WO2017041484A1 - 一种实时信息的推荐方法、装置和系统 - Google Patents

一种实时信息的推荐方法、装置和系统 Download PDF

Info

Publication number
WO2017041484A1
WO2017041484A1 PCT/CN2016/078482 CN2016078482W WO2017041484A1 WO 2017041484 A1 WO2017041484 A1 WO 2017041484A1 CN 2016078482 W CN2016078482 W CN 2016078482W WO 2017041484 A1 WO2017041484 A1 WO 2017041484A1
Authority
WO
WIPO (PCT)
Prior art keywords
interest
real
user
time information
information
Prior art date
Application number
PCT/CN2016/078482
Other languages
English (en)
French (fr)
Inventor
胡雨成
Original Assignee
腾讯科技（深圳）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 腾讯科技（深圳）有限公司 filed Critical 腾讯科技（深圳）有限公司
Publication of WO2017041484A1 publication Critical patent/WO2017041484A1/zh
Priority to US15/868,729 priority Critical patent/US11003726B2/en

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2457—Query processing with adaptation to user needs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2457—Query processing with adaptation to user needs
    • G06F16/24575—Query processing with adaptation to user needs using context
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2457—Query processing with adaptation to user needs
    • G06F16/24578—Query processing with adaptation to user needs using ranking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation

Definitions

  • the present invention relates to the field of communications technologies, and in particular, to a method, device, and system for recommending real-time information.
  • the existing technology proposes various information recommendation schemes.
  • the existing recommendation algorithms can be mainly divided into two categories, one is a behavior-based recommendation algorithm, and the other is a content-based recommendation algorithm.
  • the behavior-based recommendation algorithm mainly collects the similarity between each information in the recommendation pool by counting the behavior of the user, and then recommends the information with high similarity to the information corresponding to the behavior to the user.
  • the content-based recommendation algorithm mainly searches for various types of information, and mines the user's interests to determine the keywords that the user is interested in, and then based on the keywords and various types of information. The word is calculated and recommended to the user.
  • the inventors of the present invention found that the existing recommendation scheme either relies on a large number of users to participate, or does not consider the user's interest change, and therefore, the timeliness is poor, and For real-time information, such as news, it has the characteristics of one-time consumption (that is, the news will be read only once for news of the same content), and timeliness is extremely important. Therefore, for real-time information, existing recommendation schemes The recommendation is not good.
  • the embodiment of the invention provides a method, a device and a system for recommending real-time information, which can improve the timeliness, and recommend the real-time information that the user is most interested in to the user flexibly, accurately and in time, and greatly improve the recommendation effect.
  • An embodiment of the present invention provides a method for recommending real-time information, including:
  • Real-time information is recommended to the user based on the user interest.
  • the embodiment of the present invention further provides a recommendation device for real-time information, including:
  • An obtaining unit configured to acquire user behavior data
  • An operation unit configured to separately calculate a short-term interest, a long-term interest, and a real-time interest of the user according to the user behavior data
  • a determining unit configured to determine a user interest according to the short-term interest, long-term interest, and real-time interest of the user
  • a recommendation unit for recommending real-time information to the user based on the user interest.
  • the embodiment of the present invention further provides a recommendation system for real-time information, including any recommendation device for real-time information provided by the embodiment of the present invention.
  • an embodiment of the present invention further provides a storage medium, where processor executable instructions are stored, and the processor executable instructions are configured to perform the following operations:
  • Real-time information is recommended to the user based on the user interest.
  • the embodiment of the present invention uses the user behavior data to calculate the short-term interest, long-term interest, and real-time interest of the user according to the user behavior data, and then determines the user interest according to the short-term interest, long-term interest, and real-time interest of the user, and based on the user interest.
  • the user recommends real-time information; because the program calculates the user's interest, it not only considers the user's long-term interest as a consideration, but also considers the user's short-term interest and real-time interest as factors to reflect the user's interest changes over time. In this case, compared with the current technology, the real-time information that the user is most interested in can be recommended to the user more flexibly, accurately, and in a timely manner, and the recommendation effect is greatly improved while improving the timeliness.
  • FIG. 1 is a schematic diagram of a scenario of a recommendation system for real-time information according to an embodiment of the present invention
  • FIG. 1b is a flowchart of a method for recommending real-time information according to an embodiment of the present invention
  • FIG. 2 is another flowchart of a method for recommending real-time information according to an embodiment of the present invention
  • FIG. 3 is a structural diagram of a device for recommending real-time information according to an embodiment of the present invention
  • FIG. 3b is another structural diagram of a device for recommending real-time information according to an embodiment of the present invention.
  • Embodiments of the present invention provide a method, an apparatus, and a system for recommending real-time information.
  • the recommendation system of the real-time information includes any recommendation device for real-time information provided by the embodiment of the present invention, and the recommendation device of the real-time information may be specifically integrated into a server, such as a recommendation server.
  • the recommendation system of the real-time information may further include other devices, such as user equipment, and a user server for storing user behavior data, an information server for storing original real-time information, and the like.
  • the recommendation device of the real-time information is integrated in the recommendation server, and when the real-time information needs to be recommended, the recommendation server may acquire user behavior data from the user server, and calculate the short-term interest of the user according to the user behavior data. Long-term interest and real-time interest, then determine user interest based on the user's short-term interests, long-term interests, and real-time interests, obtain real-time information from the information server, and recommend real-time information, such as news, to the user device based on the user's interest.
  • the short-term interest refers to the interest weight calculated by the user in a short period, and the interest value of the user in each of the preset periods (such as 30 days) can be calculated according to the obtained user behavior data.
  • Obtaining the interest value of the day interest, and obtaining the interest weight value of the day is attenuated according to time;
  • the long-term interest refers to the interest weight calculated by the user in a longer period, for example, the user may be calculated according to the user behavior data.
  • the interest weight of one year, and so on; and real-time interest refers to the user's current interest weight, such as a keyword or label that the user currently clicks, and so on.
  • This embodiment will be described from the perspective of a recommendation device for real-time information, and the recommendation device of the real-time information may be specifically integrated in a server, such as a recommendation server or the like.
  • a method for recommending real-time information includes: acquiring user behavior data; calculating short-term interests, long-term interests, and real-time interests of the user according to the user behavior data; determining user interests according to the short-term interests, long-term interests, and real-time interests of the users; This user interest recommends real-time information to the user.
  • the specific process of the recommended method for real-time information can be as follows:
  • user behavior data refers to relevant data that can be analyzed by user behavior, such as user browsing records, click records, and/or download records. These user behavior data may be stored in the recommendation device of the real-time information, or may be stored in other devices, such as a user server.
  • the user behavior is stored in the user server.
  • the user behavior data can be obtained from the user server.
  • the details can be as follows:
  • the interest weight that needs to be attenuated is attenuated according to time, and the weighted interest value is obtained.
  • the date difference between the date of the interest weight that needs to be attenuated and the current date calculate the product of the date difference and the preset attenuation coefficient, calculate the difference between 1 and the product, and then attenuate the need.
  • the interest weight is multiplied by the difference to obtain the attenuated interest weight.
  • the attenuation coefficient can be set according to the requirements of the actual application, and details are not described herein again.
  • the interest weights for each day are attenuated according to time, and the weighted interest value for each day is obtained.
  • the preset period can be set according to the requirements of the actual application.
  • the preset period can be generally set to 7 days, 15 days, or 30 days.
  • the preset time range may be set according to the requirements of the actual application, at least one day, for example, may be set to one quarter, one year or two years, and the like.
  • the user behavior of the user in each month of the current date may be counted according to the user behavior data; the weight of each interest in the current month is calculated according to the user behavior of each month; The interest weights of the current month are calculated as the average weight of each interest in a year; the average weight is counted to obtain the user's long-term interest.
  • the weight of each interest in the current month may be attenuated according to time, the weight of interest after attenuation is obtained, and then the weighted interest weights are counted to obtain long-term interest of the user, and the like.
  • the keyword determines that the user's current interest is "NBA”, so the weight of the "NBA” keyword can be calculated, and so on, the user's real-time interest can be obtained.
  • the user's short-term interests, long-term interests, and real-time interests can be combined according to a preset strategy to obtain user interest.
  • the real-time information may specifically be information such as news.
  • the corresponding real-time information may be recalled from the inverted index of the real-time information to obtain candidate recommendation information, and the real-time information is recommended to the user based on the candidate recommendation information, which may be as follows:
  • the newer real-time information such as the newer news, the higher the newness.
  • C Determine a click rate of each real-time information in the candidate recommendation information, and calculate a click model factor (CM) according to the click rate.
  • CM click model factor
  • the real-time information with larger click-through rate has a larger model factor.
  • the real-time information in the candidate recommendation information may be scored according to the interest correlation, the newness, and the click model factor, and then the real-time information with the score higher than the preset threshold is determined as the recommendation information.
  • the preset threshold can be set according to the requirements of the actual application.
  • the information quality of each real-time information in the candidate recommendation information may be evaluated.
  • a quality factor of the news may be determined by means of text recognition, where The quality of junk articles and advertising articles is low. That is, before the step of determining the recommendation information from the candidate recommendation information according to the interest correlation, the newness, and the click model factor, the recommendation method of the real-time information may further include:
  • the step “determining the recommendation information from the candidate recommendation information according to the interest correlation, the newness, and the click model factor” may include: according to the interest correlation, the newness, the click model factor, and the information quality.
  • the recommendation information is determined in the candidate recommendation information, for example, as follows:
  • Real-time information in the candidate recommendation information is scored according to the interest correlation, the newness, the click model factor, and the information quality; and the real-time information whose score is higher than the preset threshold is determined as the recommendation information.
  • the inverted index of the real-time information can be obtained by collecting and counting the original real-time information, that is, in the step “recalling the corresponding real-time information from the inverted index of the real-time information according to the user interest,
  • the recommended method of the real-time information may further include:
  • the step "recalling the corresponding real-time information from the inverted index of the real-time information according to the user interest to obtain the candidate recommendation information" may specifically be: determining the category, topic, and/or key of the user's interest according to the user's interest.
  • the word obtains the original real-time information that is the same, similar, or similar to the category, topic, and/or keyword of the user's interest from the inverted index of the real-time information, and obtains the candidate recommendation information.
  • synonyms and/or synonyms may be set for the words involved in the categories, topics, and/or keywords of interest to the user, if such synonyms and/or keywords are included in the categories, topics, and/or keywords of the original real-time information. / or a word with the same synonym, then the original real-time information is determined to be original real-time information similar or similar to the category, subject, and/or keyword of interest to the user, and so on.
  • the embodiment uses the user behavior data to calculate the short-term interest, long-term interest, and real-time interest of the user according to the user behavior data, and then determines the user interest according to the short-term interest, long-term interest, and real-time interest of the user, and based on the User interest recommends real-time information to users; because the program calculates the user's interest, it not only considers the user's long-term interest as a consideration, but also considers the user's short-term interest and real-time interest as factors to reflect the user's interest. As time goes by, therefore, compared with the current technology, the real-time information that the user is most interested in can be recommended to the user more flexibly, accurately and in a timely manner, and the recommendation effect is greatly improved while improving the timeliness.
  • the recommendation device of the real-time information is specifically integrated into the recommendation server, and the real-time information is specifically described as an example of the news.
  • a method for recommending real-time information the specific process can be as follows:
  • the recommendation server obtains user behavior data from the user server.
  • user behavior data refers to relevant data that can be analyzed by user behavior, such as user browsing records, click records, and/or download records.
  • the recommendation server calculates, according to the user behavior data, the interest weight of the user every day for 30 days, obtains the interest weight of the day, and attenuates the interest weight of the day according to time, and obtains the short-term interest of the user.
  • the details can be as follows:
  • the interest weight that needs to be attenuated is attenuated according to time, and the weighted interest value is obtained.
  • ⁇ is the attenuation coefficient
  • D is the date difference between the date of the interest weight that needs to be attenuated and the current date. Indicates the interest weight that needs to be attenuated, which is the weight of interest after attenuation.
  • the attenuation coefficient can be set according to the requirements of the actual application, and details are not described herein again.
  • the interest weights for each day are attenuated according to time, and the weighted interest value for each day is obtained.
  • the recommendation server collects the interest weight of the user within one year of the current date according to the user behavior data, and obtains the long-term interest of the user.
  • the user behavior may be counted according to the user behavior data for each month of the current date, and the weight of each interest in the current month is calculated according to the user behavior of each month, according to the interest of each month in the current month.
  • the weight is calculated as the average weight of each interest in a year, and the average weight is counted to obtain the user's long-term interest.
  • the weight of each interest in the current month is attenuated according to time, and the weight of interest after attenuation is obtained, and then these decayed interest weights are counted to obtain long-term interest of the user, and so on.
  • the recommendation server determines, according to the user behavior data, an interest weight value currently clicked by the user, and obtains real-time interest of the user.
  • the message currently clicked by the user includes the “NBA” keyword (or label)
  • the “NBA” keyword or label
  • the “NBA” can be calculated.
  • the weight of the keyword, and so on, can get the user's real-time interest.
  • steps 202, 203, and 204 may be performed in no particular order.
  • the recommendation server determines the user interest according to the short-term interest, long-term interest, and real-time interest of the user.
  • the user's short-term interests, long-term interests, and real-time interests can be combined according to a preset strategy to obtain user interest.
  • the recommendation server obtains an inverted index of the news.
  • the details can be as follows:
