心点
Figure PCTCN2018083108-appb-000019
Figure PCTCN2018083108-appb-000020
之间的距离,作为最新用户数据样本X q+i与各个样本类别C 1至C r之间的距离;将最新用户数据样本X q+i添加到与其距离最近的一个样本类别,例如,样本类别C j中;利用最新用户数据样本X q+i的加权值S q+i更新样本类别C j的聚类中心点
Figure PCTCN2018083108-appb-000021
并利用最新用户数据样本X q+i所对应的内容特征,例如,样本特征Y i更新与样本类别C j相关联的各项内容特征在样本类别C j中出现的概率。
在一些实施例中,在终端设备的用户通过一种或多种预定方式触发内容推荐的情况下,特征推荐单元204可以通过以下处理从内容推荐模型提取表征针对终端设备的用户的推荐内容的至少一项内容特征:利用当前用户数据样本X t(即,当前用户数据仅包括该用户数据样本)所对应的特征权值向量W t和当前用户数据样本X t中包括的样本特征值
Figure PCTCN2018083108-appb-000022
获取当前用户数据样本X t)的加权值S t;计算当前用户数据样本X t的加权值S t与各个样本类别C 1至C r的聚类中心点
Figure PCTCN2018083108-appb-000023
Figure PCTCN2018083108-appb-000024
之间的距离,作为当前用户数据样本X t与各个样本类别C 1至C r之间的相似度;找出与当前用户数据样本X t之间的相似度最高的一个样本类别,例如,样本类别C j,并提取与样本类别C j相关联的至少一项内容特征,作为表征针对终端设备的用户的推荐内容的至少一项内容特征。
例如,特征推荐单元204在找出与当前用户数据样本X t之间的相似度最高的样本类别 C j后,可以根据TOP-K推荐规则提取与样本类别C j相关联的K项内容特征,作为表征针对终端设备的用户的推荐内容的内容特征。例如,特征推荐单元204可以提取在样本类别C j中出现频率最高的两项内容特征,作为表征针对终端设备的用户的推荐内容的内容特征。
图4示出了由图2所示的服务器执行的内容推荐过程的流程图。下面,结合附图2和图4,详细描述上述内容推荐方法中由服务器执行的具体处理的实现细节。
如图2所示,服务器106包括云侧通信单元402和内容查询单元404,其中:云侧通信单元402被配置为从端侧通信单元206接收表征针对终端设备的用户的推荐内容的至少一项内容特征(即,执行步骤S402)以及向端侧通信单元206发送针对终端设备的用户的推荐内容(即,执行步骤S406);内容查询单元404被配置为基于表征针对终端设备的用户的推荐内容的至少一项内容特征,在预先存储有内容特征、内容特征所表征的内容、以及它们之间的对应关系的存储器中查找针对终端设备的用户的推荐内容(即,执行步骤S404)。
在一些实施例中,内容查询单元404可以基于表征针对终端设备的用户的推荐内容的至少一项内容特征,使用搜索引擎的查询功能和数据库的查询功能中的一种或两种查找针对终端设备的用户的推荐内容。例如,当表征针对终端设备的用户的推荐内容的内容特征被表示为特征关键字时,内容查询单元404可以使用搜索引擎的查询功能基于特征关键字搜索针对终端设备的用户的推荐内容;当表征针对终端设备的用户的推荐内容的内容特征被表示为关键字ID时,内容查询单元404可以先在存储有关键字ID、特征关键字、及它们之间的对应关系的表格(即,特征表)中查找相应的特征关键字,然后使用搜索引擎的查询功能基于查找到的特征关键字搜索针对终端设备的用户的推荐内容;或者当表征针对终端设备的推荐内容的内容特征被表示为关键字ID时,内容查询单元404可以先在存储有关键字ID、特征关键字、及它们之间的对应关系的表格中查找相应的特征关键字,然后使用数据库的查询功能从存储有特征关键字、内容、以及它们之间的对应关系的表格(即,内容表)中查找针对终端设备的用户的推荐内容。
表5特征表
Figure PCTCN2018083108-appb-000025
表6内容
Figure PCTCN2018083108-appb-000026
下面以推荐应用程序为例,说明作为终端设备102与服务器106协同工作实现内容推荐的具体过程。首先,终端设备102执行以下处理:采集用户数据并用采集到的用户数据更新内容推荐模型;基于用户数据从内容推荐模型提取表征针对终端设备的用户的推荐内容的关键字ID,即,内容特征由关键字ID表征,这里假设内容特征为“游戏,棋类”并且对应的关键字ID为“ID:2”;将所提取的关键字ID,即“ID:2”发送给服务器106。接着,服务器106执行以下处理:接收关键字ID,即“ID:2”,并查询内容特征和内容数据,得到“象棋,游戏,棋类”和“围棋,游戏,棋类”两个结果,“游戏,棋类”是内容特征所表征的内容;将“象棋,游戏,棋类”和“围棋,游戏,棋类”内容发送回终端设备102。然后,终端设备102在接收到服务器106反馈的内容后,将“象棋、围棋”两个应用程序推荐给终端设备的用户,完成整个推荐过程。
从以上示例中可以看出,不需要将任何包含有终端设备的用户的个人信息的用户数据从客户端发送到服务器,因此可以充分保护终端设备的用户的个人隐私安全。
在根据本申请实施例的内容推荐方案中,由终端设备完成用户数据的采集、内容推荐模型的建模、以及内容特征的推荐计算,而不需要将用户数据从终端设备上传到服务器,因此充分保证了终端设备的用户的个人隐私安全;终端设备与服务器协同工作,通过终端设备对内容特征的提取和服务器基于内容特征的内容查询,解决了终端设备的推荐内容匮乏的问题;另外,利用聚类算法进行内容推荐模型的建模,避免了矩阵运算,计算量小,既利用了终端设备的计算能力,又避免了终端设备的计算量大可能带来的能耗过高的问题。根据本申请实施例的内容推荐方案适合所有终端设备推荐场景,尤其适合终端设备制造商使用。
图5示出了用于实现根据本申请实施例的内容推荐方法的终端设备的示例性硬件架构的示意图。如图5所示,终端设备102的硬件架构可以包括应用处理器510、存储器520、通信子系统530、和电源管理子系统540,其中:存储器520存储有计算机可执行程序,该计算机可执行程序包括操作系统、协议栈程序、和应用程序;电源管理子系统540用于为整个终端设备102供电,具体可以为电源管理芯片;通信子系统530为终端设备102的基本通信单元。
在一些实施例中,通信子系统530为无线调制解调器(Modem),主要用于实现基带处理、调制解调、信号放大和滤波、均衡等功能。例如,通信子系统530可以包括基带处理器531、射频模块532、和天线533,其中,基带处理器531和应用处理器533可以集成在同一芯片中。在一些实施例中,基带处理器531和应用处理器510也可以分离式部署,例如,基带处理器531和应用处理器510作为两个独立的芯片通过核间通信方式进行信息的交互。在采用分离式部署方式的情况下,基带处理器531相当于应用处理器510的一个外设,两个处理器需要各自独立的外接存储器、以及软件升级接口。
在一些实施例中,射频模块532主要负责信号发送和接收;基带处理器531主要负责信号的处理,例如,信号的A/D、D/A转换、信号的编解码、信道编解码。基带处理器531支持无线通信标准中的一种或多种,这里的无线通信标准包括但不限于GSM、CDMA 1x、CDMA2000、WCDMA、HSPA、LTE等。在一些实施例中,射频模块532包括实现射频收发、频 率合成、功率放大等功能的射频电路,该射频电路可以封装在射频芯片中。在另一些实施例中,射频模块532包含的部分或全部射频电路和基带处理器531一起集成在基带芯片中。
存储器520一般包括内存和外存,其中,内存可以为随机存储器(RAM)、只读存储器(ROM)、以及高速缓存(CACHE)等,外存可以为硬盘、光盘、USB盘、软盘或磁带机等。计算机可执行程序通常被存储在外存上,应用处理器510会将计算机可执行程序从外存加载到内存后再执行。
可以理解的是,通信子系统530用于从外部接收数据或者将终端设备102的数据发送至外部设备。终端设备102通常会同时包含通信子系统530和Wi-Fi模块550,以同时支持蜂窝网络接入和WLAN接入。但出于成本或其它因素的考虑,终端设备500也可以只包含通信子系统530和Wi-Fi模块550中的一个。
可选地,终端设备102还包括显示器560,用于显示由用户输入的信息或提供给用户的信息以及终端设备102的各种菜单界面等。显示器560可为液晶显示器(Liquid Crystal Display,LED)或有机发光二极管(Organic Light-Emitt ing Diode,OLED)等。在其他一些实施例中,显示器560上可以覆盖触控面板,以形成触摸显示屏。
除以上之外,终端设备102还可以包括用于拍摄照片或视频的摄像头580、一个或多个传感器570,例如,重力传感器、加速度传感器、光传感器等。
此外,所属领域的技术人员可以理解的是,终端设备102可包括比图5中所示部件更少或更多的部件,图5所示的终端设备仅示出了与本申请实施例所公开的多个实现方式更加相关的部件。