  • the original news from the original real-time information database, extracting the extracted original news, and classifying and predicting the original news according to the extracted features to determine the category and theme of the original news; After the content of the original news is subjected to part-of-speech weighting, text field weighting is performed to determine the keyword to which the original news belongs; the inverted index of the original news in the original news database is calculated according to the category, theme and keyword of the original news, and the news is obtained. Inverted index.
  • the original real-time information base may be stored in the recommendation server, or may be stored in other devices, such as an information server.
  • liblinear a technique for classifier generation
  • liblinear a technique for classifier generation
  • the theme model (LDA, Latent Dirichlet) can be used. Allocation) to predict the subject of the original news (ie, topic prediction), that is, LDA can be used to identify the subject information hidden in the original news document, and get the theme of the original news.
  • LDA Latent Dirichlet
  • TF-IDF word frequency-reverse file frequency
  • Document Frequency Document Frequency
  • step 206 and steps 201-205 may be performed in no particular order.
  • the recommendation server recalls the corresponding news from the inverted index of the news according to the user interest, and obtains the candidate recommendation information.
  • the category, topic, and/or keyword that the user is interested in may be determined according to the user's interest, and the inverted index of the news is obtained from the same, similar, or similar category, topic, and/or keyword that the user is interested in.
  • the original news get the candidate recommendation information.
  • the recommendation server recommends news to the user based on the candidate recommendation information.
  • the details can be as follows:
  • C Determine a click rate of each news in the candidate recommendation information, and calculate a click model factor (CM) according to the click rate.
  • CM click model factor
  • the quality factor of a news can be determined by means of text recognition, wherein the quality of the junk articles and the advertisement articles is low.
  • Determining the recommended news from the candidate recommendation information according to the interest correlation, the newness, the click model factor, and the information quality for example, the following may be:
  • the news in the candidate recommendation information is scored according to the interest correlation, the newness, the click model factor, and the information quality; and the news with the score higher than the preset threshold is determined as the recommended news.
  • steps A, B, C, and D may be in no particular order.
  • the embodiment uses the user behavior data to calculate the short-term interest, long-term interest, and real-time interest of the user according to the user behavior data, and then determines the user interest according to the short-term interest, long-term interest, and real-time interest of the user, and based on the User interest recommends news to users; because the program calculates the user's interest, it not only considers the user's long-term interest as a consideration, but also considers the user's short-term interest and real-time interest as factors to reflect the user's interest.
  • the time changes therefore, compared with the current technology, the user's current most interesting news can be recommended to the user more flexibly, accurately and in a timely manner, and the recommendation effect is greatly improved while improving the timeliness.
  • the embodiment of the present invention further provides a recommendation device for real-time information.
  • the recommendation device of the real-time information includes an obtaining unit 301, an operation unit 302, a determining unit 303, and a recommending unit 304. ,as follows:
  • the obtaining unit 301 is configured to acquire user behavior data.
  • user behavior data refers to relevant data that can be analyzed by user behavior, such as user browsing records, click records, and/or download records.
  • user behavior is stored in the user server.
  • the user behavior data can be obtained from the user server.
  • the operation unit 302 is configured to separately calculate short-term interests, long-term interests, and real-time interests of the user according to the user behavior data.
  • the operation unit 302 may include a first calculation subunit, a second calculation subunit, and a third calculation subunit, as follows:
  • a first calculating sub-unit configured to calculate, according to the user behavior data, an interest weight of each day of the user in the preset period, obtain a daily interest weight, and attenuate the interest weight of the day according to time, thereby obtaining a short-term interest of the user.
  • the first calculating sub-unit may be specifically configured to determine, according to the obtained Tianyi interest weight, an interest weight that needs to be attenuated, and the interest weight that needs to be attenuated is attenuated according to time, and the reduced interest right is obtained. Returning the operation of determining the interest weight that needs to be attenuated according to the interest value of the day, until all the interest weights in the interest weights that need to be attenuated are attenuated; and all the obtained attenuation interest values are counted. , get the user's short-term interest.
  • the first calculating sub-unit may be specifically configured to determine a date difference between a date on which the interest weight that needs to be attenuated and the current date is attenuated; and calculating the date difference and The product of the preset attenuation coefficient is calculated, and the difference between the product and the product is calculated.
  • the interest weight that needs to be attenuated is multiplied by the difference to obtain the weighted interest value. For details, refer to the previous method embodiment. Let me repeat.
  • the preset period can be set according to the requirements of the actual application.
  • the preset period can be generally set to 7 days, 15 days, or 30 days.
  • a second calculating sub-unit configured to calculate, according to the user behavior data, an interest weight of the user within a preset time range, to obtain a long-term interest of the user.
  • the preset time range may be set according to the requirements of the actual application, at least one day, for example, may be set to one quarter, one year or two years, and the like.
  • the second calculating sub-unit may be specifically configured to calculate, according to the user behavior data, the user behavior of the user in each month of the current date; and calculate the user behavior according to the monthly user behavior.
  • the weight of interest in the current month; the average weight of each interest in the year is calculated according to the weight of each interest in the current month, and the average weight is counted to obtain the long-term interest of the user.
  • the second calculating subunit may also attenuate the weight of each interest in the current month according to time, obtain the attenuated interest weight, and then perform statistics on the attenuated interest weights to obtain long-term interests of the user, etc. .
  • the third calculating subunit is configured to determine, according to the user behavior data, an interest weight value currently clicked by the user, to obtain real-time interest of the user.
  • the determining unit 303 is configured to determine the user interest according to the short-term interest, long-term interest, and real-time interest of the user.
  • the determining unit 303 may be specifically configured to fuse the short-term interest, the long-term interest, and the real-time interest of the user according to the preset policy to obtain the user interest.
  • the recommendation unit 304 is configured to recommend real-time information to the user based on the user interest.
  • the real-time information may specifically be information such as news.
  • the recommendation unit 304 can include a recall subunit and a recommendation subunit, as follows:
  • the recalling sub-unit is configured to recall corresponding real-time information from the inverted index of the real-time information according to the user interest, to obtain candidate recommendation information.
  • the recommendation subunit is configured to recommend real-time information to the user based on the candidate recommendation information.
  • the details can be as follows:
  • the recommendation sub-unit may be specifically configured to calculate a matching degree between each real-time information in the candidate recommendation information and the user interest, and obtain an interest correlation of the real-time information; and determine a release time of each real-time information in the candidate recommendation information to determine each real-time information. Freshness, obtaining the newness of real-time information; determining the click rate of each real-time information in the candidate recommendation information, and calculating a click model factor according to the click rate; according to the interest correlation, the newness and the click model factor
  • the recommendation information is determined in the candidate recommendation information; the recommendation information is recommended to the user.
  • the information quality of each real-time information in the candidate recommendation information may be evaluated.
  • a quality factor of the news may be determined by means of text recognition, where The quality of junk articles and advertising articles is low. which is:
  • the recommendation sub-unit may be specifically configured to calculate a matching degree between each real-time information in the candidate recommendation information and the user interest, and obtain an interest correlation of the real-time information; and determine a release time of each real-time information in the candidate recommendation information to determine each real-time information.
  • the inverted index of the real-time information can be obtained by collecting and counting the original real-time information, that is, as shown in FIG. 3b, the recommendation device of the real-time information may further include an inverted index determining unit 305, as follows. :
  • the inverted index determining unit 305 can be configured to obtain original real-time information from the original real-time information base, perform feature extraction on the acquired original real-time information, and perform classification prediction and topic prediction on the original real-time information according to the extracted features. Determining the original real-time information category and subject; after performing the part-of-speech weighting on the content of the obtained original real-time information, performing text field weighting to determine the keyword to which the original real-time information belongs; according to the original real-time information category, subject, and keyword Calculating an inverted index of the original real-time information in the original real-time information database, and obtaining an inverted index of the real-time information;
  • the sub-unit is recalled, specifically for determining the category, topic, and/or keyword that the user is interested in according to the user's interest, and obtaining the category, topic, and/or interest of the user from the inverted index of the real-time information. Or the original real-time information with the same keyword, similarity or similarity, and the candidate recommendation information is obtained.
  • the foregoing units may be implemented as a separate entity, or may be implemented in any combination, and may be implemented as the same or a plurality of entities.
  • the foregoing method embodiments and details are not described herein.
  • the recommendation device of the real-time information may be specifically integrated in a server, such as a recommendation server or the like.
  • the acquiring unit 301 of the real-time information recommendation apparatus of the present embodiment can acquire user behavior data, and then the computing unit 302 calculates the short-term interest, long-term interest, and real-time interest of the user according to the user behavior data, and then, by The determining unit 303 determines the user interest according to the short-term interest, long-term interest, and real-time interest of the user, and the recommendation unit 304 recommends real-time information to the user based on the user interest; since the solution calculates the user's interest, not only the long-term interest of the user is taken as Considering the factors, and also taking the user's short-term interest and real-time interest as factors to reflect the user's interest changes over time, therefore, compared to the current technology, the user can be more flexible, accurate and timely The most interesting real-time information is currently recommended to the user, and the recommendation effect is greatly improved while improving the timeliness.
  • the recommendation unit 304 can consider not only the interest correlation of the real-time information, but also the timeliness of the real-time information, the click rate and the information quality of the keyword, and the like. A more accurate description of the relationship between this information and user interest, further improving the quality of recommendations and recommendations.
  • the embodiment of the present invention further provides a recommendation system for real-time information, which may include any recommendation device for real-time information provided by the embodiment of the present invention.
  • a recommendation system for real-time information may include any recommendation device for real-time information provided by the embodiment of the present invention.
  • the recommendation device of the real-time information may be integrated into a server, such as a recommendation server, to be integrated into the recommendation server, and the specific information may be as follows:
  • a recommendation server configured to obtain user behavior data; calculate short-term interests, long-term interests, and real-time interests of the user according to the user behavior data; determine user interest according to the short-term interest, long-term interest, and real-time interest of the user; and Recommend real-time information.
  • the operation of calculating the user's short-term interest, long-term interest and real-time interest can be as follows:
  • the user behavior data calculates the interest weight of the user in each day of the preset period, obtains the interest weight of the day, and attenuates the interest weight of the day according to time, and obtains the short-term interest of the user.
  • the preset period can be set according to the requirements of the actual application.
  • the preset period can be generally set to 7 days, 15 days, or 30 days.
  • the preset time range may be set according to the requirements of the actual application, at least one day, for example, may be set to one quarter, one year or two years, and the like.
  • the interest weight value currently clicked by the user is determined according to the user behavior data, and the real-time interest of the user is obtained.
  • the recommendation server may retrieve the corresponding real-time information from the inverted index of the real-time information according to the user interest, obtain candidate recommendation information, and then recommend real-time information to the user based on the candidate recommendation information. For example, parameters such as interest correlation, newness, click model factor, and information quality of real-time information can be calculated, and then determined from the candidate recommendation information according to the interest correlation, the newness, the click model factor, and the information quality. Recommended information.
  • the inverted index of the real-time information can be obtained by collecting and counting the original real-time information, for example, as follows:
  • the recommendation server can also be used to obtain the original real-time information from the original real-time information base; perform feature extraction on the obtained original real-time information; perform classification prediction and topic prediction on the original real-time information according to the extracted features to determine the original real-time Information category and subject; after performing the part-of-speech weighting on the content of the original real-time information obtained, performing text field weighting to determine the keyword to which the original real-time information belongs; calculating the original according to the original real-time information category, subject and keyword The inverted index of the original real-time information in the real-time information database, and the inverted index of the real-time information is obtained.
  • the recommendation system of the real-time information may further include other devices, such as user equipment, and optionally, a user server and an information server, as follows:
  • the user equipment can be used to receive real-time information recommended by the recommendation server.
  • a user server that can be used to store user behavior data and provide user behavior data to the recommendation server.
  • the information server can be used to save the original real-time information and provide the saved original real-time information to the recommendation server.
  • the recommendation system of the real-time information may include any one of the recommended devices for real-time information provided by the embodiments of the present invention. Therefore, the beneficial effects of any real-time information recommendation device provided by the embodiments of the present invention may be implemented. See the previous embodiment, and details are not described herein.
  • the program may be stored in a computer readable storage medium, and the storage medium may include: Read only memory (ROM, Read Only Memory), Random Access Memory (RAM), disk or CD.
  • ROM Read only memory
  • RAM Random Access Memory