具体地,如图5所示,存储器520存储的计算机可执行程序包括操作系统和应用程序。在一些场景下,协议栈程序为一个独立的计算机可执行程序,操作系统通过接口调用协议栈程序以进行报文处理。在一些场景下,协议栈程序也可以被包含在操作系统中,作为操作系统内核的一部分。其中,协议栈程序按照协议层级或功能又可以分为多个模块,每一个模块实现一层协议的功能,比如,网络层模块用于实现网络层协议(例如,IP协议),传输层模块用于实现传输层协议(例如,TCP或者UDP协议),等等。当通信子系统530或Wi-Fi模块550接收到报文后,终端设备102的硬件驱动程序将报文加入缓存队列,并通知操作系统,操作系统通过系统调用接口调度协议栈的各个模块,以执行图3相关的实施例所描述的工作流程。
需要说明的是,本发明实施例所使用的术语“计算机可执行程序”应被广泛地解释为包括但不限于:指令、指令集、代码、代码段、子程序、软件模块、应用、软件包、线程、进程、函数、固件、中间件等。
图6示出了用于实现根据本申请实施例的内容推荐方法的服务器的示例性硬件架构的结构图。如图6所示,服务器106可以包括输入设备601、输入接口602、中央处理器603、存储器604、输出接口605、以及输出设备606。其中,输入接口602、中央处理器603、存储器604、以及输出接口605通过总线610相互连接,输入设备601和输出设备606分别通过输入接口602和输出接口605与总线610连接,进而与计算设备600的其他组件连接。
也就是说,终端设备102和服务器106也可以被实现为包括:存储有计算机可执行指 令的存储器;以及至少一个处理器,该处理器在执行计算机可执行指令时可以实现结合图3或图4描述的内容推荐过程。
在一些实施例中,用于实现根据本申请实施例的内容推荐功能的服务器106可以为虚拟功能单元,例如,通过虚拟化技术在通用硬件资源上构建的虚拟网络功能(virtual network funct ion,VNF)或者容器。图7示出了根据本申请实施例的服务器的另一示例架构的示意图。
如图7所示,由一台有多台物理主机的硬件资源组成硬件层710(也称为基础设施层),虚拟机监视器(Virtual Machine Monitor,VMM)720以及若干虚拟机(Virtual Machine,VM)运行于硬件层710之上,每一台虚拟机可以看成是一台独立的计算机,可以调用硬件层的资源来实现特定的功能。其中,硬件层包括但不限于I/O设备、CPU和存储器。服务器106具体可以为一台虚拟机,例如VM 740。VM 740还运行有计算机可执行程序,VM 740通过调用硬件层710中的CPU、存储器等资源,来运行该计算机可执行程序,从而实现结合图2和图4描述的内容查询以及通信功能。
本领域普通技术人员可以意识到,结合本文中所公开的实施例描述的各示例的单元及算法步骤,能够以电子硬件、计算机软件或者二者的结合来实现,为了清楚地说明硬件和软件的可互换性,在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行,取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能,但是这种实现不应认为超出本发明的范围。
所属领域的技术人员可以清楚地了解到,为了描述的方便和简洁,上述描述的系统、装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
在本申请所提供的几个实施例中,应该理解到,所揭露的系统、装置和方法,可以通过其它的方式实现。例如,以上所描述的装置实施例仅仅是示意性的,例如,所述单元的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如多个单元或组件可以结合或者可以集成到另一个系统,或一些特征可以忽略,或不执行。另外,所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口、装置或单元的间接耦合或通信连接,也可以是电的,机械的或其它的形式连接。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本发明实施例方案的目的。
另外,在本发明各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以是两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用软件功能单元的形式实现。
以上所述,仅为本发明的具体实施方式,但本发明的保护范围并不局限于此,任何熟悉本技术领域的技术人员在本发明揭露的技术范围内,可轻易想到各种等效的修改或替换,这些修改或替换都应涵盖在本发明的保护范围之内。因此,本发明的保护范围应以权利要求的保护范围为准。

Claims (13)

  1. 一种内容推荐方法,包括:
    在第一时段内采集终端设备的第一用户数据,所述第一用户数据包括在所述第一时段内使用所述终端设备的用户的个人数据、行为信息、以及实时情境相关信息中的至少一种;
    基于所述第一用户数据,从所述终端设备保存的内容推荐模型中提取至少一项内容特征,所述至少一项内容特征为表征针对所述用户的推荐内容的内容特征,其中,所述内容推荐模型是所述终端设备根据历史用户数据构建的;以及
    将所述至少一项内容特征发送给服务器,并从所述服务器接收针对所述用户的推荐内容,其中,针对所述用户的推荐内容是所述服务器根据所述至少一项内容特征查询得到的。
  2. 根据权利要求1所述的内容推荐方法,其特征在于,还包括:
    在第二时段内采集所述终端设备的第二用户数据,所述第二时段早于所述第一时段,所述第二用户数据为所述历史用户数据;以及
    基于所述历史用户数据、以及所述历史用户数据所对应的多项内容特征,构建所述内容推荐模型。
  3. 根据权利要求2所述的内容推荐方法,其特征在于,还包括:
    在第三时段内采集所述终端设备的第三用户数据,所述第三时段晚于所述第二时段;以及
    基于所述第三用户数据、以及所述第三用户数据所对应的多项内容特征,更新所述内容推荐模型。
  4. 根据权利要求2或3所述的内容推荐方法,其特征在于,所述历史用户数据包括多个历史用户数据样本,所述多个历史用户数据样本中的每个历史用户数据样本包括多个样本特征值,所述内容推荐模型包括所述多个历史用户数据样本所归属的各个样本类别与内容特征之间的对应关系,
    基于所述历史用户数据、以及所述历史用户数据所对应的多项内容特征,构建所述内容推荐模型包括:
    对于所述多个历史用户数据样本中的每个历史用户数据样本,利用所述历史用户数据样本所对应的特征权值向量和所述历史用户数据样本中包括的多个样本特征值,获取所述历史用户数据样本的加权值,其中,所述历史用户数据样本所对应的特征权值向量包括分别与所述历史用户数据样本中包括的多个样本特征值相对应的多个权重值;
    基于所述多个历史用户数据样本的加权值,将所述多个历史用户数据样本划分为多个样本类别;以及
    对于所述多个样本类别中的每个样本类别,将归属于所述样本类别的各个历史用户数据样本所对应的内容特征与所述样本类别相关联。
  5. 根据权利要求4所述的内容推荐方法,其特征在于,将所述多个历史用户数据样本划分为多个样本类别包括:通过聚类算法将所述多个历史用户数据样本划分为所述多个样本类别,所述内容推荐模型还包括所述多个样本类别中的每个样本类别的聚类中心点、归属于所述样本类别的用户数据样本的数目、以及与所述样本类别相关联的各项内容特征在 所述样本类别中出现的概率。
  6. 根据权利要求3至5任一项所述的内容推荐方法,其特征在于,所述第三用户数据包括多个第三用户数据样本,所述多个第三用户数据样本中的每个第三用户数据样本包括多个样本特征值,
    基于所述多个第三用户数据样本中的任意一个第三用户数据样本、以及所述第三用户数据样本所对应的内容特征,更新所述内容推荐模型包括:
    利用所述第三用户数据样本所对应的特征权值向量和所述第三用户数据样本中包括的多个样本特征值,获取所述第三用户数据样本的加权值,其中,所述第三用户数据样本所对应的特征权值向量包括分别与所述第三用户数据样本中包括的多个样本特征值相对应的多个权重值;
    计算所述第三用户数据样本的加权值与各个样本类别的聚类中心点之间的距离;
    将所述第三用户数据样本添加到与其距离最近的一个样本类别中;
    利用所述第三用户数据样本的加权值更新所述样本类别的聚类中心点,并利用所述第三用户数据样本所对应的内容特征更新与所述样本类别相关联的各项内容特征在所述样本类别中出现的概率。
  7. 根据权利要求1至6任一项所述的内容推荐方法,其特征在于,所述第一用户数据包括一个第一用户数据样本,所述第一用户数据样本包括多个样本特征值,
    基于所述第一用户数据,从所述内容推荐模型中提取所述至少一项内容特征包括:
    利用所述第一用户数据样本所对应的特征权值向量和所述第一用户数据样本中包括的多个样本特征值,获取所述第一用户数据样本的加权值,其中,所述第一用户数据样本所对应的特征权值向量包括分别与所述第一用户数据样本中包括的多个样本特征值相对应的多个权重值;
    计算所述第一用户数据样本的加权值与各个样本类别的聚类中心点之间的距离,作为所述第一用户数据样本与各个样本类别之间的相似度;
    找出与所述第一用户数据样本之间的相似度最高的一个样本类别,并提取与所述样本类别相关联的至少一项内容特征,作为表征针对所述用户的推荐内容的内容特征。
  8. 一种内容推荐方法,包括:
    接收表征针对终端设备的用户的推荐内容的至少一项内容特征;
    基于所述至少一项内容特征,在预先存储有内容特征、内容特征所表征的内容、以及它们之间的对应关系的存储器中查找针对所述用户的推荐内容;以及
    将针对所述用户的推荐内容返回给所述终端设备。
  9. 根据权利要求8所述的内容推荐方法,其特征在于,使用搜索引擎的查询功能和数据库的查询功能中的至少一种,查找针对所述终端设备用户的推荐内容。
  10. 一种终端设备,包括至少一个处理器、存储器、以及存储在所述存储器上并可被所述至少一个处理器执行的计算机可执行指令,其特征在于,所述至少一个处理器执行所述计算机可执行指令,以实现权利要求1至7中任一项所述的内容推荐方法。
  11. 一种服务器,包括至少一个处理器、存储器、以及存储在所述存储器上并可被所 述至少一个处理器执行的计算机可执行指令,其特征在于,所述至少一个处理器执行所述计算机可执行指令,以实现权利要求8或9所述的内容推荐方法。
  12. 一种计算机可读存储介质,其上存储有计算机程序,其特征在于,该计算机程序在被处理器执行时实现权利要求1至7中任一项所述的内容推荐方法。
  13. 一种内容推荐系统,其特征在于,包括如权利要求10所述的终端设备、以及如权利要求11所述的服务器。
PCT/CN2018/083108 2017-06-09 2018-04-13 内容推荐方法和系统 WO2018223772A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710433188.2 2017-06-09
CN201710433188.2A CN109033103A (zh) 2017-06-09 2017-06-09 内容推荐方法和系统