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种实时信息的推荐方法，包括：获取用户行为数据，根据该用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣，然后根据用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣，并基于该用户兴趣向用户推荐实时信息；此外，还提供相应的装置和系统。

Description

一种实时信息的推荐方法、装置和系统
本申请要求于2015年9月8日提交中国专利局、申请号为2015105644813、发明名称为“一种实时信息的推荐方法、装置和系统”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及通信技术领域，具体涉及一种实时信息的推荐方法、装置和系统。
 背景技术
互联网信息的急速增长，使得人们淹没在资讯的海洋中，如何在浩瀚的信息海洋中快速地找到需要的信息，是一个极为重要的问题。
为了解决互联网海量信息资源出现的“信息过载”问题，协助用户从玲琅满目的海量信息中快速获取到自己所需的信息，现有技术提出了各种信息的推荐方案，所谓信息推荐，指的是根据用户的兴趣特点和行为，向用户推荐用户感兴趣的信息，为此，现有的推荐算法主要可分为两大类，一是基于行为的推荐算法、二是基于内容的推荐算法。其中，基于行为的推荐算法主要是通过统计用户对信息的行为，以及计算推荐池中各信息的相似度，然后将与该行为所对应的信息相似度较高的信息推荐给用户。而基于内容的推荐算法则主要是通过对各类信息进行打关键字，以及对用户的兴趣进行挖掘，以确定用户所感兴趣的关键字，然后基于该感兴趣的关键字和各类信息的关键字计算推荐列表，并推荐给用户。
 技术问题
在对现有技术的研究和实践过程中，本发明的发明人发现，现有的推荐方案或是需要依赖大量的用户参与，或时没有考虑用户的兴趣变化，因此，时效性较差，而对于实时信息，如新闻类的信息，均具有一次性消费的特点（即对于相同内容的新闻，用户只会阅读一次），时效性极为重要，因此，对于实时信息而言，现有的推荐方案的推荐效果并不佳。
 技术解决方案
本发明实施例提供一种实时信息的推荐方法、装置和系统，可以提高其时效性，灵活、准确且及时地将用户当前最感兴趣的实时信息推荐给用户，大大改善推荐效果。
本发明实施例提供一种实时信息的推荐方法，包括：
获取用户行为数据；
根据所述用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣；
根据所述用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣；
基于所述用户兴趣向用户推荐实时信息。
相应的，本发明实施例还提供一种实时信息的推荐装置，包括：
获取单元，用于获取用户行为数据；
运算单元，用于根据所述用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣；
确定单元，用于根据所述用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣；
推荐单元，用于基于所述用户兴趣向用户推荐实时信息。
此外，本发明实施例还提供一种实时信息的推荐系统，包括本发明实施例提供的任一种实时信息的推荐装置。
另外，本发明实施例还提供一种存储介质，其内存储有处理器可执行指令，所述处理器可执行指令用于执行如下操作：
获取用户行为数据；
根据所述用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣；
根据所述用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣；
基于所述用户兴趣向用户推荐实时信息。
 有益效果
本发明实施例采用获取用户行为数据，根据该用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣，然后根据用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣，并基于该用户兴趣向用户推荐实时信息；由于该方案在计算用户的兴趣时，不仅将用户的长期兴趣作为考量因素，而且，也将用户的短期兴趣与实时兴趣也作为考量因素，以反映出用户的兴趣随时间变化的情况，因此，相对于现在技术而言，可以更加灵活、准确且及时地将用户当前最感兴趣的实时信息推荐给用户，在提高时效性的同时，大大改善推荐效果。
 附图说明
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1a是本发明实施例提供的实时信息的推荐系统的场景示意图；
图1b是本发明实施例提供的实时信息的推荐方法的流程图；
图2是本发明实施例提供的实时信息的推荐方法的另一流程图；
图3a是本发明实施例提供的实时信息的推荐装置的结构图；
图3b是本发明实施例提供的实时信息的推荐装置的另一结构图。
 本发明的最佳实施方式
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清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
本发明实施例提供一种实时信息的推荐方法、装置和系统。
参见图1a，该实时信息的推荐系统包括本发明实施例所提供的任一种实时信息的推荐装置，该实时信息的推荐装置具体可以集成在服务器，比如推荐服务器中。此外，该实时信息的推荐系统还可以包括其他的设备，比如用户设备，以及用于保存用户行为数据的用户服务器，以及用于保存原始实时信息的信息服务器等。
例如，以该实时信息的推荐装置集成在推荐服务器中为例，则当需要对实时信息进行推荐时，该推荐服务器可以从用户服务器获取用户行为数据，根据该用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣，然后根据该用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣，从信息服务器获取实时信息，并基于该用户兴趣向用户设备推荐实时信息，比如新闻等。
其中，短期兴趣指的是用户在较短的一个周期内所计算得到的兴趣权值，具体可以根据获取到的用户行为数据计算用户在预置周期（比如30天）中每一天的兴趣权值，得到天兴趣权值，并对该天兴趣权值按照时间进行衰减来得到；长期兴趣指的是用户在较长一个周期内所计算得到的兴趣权重，比如可以根据该用户行为数据计算用户在一年内的兴趣权值，等等；而实时兴趣则指的是用户当前兴趣权值，比如用户当前点击的了某个关键字或标签，等等。
以下将分别进行详细说明。
实施例一、
本实施例将从实时信息的推荐装置的角度进行描述，该实时信息的推荐装置具体可以集成在服务器，比如推荐服务器等设备中。
一种实时信息的推荐方法，包括：获取用户行为数据；根据该用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣；根据该用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣；基于该用户兴趣向用户推荐实时信息。
如图1b所示，该实时信息的推荐方法的具体流程可以如下：
101、获取用户行为数据。
其中，用户行为数据指的是可供用户行为分析的相关数据，比如用户的浏览记录、点击记录和/或下载记录等数据。这些用户行为数据可以存储在该实时信息的推荐装置中，也可以存储在其他的设备，比如用户服务器等设备中。
以存储在用户服务器为例，则此时，具体可以从用户服务器中获取用户行为数据。
102、根据该用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣。例如，具体可以如下：
（1）根据该用户行为数据计算用户在预置周期中每一天的兴趣权值，得到天兴趣权值，并对该天兴趣权值按照时间进行衰减，得到用户的短期兴趣。
其中，按时间进行衰减的方式可以有多种，例如，具体可以如下：
A、根据该天兴趣权值确定当前需要进行衰减的兴趣权值。
B、对该需要进行衰减的兴趣权值按照时间进行衰减，得到衰减后的兴趣权值。
比如，可以确定该需要进行衰减的兴趣权值所在的日期与当前日期的日期差，计算该日期差与预置衰减系数的乘积，并计算1与该乘积的差，然后将该需要进行衰减的兴趣权值乘以该差，得到衰减后的兴趣权值。
[援引加入(细则20.6) 23.05.2016]　
比如，若用μ来表示衰减系数，用D表示需要进行衰减的兴趣权值所在的日期与当前日期的日期差，final_weightk表示需要进行衰减的兴趣权值，则衰减后兴趣权值 final_weightk用公式表示为：
[援引加入(细则20.6) 23.05.2016]　
final_weightk=interest_weightk*(1-D*μ)。
其中，衰减系数可以根据实际应用的需求进行设置，在此不再赘述。
C、返回执行根据该天兴趣权值确定当前需要进行衰减的兴趣权值的步骤，直至该天兴趣权值中所有需要进行衰减的兴趣权重衰减完毕。
即对每一天的兴趣权值均按照时间进行衰减处理，得到对每一天应的衰减后的兴趣权值。
D、对得到的所有衰减后的兴趣权值进行统计，得到用户的短期兴趣。
其中，预置周期可以根据实际应用的需求进行设置，比如，一般可以设置为7天、15天或30天等。
[援引加入(细则20.6) 23.05.2016]　
例如，以30天为例，则若当前日期为8月10号，则可以以8月10为基准，获取30天，即7月12号至8月10号这段日期中每一天的兴趣权值，然后，对该每一天的兴趣权值按照时间进行衰减，比如，以8月8日为例，由于8月8日与8月10号的日期差为两天，则8月8日这一天的衰减后兴趣权值final_weight28为：
[援引加入(细则20.6) 23.05.2016]　
final_weight28=interest_weight28*(1-2*μ)。
当然，也可以采用其他的衰减方式，在此不再赘述。
（2）根据该用户行为数据计算用户在预置时间范围内的兴趣权值，得到用户的长期兴趣。
其中，该预置时间范围可以根据实际应用的需求进行设置，至少大于一天，比如，可以设置为一个季度，一年或两年，等等。
例如，以一年为例，则可以根据该用户行为数据统计用户在当前日期的一年内每个月的用户行为；根据该每个月的用户行为计算每个兴趣在当月的权重；根据该每个兴趣在当月的权重计算一年内每个兴趣的平均权重；对该平均权重进行统计，得到用户的长期兴趣。
可选的，也可以对每个兴趣在当月的权重按照时间进行衰减，得到衰减后的兴趣权重，然后对这些衰减后的兴趣权重进行统计，得到用户的长期兴趣，等等。
（3）根据该用户行为数据确定用户当前点击的兴趣权值，得到用户的实时兴趣。
例如，若用户当前点击的消息包括了“NBA” 关键字（或标签），则确定用户当前的兴趣为“NBA”，于是，可以计算该“NBA” 关键字的权值，以此类推，便可以得到用户的实时兴趣。
103、根据该用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣。
例如，可以根据预置的策略将用户的短期兴趣、长期兴趣和实时兴趣进行融合，得到用户兴趣。
其中，融合的方式可以有多种，比如，可以分别为短期兴趣、长期兴趣和实时兴趣设置权重，然后根据这些权重将短期兴趣、长期兴趣和实时兴趣融合在一起，又比如，还可以为短期兴趣、长期兴趣和实时兴趣之间的关系设置一个函数，然后通过该函数将短期兴趣、长期兴趣和实时兴趣融合在一起，等等，在此不再赘述。
104、基于该用户兴趣向用户推荐实时信息。其中，该实时信息具体可以为新闻等信息。
例如，可以根据该用户兴趣，从实时信息的倒排索引中召回相应的实时信息，得到候选推荐信息，基于该候选推荐信息推荐实时信息给该用户，具体可以如下：
A、计算该候选推荐信息中各实时信息与该用户兴趣的匹配度，得到实时信息的兴趣相关性。
B、确定该候选推荐信息中各实时信息的发布时间确定各实时信息的新鲜度，得到实时信息的时新性。
其中，越新的实时信息，比如越新的新闻，其时新性越高。
C、确定该候选推荐信息中各实时信息的点击率，并根据该点击率计算点击模型因子（CM）。
其中，点击率越大的实时信息，其模型因子越大。
D、根据该兴趣相关性、时新性和点击模型因子从该候选推荐信息中确定推荐信息。
比如，可以根据该兴趣相关性、时新性和点击模型因子对该候选推荐信息中的实时信息进行评分，然后将评分高于预置阈值的实时信息确定为推荐信息。
其中，该预置阈值可以根据实际应用的需求进行设置。
E、将该推荐信息推荐给该用户。
可选的，为了提高推荐的信息质量，在确定推荐消息之前，还可以对候选推荐信息中各实时信息的信息质量进行评判，比如，可以通过文本识别的方式确定一个新闻的质量因子，其中，垃圾文章、广告文章的质量分低。即在步骤“根据该兴趣相关性、时新性和点击模型因子从该候选推荐信息中确定推荐信息”之前，该实时信息的推荐方法还可以包括：
确实该候选推荐信息中各实时信息的信息质量。
则此时，步骤“根据该兴趣相关性、时新性和点击模型因子从该候选推荐信息中确定推荐信息”可以包括：根据上该兴趣相关性、时新性、点击模型因子和信息质量从该候选推荐信息中确定推荐信息，比如，具可以如下：
根据该兴趣相关性、时新性、点击模型因子和信息质量对该候选推荐信息中的实时信息进行评分；将评分高于预置阈值的实时信息确定为推荐信息。
需说明的是，其中，实时信息的倒排索引可以通过对原始实时信息进行收集和统计来得到，即在步骤“根据该用户兴趣，从实时信息的倒排索引中召回相应的实时信息，得到候选推荐信息”之前，该实时信息的推荐方法还可以包括：
从原始实时信息库中获取原始实时信息；对获取到的原始实时信息进行特征提取；根据提取到的特征对该原始实时信息进行分类预测和话题预测，以确定原始实时信息类目和主题；对获取到的原始实时信息的内容进行词性加权处理后，进行文本域加权，以确定原始实时信息所属的关键字；根据原始实时信息类目、主题和关键字计算该原始实时信息库中原始实时信息的倒排索引，得到实时信息的倒排索引。
则此时，步骤“根据该用户兴趣，从实时信息的倒排索引中召回相应的实时信息，得到候选推荐信息”具体可以为：根据用户兴趣确定用户所感兴趣的类目、主题和/或关键字，从该实时信息的倒排索引中获取与用户所感兴趣的类目、主题和/或关键字相同、相似或相近的原始实时信息，得到候选推荐信息。
例如，可以为用户所感兴趣的类目、主题和/或关键字中所涉及的词语设置同义词和/或近义词，若原始实时信息的类目、主题和/或关键字中包括有与这些同义词和/或近义词相同的词语，则确定该原始实时信息为与用户所感兴趣的类目、主题和/或关键字相似或相近的原始实时信息，等等。
由上可知，本实施例采用获取用户行为数据，根据该用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣，然后根据用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣，并基于该用户兴趣向用户推荐实时信息；由于该方案在计算用户的兴趣时，不仅将用户的长期兴趣作为考量因素，而且，也将用户的短期兴趣与实时兴趣也作为考量因素，以反映出用户的兴趣随时间变化的情况，因此，相对于现在技术而言，可以更加灵活、准确且及时地将用户当前最感兴趣的实时信息推荐给用户，在提高时效性的同时，大大改善推荐效果。
此外，在计算候选推荐信息的过程中，不仅可以考虑实时信息的兴趣相关性，同时也可以考虑实时信息的时新性，以及关键字的点击率和信息质量等，因此，可以更加准确的描述这些信息与用户兴趣之间关系强弱，进一步提高了推荐质量和推荐效果。
实施例二、
根据实施例一所描述的方法，以下将以该实时信息的推荐装置具体集成在推荐服务器，且该实时信息具体为新闻为例进行详细说明。
如图2所示，一种实时信息的推荐方法，具体流程可以如下：
201、推荐服务器从用户服务器中获取用户行为数据。
其中，用户行为数据指的是可供用户行为分析的相关数据，比如用户的浏览记录、点击记录和/或下载记录等数据。
202、推荐服务器根据该用户行为数据计算用户在30天内每一天的兴趣权值，得到天兴趣权值，并对该天兴趣权值按照时间进行衰减，得到用户的短期兴趣。例如，具体可以如下：
A、根据该天兴趣权值确定当前需要进行衰减的兴趣权值。
B、对该需要进行衰减的兴趣权值按照时间进行衰减，得到衰减后的兴趣权值。
比如，可以确定该需要进行衰减的兴趣权值所在的日期与当前日期的日期差，计算该日期差与预置衰减系数的乘积，并计算1与该乘积的差，然后将该需要进行衰减的兴趣权值乘以该差，得到衰减后的兴趣权值，用公式表示即为：
[援引加入(细则20.6) 23.05.2016]　
final_weightK=interest_weightK*(1-D*μ)。
其中，μ为衰减系数，D为需要进行衰减的兴趣权值所在的日期与当前日期的日期差， 表示需要进行衰减的兴趣权值， 为衰减后兴趣权值。其中，衰减系数可以根据实际应用的需求进行设置，在此不再赘述。
C、返回执行根据该天兴趣权值确定当前需要进行衰减的兴趣权值的步骤，直至该天兴趣权值中所有需要进行衰减的兴趣权重衰减完毕。
即对每一天的兴趣权值均按照时间进行衰减处理，得到对每一天应的衰减后的兴趣权值。
D、对得到的所有衰减后的兴趣权值进行统计，得到用户的短期兴趣。
[援引加入(细则20.6) 23.05.2016]　
例如，若当前日期为8月10号，则可以以8月10为基准，获取30天，即7月12号至8月10号这段日期中每一天的兴趣权值，然后，对该每一天的兴趣权值按照时间进行衰减，比如，以8月8日为例，由于8月8日与8月10号的日期差为两天，则8月8日这一天的衰减后兴趣权值final_weight28为：
[援引加入(细则20.6) 23.05.2016]　
final_weight28=interest_weight28*(1-2*μ)。
当然，也可以采用其他的衰减方式，在此不再赘述。
203、推荐服务器根据该用户行为数据统计用户在当前日期的一年内的兴趣权值，得到用户的长期兴趣。
例如，具体可以根据该用户行为数据统计用户在当前日期的一年内每个月的用户行为，并根据该每个月的用户行为计算每个兴趣在当月的权重，根据该每个兴趣在当月的权重计算一年内每个兴趣的平均权重，对该平均权重进行统计，得到用户的长期兴趣。
每个兴趣在当月的权重按照时间进行衰减，得到衰减后的兴趣权重，然后对这些衰减后的兴趣权重进行统计，得到用户的长期兴趣，等等。
204、推荐服务器根据该用户行为数据确定用户当前点击的兴趣权值，得到用户的实时兴趣。
例如，若用户当前点击的消息包括了“NBA”关键字（或标签），则确定用户当前的兴趣为“NBA”，于是，可以计算该“NBA” 关键字的权值，以此类推，便可以得到用户的实时兴趣。
其中，步骤202、203和204的执行可以不分先后。
205、推荐服务器根据该用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣。
例如，可以根据预置的策略将用户的短期兴趣、长期兴趣和实时兴趣进行融合，得到用户兴趣。
其中，融合的方式可以有多种，比如，可以分别为短期兴趣、长期兴趣和实时兴趣设置权重，然后根据这些权重将短期兴趣、长期兴趣和实时兴趣融合在一起，又比如，还可以为短期兴趣、长期兴趣和实时兴趣之间的关系设置一个函数，然后通过该函数将短期兴趣、长期兴趣和实时兴趣融合在一起，等等，在此不再赘述。
206、推荐服务器获取新闻的倒排索引。例如，具体可以如下：
从原始实时信息库中获取原始新闻，对获取到的原始新闻进行特征提取，根据提取到的特征对该原始新闻进行分类预测和话题预测，以确定原始新闻的类目和主题；对获取到的原始新闻的内容进行词性加权处理后，进行文本域加权，以确定原始新闻所属的关键字；根据原始新闻的类目、主题和关键字计算该原始新闻库中原始新闻的倒排索引，得到新闻的倒排索引。
其中，该原始实时信息库可以存储在该推荐服务器中，也可以存储在其他的设备，比如信息服务器中。
其中，可以采用liblinear（一种用来进行分类器生成的技术）来对原始新闻的分类进行预测（即分类预测），得到原始新闻的类目
其中，可以采用主题模型（LDA，Latent Dirichlet Allocation）来对原始新闻的主题进行预测（即话题预测），即可以采用LDA来识别原始新闻的文档中潜藏的主题信息，得到原始新闻的主题。
其中，可以采用词频-逆向文件频率（TF-IDF，term frequency–inverse document frequency）来对获取到的原始新闻的内容进行词性加权处理，其中，TF-IDF是一种用于信息检索与数据挖掘的常用加权技术，它也是一种统计方法，用以评估一字词对于一个文件集或一个语料库中的其中一份文件的重要程度。
其中，步骤206与步骤201~205的执行可以不分先后。
207、推荐服务器根据该用户兴趣，从新闻的倒排索引中召回相应的新闻，得到候选推荐信息。
例如，具体可以根据用户兴趣确定用户所感兴趣的类目、主题和/或关键字，从该新闻的倒排索引中获取与用户所感兴趣的类目、主题和/或关键字相同、相似或相近的原始新闻，得到候选推荐信息。
208、推荐服务器基于该候选推荐信息推荐新闻给该用户。例如，具体可以如下：
A、计算该候选推荐信息中各新闻与该用户兴趣的匹配度，得到新闻与用户的兴趣相关性。
B、确定该候选推荐信息中各新闻的发布时间确定各新闻的新鲜度，得到新闻的时新性。其中，越新的新闻，其时新性越高。
C、确定该候选推荐信息中各新闻的点击率，并根据该点击率计算点击模型因子（CM）。其中，点击率越大的新闻，其模型因子越大。
D、确实该候选推荐信息中各新闻的信息质量。
比如，可以通过文本识别的方式确定一个新闻的质量因子，其中，垃圾文章、广告文章的质量分低。
E、根据上该兴趣相关性、时新性、点击模型因子和信息质量从该候选推荐信息中确定推荐新闻，比如，具可以如下：
根据该兴趣相关性、时新性、点击模型因子和信息质量对该候选推荐信息中的新闻进行评分；将评分高于预置阈值的新闻确定为推荐新闻。
F、将该推荐新闻推荐给该用户。
其中，步骤A、B、C和D的执行步骤可以不分先后。
由上可知，本实施例采用获取用户行为数据，根据该用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣，然后根据用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣，并基于该用户兴趣向用户推荐新闻；由于该方案在计算用户的兴趣时，不仅将用户的长期兴趣作为考量因素，而且，也将用户的短期兴趣与实时兴趣也作为考量因素，以反映出用户的兴趣随时间变化的情况，因此，相对于现在技术而言，可以更加灵活、准确且及时地将用户当前最感兴趣的新闻推荐给用户，在提高时效性的同时，大大改善推荐效果。