Publications (1)

Publication Number Publication Date
WO2018223772A1 true WO2018223772A1 (zh) 2018-12-13

Family

ID=64566716

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/083108 WO2018223772A1 (zh) 2017-06-09 2018-04-13 内容推荐方法和系统

Country Status (2)

Country Link
CN (1) CN109033103A (zh)
WO (1) WO2018223772A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109948057A (zh) * 2019-03-21 2019-06-28 北京地平线机器人技术研发有限公司 感兴趣内容推送方法、装置及电子设备和介质
CN112052387A (zh) * 2020-08-17 2020-12-08 腾讯科技(深圳)有限公司 一种内容推荐方法、装置和计算机可读存储介质
CN116600020A (zh) * 2023-07-13 2023-08-15 支付宝(杭州)信息技术有限公司 协议生成方法、端云协同推荐方法及装置

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109684549A (zh) * 2018-12-24 2019-04-26 拉扎斯网络科技(上海)有限公司 目标数据预测方法、装置、电子设备及计算机存储介质
CN113573106B (zh) * 2020-04-28 2023-03-21 北京达佳互联信息技术有限公司 用于多媒体信息的模型更新方法、装置及服务器
CN112084413B (zh) * 2020-09-16 2024-02-13 腾讯科技(深圳)有限公司 一种信息推荐的方法、装置及存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104036002A (zh) * 2014-06-16 2014-09-10 深圳市英威诺科技有限公司 一种智能推荐数据的技术方法
CN105046600A (zh) * 2015-07-09 2015-11-11 传成文化传媒(上海)有限公司 一种酒店用电视的个性化信息推荐方法及系统
CN105912667A (zh) * 2016-04-12 2016-08-31 玉环看知信息科技有限公司 一种信息推荐方法、装置及移动终端
CN106201465A (zh) * 2016-06-23 2016-12-07 扬州大学 面向开源社区的软件项目个性化推荐方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102609523B (zh) * 2012-02-10 2018-03-06 上海视畅信息科技有限公司 基于物品分类和用户分类的协同过滤推荐方法
CN103106259B (zh) * 2013-01-25 2016-01-20 西北工业大学 一种基于情境的移动网页内容推荐方法
CN105608121B (zh) * 2015-12-14 2020-09-25 东软集团股份有限公司 一种个性化推荐方法及装置
CN105653626A (zh) * 2015-12-28 2016-06-08 深圳市金立通信设备有限公司 一种内容推送方法及终端