此外，在计算新闻的过程中，不仅可以考虑新闻与用户的兴趣相关性，同时也可以考虑新闻的时新性，以及关键字的点击率和新闻质量等，因此，可以更加准确的描述这些信息与用户兴趣之间关系强弱，进一步提高了推荐质量和推荐效果。
实施例三、
为了更好地实施以上方法，本发明实施例还提供一种实时信息的推荐装置，如图3a所示，该实时信息的推荐装置包括获取单元301、运算单元302、确定单元303和推荐单元304，如下：
（1）获取单元301；
获取单元301，用于获取用户行为数据。
其中，用户行为数据指的是可供用户行为分析的相关数据，比如用户的浏览记录、点击记录和/或下载记录等数据。以存储在用户服务器为例，则此时，具体可以从用户服务器中获取用户行为数据。
（2）运算单元302；
运算单元302，用于根据该用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣。
例如，该运算单元302可以包括第一计算子单元、第二计算子单元和第三计算子单元，如下：
第一计算子单元，用于根据该用户行为数据计算用户在预置周期中每一天的兴趣权值，得到天兴趣权值，并对该天兴趣权值按照时间进行衰减，得到用户的短期兴趣。
例如，该第一计算子单元，具体可以用于根据得到的天兴趣权值确定当前需要进行衰减的兴趣权值，对该需要进行衰减的兴趣权值按照时间进行衰减，得到衰减后的兴趣权值；返回执行根据天兴趣权值确定当前需要进行衰减的兴趣权值的操作，直至该天兴趣权值中所有需要进行衰减的兴趣权重衰减完毕；对得到的所有衰减后的兴趣权值进行统计，得到用户的短期兴趣。
其中，在对需要进行衰减的兴趣权值进行衰减时，该第一计算子单元，具体可以用于确定该需要进行衰减的兴趣权值所在的日期与当前日期的日期差；计算该日期差与预置衰减系数的乘积，并计算1与该乘积的差；将该需要进行衰减的兴趣权值乘以该差，得到衰减后的兴趣权值，具体可参见前面的方法实施例，在此不再赘述。
其中，预置周期可以根据实际应用的需求进行设置，比如，一般可以设置为7天、15天或30天等。
第二计算子单元，用于根据该用户行为数据计算用户在预置时间范围内的兴趣权值，得到用户的长期兴趣。
其中，该预置时间范围可以根据实际应用的需求进行设置，至少大于一天，比如，可以设置为一个季度，一年或两年，等等。
例如，以一年为例，则该第二计算子单元，具体可以用于根据该用户行为数据统计用户在当前日期的一年内每个月的用户行为；根据该每个月的用户行为计算每个兴趣在当月的权重；根据该每个兴趣在当月的权重计算一年内每个兴趣的平均权重，对该平均权重进行统计，得到用户的长期兴趣。
可选的，第二计算子单元也可以对每个兴趣在当月的权重按照时间进行衰减，得到衰减后的兴趣权重，然后对这些衰减后的兴趣权重进行统计，得到用户的长期兴趣，等等。
第三计算子单元，用于根据该用户行为数据确定用户当前点击的兴趣权值，得到用户的实时兴趣。
（3）确定单元303；
确定单元303，用于根据该用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣。
例如，确定单元303，具体可以用于根据预置的策略将用户的短期兴趣、长期兴趣和实时兴趣进行融合，得到用户兴趣。
其中，融合的方式可以有多种，比如，可以分别为短期兴趣、长期兴趣和实时兴趣设置权重，然后根据这些权重将短期兴趣、长期兴趣和实时兴趣融合在一起，又比如，还可以为短期兴趣、长期兴趣和实时兴趣之间的关系设置一个函数，然后通过该函数将短期兴趣、长期兴趣和实时兴趣融合在一起，等等，在此不再赘述。
（4）推荐单元304；
推荐单元304，用于基于该用户兴趣向用户推荐实时信息。
其中，该实时信息具体可以为新闻等信息。
例如，该推荐单元304可以包括召回子单元和推荐子单元，如下：
该召回子单元，用于根据该用户兴趣，从实时信息的倒排索引中召回相应的实时信息，得到候选推荐信息。
该推荐子单元，用于基于该候选推荐信息推荐实时信息给该用户。具体可以如下：
该推荐子单元，具体可以用于计算该候选推荐信息中各实时信息与该用户兴趣的匹配度，得到实时信息的兴趣相关性；确定该候选推荐信息中各实时信息的发布时间确定各实时信息的新鲜度，得到实时信息的时新性；确定该候选推荐信息中各实时信息的点击率，并根据该点击率计算点击模型因子；根据该兴趣相关性、时新性和点击模型因子从该候选推荐信息中确定推荐信息；将该推荐信息推荐给该用户。
可选的，为了提高推荐的信息质量，在确定推荐消息之前，还可以对候选推荐信息中各实时信息的信息质量进行评判，比如，可以通过文本识别的方式确定一个新闻的质量因子，其中，垃圾文章、广告文章的质量分低。即：
该推荐子单元，具体可以用于计算该候选推荐信息中各实时信息与该用户兴趣的匹配度，得到实时信息的兴趣相关性；确定该候选推荐信息中各实时信息的发布时间确定各实时信息的新鲜度，得到实时信息的时新性；确定该候选推荐信息中各实时信息的点击率，并根据该点击率计算点击模型因子；确实该候选推荐信息中各实时信息的信息质量；根据该兴趣相关性、时新性、点击模型因子和信息质量从该候选推荐信息中确定推荐信息；将该推荐信息推荐给该用户。
需说明的是，其中，实时信息的倒排索引可以通过对原始实时信息进行收集和统计来得到，即如图3b所示，该实时信息的推荐装置还可以包括倒排索引确定单元305，如下：
倒排索引确定单元305，可以用于从原始实时信息库中获取原始实时信息；对获取到的原始实时信息进行特征提取；根据提取到的特征对该原始实时信息进行分类预测和话题预测，以确定原始实时信息类目和主题；对获取到的原始实时信息的内容进行词性加权处理后，进行文本域加权，以确定原始实时信息所属的关键字；根据原始实时信息类目、主题和关键字计算该原始实时信息库中原始实时信息的倒排索引，得到实时信息的倒排索引；
则此时，召回子单元，具体用于根据用户兴趣确定用户所感兴趣的类目、主题和/或关键字，从该实时信息的倒排索引中获取与用户所感兴趣的类目、主题和/或关键字相同、相似或相近的原始实时信息，得到候选推荐信息。
具体实施时，以上各个单元可以作为独立的实体来实现，也可以进行任意组合，作为同一或若干个实体来实现，以上各个单元的具体实施可参见前面的方法实施例，在此不再赘述。
该实时信息的推荐装置具体可以集成在服务器，比如推荐服务器等设备中。
由上可知，本实施例的实时信息的推荐装置的获取单元301可以获取用户行为数据，然后由运算单元302根据该用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣，再然后，由确定单元303根据用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣，并由推荐单元304基于该用户兴趣向用户推荐实时信息；由于该方案在计算用户的兴趣时，不仅将用户的长期兴趣作为考量因素，而且，也将用户的短期兴趣与实时兴趣也作为考量因素，以反映出用户的兴趣随时间变化的情况，因此，相对于现在技术而言，可以更加灵活、准确且及时地将用户当前最感兴趣的实时信息推荐给用户，在提高时效性的同时，大大改善推荐效果。
此外，在推荐单元304在计算候选推荐信息的过程中，不仅可以考虑实时信息的兴趣相关性，同时也可以考虑实时信息的时新性，以及关键字的点击率和信息质量等，因此，可以更加准确的描述这些信息与用户兴趣之间关系强弱，进一步提高了推荐质量和推荐效果。
实施例四、
此外，本发明实施例还提供一种实时信息的推荐系统，可以包括本发明实施例提供的任一种实时信息的推荐装置，具体可参见实施例三。其中，该实时信息的推荐装置具体可以集成在服务器，如推荐服务器中，以集成在推荐服务器中，则具体可以如下：
推荐服务器，用于获取用户行为数据；根据该用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣；根据该用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣；基于该用户兴趣向用户推荐实时信息。
其中，计算用户的短期兴趣、长期兴趣和实时兴趣的操作具体可以如下：
（1）短期兴趣；
用户行为数据计算用户在预置周期中每一天的兴趣权值，得到天兴趣权值，并对该天兴趣权值按照时间进行衰减，得到用户的短期兴趣。
其中，预置周期可以根据实际应用的需求进行设置，比如，一般可以设置为7天、15天或30天等。
（2）长期兴趣；
根据该用户行为数据计算用户在预置时间范围内的兴趣权值，得到用户的长期兴趣。
其中，该预置时间范围可以根据实际应用的需求进行设置，至少大于一天，比如，可以设置为一个季度，一年或两年，等等。
（3）实时兴趣；
根据该用户行为数据确定用户当前点击的兴趣权值，得到用户的实时兴趣。
其中，推荐服务器在向用户推荐实时信息时，具体可以根据该用户兴趣，从实时信息的倒排索引中召回相应的实时信息，得到候选推荐信息，然后基于该候选推荐信息推荐实时信息给该用户，比如，可以计算实时信息的兴趣相关性、时新性、点击模型因子和信息质量等参数，然后根据上该兴趣相关性、时新性、点击模型因子和信息质量从该候选推荐信息中确定推荐信息。
需说明的是，其中，实时信息的倒排索引可以通过对原始实时信息进行收集和统计来得到，例如，可以如下：
推荐服务器，还可以用于从原始实时信息库中获取原始实时信息；对获取到的原始实时信息进行特征提取；根据提取到的特征对该原始实时信息进行分类预测和话题预测，以确定原始实时信息类目和主题；对获取到的原始实时信息的内容进行词性加权处理后，进行文本域加权，以确定原始实时信息所属的关键字；根据原始实时信息类目、主题和关键字计算该原始实时信息库中原始实时信息的倒排索引，得到实时信息的倒排索引。
以上各个操作的具体实施可参见前面的方法实施例，在此不再赘述。
此外，该实时信息的推荐系统还可以包括其他的设备，比如用户设备，可选的，还可以包括用户服务器和信息服务器等，如下：
用户设备，可以用于接收推荐服务器推荐的实时信息。
用户服务器，可以用于保存用户行为数据，并向推荐服务器提供用户行为数据。
信息服务器，可以用于保存原始实时信息，并向推荐服务器提供保存的原始实时信息。
以上各个设备的具体实施，可参见前面的方法实施例，在此不再赘述。
由于该实时信息的推荐系统可以包括本发明实施例提供的任一种实时信息的推荐装置，因此，可以实现本发明实施例提供的任一种实时信息的推荐装置所能实现的有益效果，详见前面的实施例，在此不再赘述。
本领域普通技术人员可以理解上述实施例的各种方法中的全部或部分步骤是可以通过程序来指令相关的硬件来完成，该程序可以存储于一计算机可读存储介质中，存储介质可以包括：只读存储器（ROM，Read Only Memory）、随机存取记忆体（RAM，Random Access Memory）、磁盘或光盘等。
以上对本发明实施例所提供的一种实时信息的推荐方法、装置和系统进行了详细介绍，本文中应用了具体个例对本发明的原理及实施方式进行了阐述，以上实施例的说明只是用于帮助理解本发明的方法及其核心思想；同时，对于本领域的技术人员，依据本发明的思想，在具体实施方式及应用范围上均会有改变之处，综上所述，本说明书内容不应理解为对本发明的限制。

Claims (22)

  1. 一种实时信息的推荐方法，其特征在于，包括：
    获取用户行为数据；
    根据所述用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣；
    根据所述用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣；
    基于所述用户兴趣向用户推荐实时信息。
  2. 根据权利要求1所述的方法，其特征在于，所述根据所述用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣，包括：
    根据所述用户行为数据计算用户在预置周期中每一天的兴趣权值，得到天兴趣权值，并对所述天兴趣权值按照时间进行衰减，得到用户的短期兴趣；
    根据所述用户行为数据计算用户在预置时间范围内的兴趣权值，得到用户的长期兴趣，所述预置时间范围大于一天；
    根据所述用户行为数据确定用户当前点击的兴趣权值，得到用户的实时兴趣。
  3. 根据权利要求2所述的方法，其特征在于，所述对所述天兴趣权值按照时间进行衰减，得到用户的短期兴趣，包括：
    根据所述天兴趣权值确定当前需要进行衰减的兴趣权值；
    对所述需要进行衰减的兴趣权值按照时间进行衰减，得到衰减后的兴趣权值；
    返回执行根据所述天兴趣权值确定当前需要进行衰减的兴趣权值的步骤，直至所述天兴趣权值中所有需要进行衰减的兴趣权重衰减完毕；
    对得到的所有衰减后的兴趣权值进行统计，得到用户的短期兴趣。
  4. 根据权利要求3所述的方法，其特征在于，所述对所述需要进行衰减的兴趣权值按照时间进行衰减，得到衰减后的兴趣权值，包括：
    确定所述需要进行衰减的兴趣权值所在的日期与当前日期的日期差；
    计算所述日期差与预置衰减系数的乘积，并计算1与所述乘积的差；
    将所述需要进行衰减的兴趣权值乘以所述差，得到衰减后的兴趣权值。
  5. 根据权利要求2所述的方法，其特征在于，所述根据所述用户行为数据计算用户在预置时间范围内的兴趣权值，得到用户的长期兴趣，包括：
    根据所述用户行为数据统计用户在当前日期的一年内，每个月的用户行为；
    根据所述每个月的用户行为计算每个兴趣在当月的权重；
    根据所述每个兴趣在当月的权重计算一年内每个兴趣的平均权重；
    对所述平均权重进行统计，得到用户的长期兴趣。
  6. 根据权利要求1至5任一项所述的方法，所述基于所述用户兴趣向用户推荐实时信息，包括：
    根据所述用户兴趣，从实时信息的倒排索引中召回相应的实时信息，得到候选推荐信息；
    基于所述候选推荐信息推荐实时信息给所述用户。
  7. 根据权利要求6所述的方法，其特征在于，所述将基于所述候选推荐信息推荐实时信息给所述用户，包括：
    计算所述候选推荐信息中各实时信息与所述用户兴趣的匹配度，得到实时信息的兴趣相关性；
    确定所述候选推荐信息中各实时信息的发布时间确定各实时信息的新鲜度，得到实时信息的时新性；
    确定所述候选推荐信息中各实时信息的点击率，并根据所述点击率计算点击模型因子；
    根据所述兴趣相关性、时新性和点击模型因子从所述候选推荐信息中确定推荐信息；
    将所述推荐信息推荐给所述用户。
  8. 根据权利要求7所述的方法，其特征在于，所述根据所述兴趣相关性、时新性和点击模型因子从所述候选推荐信息中确定推荐信息，包括：
    根据所述兴趣相关性、时新性和点击模型因子对所述候选推荐信息中的实时信息进行评分；
    将评分高于预置阈值的实时信息确定为推荐信息。
  9. 根据权利要求6所述的方法，其特征在于，所述根据所述兴趣相关性、时新性和点击模型因子从所述候选推荐信息中确定推荐信息之前，还包括：
    确实所述候选推荐信息中各实时信息的信息质量；
    所述根据所述兴趣相关性、时新性和点击模型因子从所述候选推荐信息中确定推荐信息，包括：根据上所述兴趣相关性、时新性、点击模型因子和信息质量从所述候选推荐信息中确定推荐信息。
  10. 根据权利要求9所述的方法，其特征在于，所述根据上所述兴趣相关性、时新性、点击模型因子和信息质量从所述候选推荐信息中确定推荐信息，包括：
    根据所述兴趣相关性、时新性、点击模型因子和信息质量对所述候选推荐信息中的实时信息进行评分；
    将评分高于预置阈值的实时信息确定为推荐信息。
  11. 根据权利要求6所述的方法，其特征在于，所述根据所述用户兴趣，从实时信息的倒排索引中召回相应的实时信息，得到候选推荐信息之前，还包括：
    从原始实时信息库中获取原始实时信息；
    对获取到的原始实时信息进行特征提取；
    根据提取到的特征对所述原始实时信息进行分类预测和话题预测，以确定原始实时信息类目和主题；
    对获取到的原始实时信息的内容进行词性加权处理后，进行文本域加权，以确定原始实时信息所属的关键字；
    根据原始实时信息类目、主题和关键字计算所述原始实时信息库中原始实时信息的倒排索引，得到实时信息的倒排索引；
    所述根据所述用户兴趣，从实时信息的倒排索引中召回相应的实时信息，得到候选推荐信息，具体为：根据用户兴趣确定用户所感兴趣的类目、主题和/或关键字，从所述实时信息的倒排索引中获取与用户所感兴趣的类目、主题和/或关键字相同、相似或相近的原始实时信息，得到候选推荐信息。
  12. 一种实时信息的推荐装置，其特征在于，包括：
    获取单元，用于获取用户行为数据；
    运算单元，用于根据所述用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣；
    确定单元，用于根据所述用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣；
    推荐单元，用于基于所述用户兴趣向用户推荐实时信息。
  13. 根据权利要求12所述的装置，其特征在于，所述运算单元包括第一计算子单元、第二计算子单元和第三计算子单元；
    第一计算子单元，用于根据所述用户行为数据计算用户在预置周期中每一天的兴趣权值，得到天兴趣权值，并对所述天兴趣权值按照时间进行衰减，得到用户的短期兴趣；
    第二计算子单元，用于根据所述用户行为数据计算用户在预置时间范围内的兴趣权值，得到用户的长期兴趣，所述预置时间范围大于一天；
    第三计算子单元，用于根据所述用户行为数据确定用户当前点击的兴趣权值，得到用户的实时兴趣。
  14. 根据权利要求13所述的装置，其特征在于，
    所述第一计算子单元，具体用于根据所述天兴趣权值确定当前需要进行衰减的兴趣权值，对所述需要进行衰减的兴趣权值按照时间进行衰减，得到衰减后的兴趣权值；返回执行根据所述天兴趣权值确定当前需要进行衰减的兴趣权值的操作，直至所述天兴趣权值中所有需要进行衰减的兴趣权重衰减完毕；对得到的所有衰减后的兴趣权值进行统计，得到用户的短期兴趣。
  15. 根据权利要求14所述的装置，其特征在于，
    所述第一计算子单元，具体用于确定所述需要进行衰减的兴趣权值所在的日期与当前日期的日期差；计算所述日期差与预置衰减系数的乘积，并计算1与所述乘积的差；将所述需要进行衰减的兴趣权值乘以所述差，得到衰减后的兴趣权值。
  16. 根据权利要求13所述的装置，其特征在于，
    第二计算子单元，具体用于根据所述用户行为数据统计用户在当前日期的一年内，每个月的用户行为；根据所述每个月的用户行为计算每个兴趣在当月的权重；根据所述每个兴趣在当月的权重计算一年内每个兴趣的平均权重；对所述平均权重进行统计，得到用户的长期兴趣。
  17. 根据权利要求12至16任一项所述的装置，其特征在于，所述推荐单元包括召回子单元和推荐子单元；
    所述召回子单元，用于根据所述用户兴趣，从实时信息的倒排索引中召回相应的实时信息，得到候选推荐信息；
    所述推荐子单元，用于基于所述候选推荐信息推荐实时信息给所述用户。
  18. 根据权利要求17所述的装置，其特征在于，
    所述推荐子单元，具体用于计算所述候选推荐信息中各实时信息与所述用户兴趣的匹配度，得到实时信息的兴趣相关性；确定所述候选推荐信息中各实时信息的发布时间确定各实时信息的新鲜度，得到实时信息的时新性；确定所述候选推荐信息中各实时信息的点击率，并根据所述点击率计算点击模型因子；根据所述兴趣相关性、时新性和点击模型因子从所述候选推荐信息中确定推荐信息；将所述推荐信息推荐给所述用户。
  19. 根据权利要求17所述的装置，其特征在于，
    所述推荐子单元，具体用于计算所述候选推荐信息中各实时信息与所述用户兴趣的匹配度，得到实时信息的兴趣相关性；确定所述候选推荐信息中各实时信息的发布时间确定各实时信息的新鲜度，得到实时信息的时新性；确定所述候选推荐信息中各实时信息的点击率，并根据所述点击率计算点击模型因子；确实所述候选推荐信息中各实时信息的信息质量；根据所述兴趣相关性、时新性、点击模型因子和信息质量从所述候选推荐信息中确定推荐信息；将所述推荐信息推荐给所述用户。
  20. 根据17所述的装置，其特征在于，还包括倒排索引确定单元；
    倒排索引确定单元，用于从原始实时信息库中获取原始实时信息；对获取到的原始实时信息进行特征提取；根据提取到的特征对所述原始实时信息进行分类预测和话题预测，以确定原始实时信息类目和主题；对获取到的原始实时信息的内容进行词性加权处理后，进行文本域加权，以确定原始实时信息所属的关键字；根据原始实时信息类目、主题和关键字计算所述原始实时信息库中原始实时信息的倒排索引，得到实时信息的倒排索引；
    所述召回子单元，具体用于根据用户兴趣确定用户所感兴趣的类目、主题和/或关键字，从所述实时信息的倒排索引中获取与用户所感兴趣的类目、主题和/或关键字相同、相似或相近的原始实时信息，得到候选推荐信息。
  21. 一种实时信息的推荐系统，其特征在于，包括权利要求12至20所述的任一种实时信息的推荐装置。
  22. 一种存储介质，其特征在于，其内存储有处理器可执行指令，所述处理器可执行指令用于执行如下操作：
    获取用户行为数据；
    根据所述用户行为数据分别计算用户的短期兴趣、长期兴趣和实时兴趣；
    根据所述用户的短期兴趣、长期兴趣和实时兴趣确定用户兴趣；
    基于所述用户兴趣向用户推荐实时信息。
PCT/CN2016/078482 2015-09-08 2016-04-05 一种实时信息的推荐方法、装置和系统 WO2017041484A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US15/868,729 US11003726B2 (en) 2015-09-08 2018-01-11 Method, apparatus, and system for recommending real-time information

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510564481.3 2015-09-08
CN201510564481.3A CN106503014B (zh) 2015-09-08 2015-09-08 一种实时信息的推荐方法、装置和系统

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US15/868,729 Continuation US11003726B2 (en) 2015-09-08 2018-01-11 Method, apparatus, and system for recommending real-time information

Publications (1)

Publication Number Publication Date
WO2017041484A1 true WO2017041484A1 (zh) 2017-03-16

Family

ID=58239117

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/078482 WO2017041484A1 (zh) 2015-09-08 2016-04-05 一种实时信息的推荐方法、装置和系统

Country Status (3)

Country Link
US (1) US11003726B2 (zh)
CN (1) CN106503014B (zh)
WO (1) WO2017041484A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110765310A (zh) * 2019-10-09 2020-02-07 上海麦克风文化传媒有限公司 一种基于参数配置的音频专辑推荐方法及系统