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104036002A (zh) * 2014-06-16 2014-09-10 深圳市英威诺科技有限公司 一种智能推荐数据的技术方法
CN105046600A (zh) * 2015-07-09 2015-11-11 传成文化传媒(上海)有限公司 一种酒店用电视的个性化信息推荐方法及系统
CN105912667A (zh) * 2016-04-12 2016-08-31 玉环看知信息科技有限公司 一种信息推荐方法、装置及移动终端
CN106201465A (zh) * 2016-06-23 2016-12-07 扬州大学 面向开源社区的软件项目个性化推荐方法

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109948057A (zh) * 2019-03-21 2019-06-28 北京地平线机器人技术研发有限公司 感兴趣内容推送方法、装置及电子设备和介质
CN109948057B (zh) * 2019-03-21 2022-03-01 北京地平线机器人技术研发有限公司 感兴趣内容推送方法、装置及电子设备和介质
CN112052387A (zh) * 2020-08-17 2020-12-08 腾讯科技(深圳)有限公司 一种内容推荐方法、装置和计算机可读存储介质
CN112052387B (zh) * 2020-08-17 2024-03-26 腾讯科技(深圳)有限公司 一种内容推荐方法、装置和计算机可读存储介质
CN116600020A (zh) * 2023-07-13 2023-08-15 支付宝(杭州)信息技术有限公司 协议生成方法、端云协同推荐方法及装置
CN116600020B (zh) * 2023-07-13 2023-10-10 支付宝(杭州)信息技术有限公司 协议生成方法、端云协同推荐方法及装置

Also Published As

Publication number Publication date
CN109033103A (zh) 2018-12-18

Similar Documents

Publication Publication Date Title
WO2018223772A1 (zh) 内容推荐方法和系统
CN108228270B (zh) 启动资源加载方法及装置
CN111125269B (zh) 一种数据管理方法、血缘关系显示方法和相关装置
US9477787B2 (en) Method and apparatus for information clustering based on predictive social graphs
CN105389325A (zh) 内容搜索方法和实施内容搜索方法的电子装置
EP2618278B1 (en) Synchronizing endpoint data stores having disparate schemas
CN110782289B (zh) 一种基于用户画像的业务推荐方法和系统
US20140188626A1 (en) Method and apparatus for secure advertising
CN112307311A (zh) 一种信息搜索方法、装置、设备及存储介质
CN110633438A (zh) 一种新闻事件处理的方法、终端、服务器及存储介质
CN107666515B (zh) 图像处理方法和装置、计算机设备、计算机可读存储介质
CN114694226B (zh) 一种人脸识别方法、系统及存储介质
CN117093766A (zh) 问诊平台的信息推荐方法、相关装置及存储介质
CN115658348A (zh) 微服务调用方法、相关装置及存储介质
CN116071614A (zh) 样本数据处理方法、相关设备及存储介质
CN115546516A (zh) 人员聚档方法、装置、计算机设备及存储介质
CN116028147A (zh) 应用程序的推荐方法和电子设备
CN103685472A (zh) 用于提供移动设备所对应的资源信息的方法与设备
CN114756714A (zh) 一种图数据的处理方法、装置以及存储介质
CN110322039B (zh) 一种点击率预估方法、服务器及计算机可读存储介质
CN115525554B (zh) 模型的自动化测试方法、系统及存储介质
CN116450808B (zh) 一种数据的处理方法、装置以及存储介质
CN104965715A (zh) 更新内置搜索引擎的方法和装置
CN115050079B (zh) 人脸识别方法、装置及存储介质
WO2013053254A1 (en) Index creation method and system

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18813654

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18813654

Country of ref document: EP

Kind code of ref document: A1