Families Citing this family (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108810056B (zh) * 2017-05-04 2021-05-14 腾讯科技（北京）有限公司 信息推送方法及装置
CN107545065B (zh) * 2017-09-08 2021-08-13 北京奇艺世纪科技有限公司 一种用户兴趣校准方法、系统及个性化推荐系统
CN110020117B (zh) * 2017-09-29 2022-05-03 北京搜狗科技发展有限公司 一种兴趣信息获取方法、装置及电子设备
CN107844525A (zh) * 2017-10-12 2018-03-27 广州艾媒数聚信息咨询股份有限公司 一种基于用户行为的资讯个性化推荐方法、系统及装置
CN107844598A (zh) * 2017-11-22 2018-03-27 广州优视网络科技有限公司 内容推荐方法、装置及计算机设备
CN110431535A (zh) * 2018-01-22 2019-11-08 华为技术有限公司 一种用户画像的生成方法及装置
CN108537567B (zh) * 2018-03-06 2020-08-07 阿里巴巴集团控股有限公司 一种目标用户群体的确定方法和装置
CN108805622B (zh) * 2018-06-11 2021-11-09 深圳乐信软件技术有限公司 商品推荐方法、装置、设备及存储介质
WO2020000207A1 (zh) * 2018-06-26 2020-01-02 深圳市爱的网络科技有限公司 用户兴趣采集方法、装置、计算机装置及计算机可读存储介质
CN109189951B (zh) * 2018-07-03 2021-09-03 南京尚网网络科技有限公司 一种多媒体资源推荐方法、设备及存储介质
CN109460514A (zh) * 2018-11-02 2019-03-12 北京京东尚科信息技术有限公司 用于推送信息的方法和装置
CN111241381A (zh) * 2018-11-28 2020-06-05 北京奇虎科技有限公司 信息推荐方法、装置、电子设备及计算机可读存储介质
CN109993570B (zh) * 2019-01-14 2023-09-01 深圳市东信时代信息技术有限公司 一种定向投放移动广告的方法及系统
CN110275952A (zh) * 2019-05-08 2019-09-24 平安科技（深圳）有限公司 基于用户短期兴趣的新闻推荐方法、装置及介质
CN110489639B (zh) * 2019-07-15 2022-10-25 北京奇艺世纪科技有限公司 一种内容推荐方法及装置
CN110516159B (zh) * 2019-08-30 2022-12-20 北京字节跳动网络技术有限公司 一种信息推荐方法、装置、电子设备及存储介质
CN110765309B (zh) * 2019-10-09 2023-09-01 上海麦克风文化传媒有限公司 一种基于参数配置的推荐系统召回方法及系统
US11263174B2 (en) 2019-11-08 2022-03-01 International Business Machines Corporation Reducing resource consumption in container image management
CN111061945B (zh) * 2019-11-11 2023-06-27 汉海信息技术(上海)有限公司 推荐方法、装置、电子设备，存储介质
CN111209474B (zh) * 2019-12-27 2023-04-28 广东德诚科教有限公司 在线课程的推荐方法、装置、计算机设备和存储介质
CN111310033B (zh) * 2020-01-23 2023-05-30 山西大学 基于用户兴趣漂移的推荐方法及推荐装置
CN111444419A (zh) * 2020-03-02 2020-07-24 平安国际智慧城市科技股份有限公司 资源推荐方法、装置、计算机设备和存储介质
CN111369324B (zh) * 2020-03-12 2024-01-23 苏州大学 一种目标信息确定方法、装置、设备及可读存储介质
CN111444428B (zh) * 2020-03-27 2022-08-30 腾讯科技（深圳）有限公司 基于人工智能的信息推荐方法、装置、电子设备及存储介质
CN112084404B (zh) * 2020-09-01 2024-03-01 北京百度网讯科技有限公司 内容推荐方法、装置、设备和介质
CN113744016B (zh) * 2020-11-04 2024-05-24 北京沃东天骏信息技术有限公司 一种对象推荐方法及装置、设备、存储介质
JPWO2022201435A1 (zh) * 2021-03-25 2022-09-29
CN113486250B (zh) * 2021-07-28 2023-09-05 中移（杭州）信息技术有限公司 内容推荐方法、装置、设备及计算机可读存储介质
CN115048586B (zh) * 2022-08-11 2023-02-21 广东工业大学 一种融合多特征的新闻推荐方法及系统
CN116049535A (zh) * 2022-08-18 2023-05-02 荣耀终端有限公司 信息推荐方法、装置、终端装置及存储介质
CN117828193B (zh) * 2024-03-04 2024-05-17 山东省计算中心（国家超级计算济南中心） 基于多兴趣半联合学习兴趣推荐方法、系统、设备及介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102663627A (zh) * 2012-04-26 2012-09-12 焦点科技股份有限公司 个性化推荐方法
CN102681999A (zh) * 2011-03-08 2012-09-19 阿里巴巴集团控股有限公司 一种用户行为信息收集及信息发送方法及装置
CN104090990A (zh) * 2014-07-31 2014-10-08 北京奇虎科技有限公司 新闻推送方法和系统
CN104246751A (zh) * 2011-12-02 2014-12-24 Kddi株式会社 推荐装置、推荐系统、推荐方法以及程序
US20150088911A1 (en) * 2013-09-25 2015-03-26 Alibaba Group Holding Limited Method and system for extracting user behavior features to personalize recommendations

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7853485B2 (en) * 2005-11-22 2010-12-14 Nec Laboratories America, Inc. Methods and systems for utilizing content, dynamic patterns, and/or relational information for data analysis
US8001105B2 (en) * 2006-06-09 2011-08-16 Ebay Inc. System and method for keyword extraction and contextual advertisement generation
US9405830B2 (en) * 2007-02-28 2016-08-02 Aol Inc. Personalization techniques using image clouds
JP4433326B2 (ja) * 2007-12-04 2010-03-17 ソニー株式会社 情報処理装置および方法、並びにプログラム
US8996530B2 (en) * 2012-04-27 2015-03-31 Yahoo! Inc. User modeling for personalized generalized content recommendations
US9646062B2 (en) * 2013-06-10 2017-05-09 Microsoft Technology Licensing, Llc News results through query expansion
US9514133B1 (en) * 2013-06-25 2016-12-06 Jpmorgan Chase Bank, N.A. System and method for customized sentiment signal generation through machine learning based streaming text analytics
CN104572734B (zh) * 2013-10-23 2019-04-30 腾讯科技（深圳）有限公司 问题推荐方法、装置及系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102681999A (zh) * 2011-03-08 2012-09-19 阿里巴巴集团控股有限公司 一种用户行为信息收集及信息发送方法及装置
CN104246751A (zh) * 2011-12-02 2014-12-24 Kddi株式会社 推荐装置、推荐系统、推荐方法以及程序
CN102663627A (zh) * 2012-04-26 2012-09-12 焦点科技股份有限公司 个性化推荐方法
US20150088911A1 (en) * 2013-09-25 2015-03-26 Alibaba Group Holding Limited Method and system for extracting user behavior features to personalize recommendations
CN104090990A (zh) * 2014-07-31 2014-10-08 北京奇虎科技有限公司 新闻推送方法和系统

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110765310A (zh) * 2019-10-09 2020-02-07 上海麦克风文化传媒有限公司 一种基于参数配置的音频专辑推荐方法及系统
CN110765310B (zh) * 2019-10-09 2024-01-30 上海麦克风文化传媒有限公司 一种基于参数配置的音频专辑推荐方法及系统

Also Published As

Publication number Publication date
US20180129749A1 (en) 2018-05-10
US11003726B2 (en) 2021-05-11
CN106503014B (zh) 2020-08-07
CN106503014A (zh) 2017-03-15

Similar Documents

Publication Publication Date Title
WO2017041484A1 (zh) 一种实时信息的推荐方法、装置和系统
WO2012134180A2 (ko) 문장에 내재한 감정 분석을 위한 감정 분류 방법 및 컨텍스트 정보를 이용한 다중 문장으로부터의 감정 분류 방법
WO2013117147A1 (zh) 微博排序、搜索、展示方法和系统
WO2020009297A1 (ko) 도메인 추출기반의 언어 이해 성능 향상장치및 성능 향상방법
WO2013139239A1 (en) Method for recommending users in social network and the system thereof
WO2015129989A1 (ko) 콘텐츠 및 음원 추천 장치 및 방법
WO2012091360A2 (ko) 유저 맞춤형 컨텐츠 제공 방법 및 시스템
WO2015144089A1 (en) Application recommending method and apparatus
WO2017054463A1 (zh) 事件信息推送方法、事件信息推送装置及存储介质
WO2018034426A1 (ko) 커널 ｒｄｒ을 이용한 태깅 말뭉치 오류 자동수정방법
WO2016167424A1 (ko) 답변 추천 장치, 자동 문장 완성 시스템 및 방법
WO2010024628A2 (ko) 확장 키워드 풀을 사용한 검색 방법 및 시스템
WO2012130145A1 (zh) 获取和搜索相关知识信息的方法及装置
WO2019177182A1 (ko) 속성 정보 분석을 통한 멀티미디어 컨텐츠 검색장치 및 검색방법
WO2016003219A1 (en) Electronic device and method for providing content on electronic device
WO2015131803A1 (en) Application recommending method and system
WO2019075944A1 (zh) 基于用户行为的信息屏蔽方法、装置、终端和存储介质
WO2018129978A1 (zh) 信息处理方法、装置、存储介质及计算机设备
EP3494498A1 (en) Electronic device and method for controlling the same
WO2018076811A1 (zh) 数据分享方法、装置、存储介质及电子设备
WO2020022819A1 (en) Communication via simulated user
WO2013097603A1 (zh) 评价信息生成方法及系统、计算机存储介质
WO2016093630A1 (en) Semantic enrichment of trajectory data
WO2024029922A1 (ko) 소프트웨어 기반 기업 esg 경영 관리 장치 및 그 수행 방법
WO2020153525A1 (ko) 인공지능 개체에 대한 정보를 블록체인으로 관리하는 방법 및 그 플랫폼

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: 16843423

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: 16843423

Country of ref document: EP

Kind code of ref document: A1