WO2020140400A1 - 基于用户行为的产品推荐方法、装置、设备及存储介质 - Google Patents

基于用户行为的产品推荐方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2020140400A1
WO2020140400A1 PCT/CN2019/092103 CN2019092103W WO2020140400A1 WO 2020140400 A1 WO2020140400 A1 WO 2020140400A1 CN 2019092103 W CN2019092103 W CN 2019092103W WO 2020140400 A1 WO2020140400 A1 WO 2020140400A1
Authority
WO
WIPO (PCT)
Prior art keywords
product
recommended
attitude
user
behavior
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/092103
Other languages
English (en)
French (fr)
Inventor
石志娟
黄燕霞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
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 Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2020140400A1 publication Critical patent/WO2020140400A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions

Definitions

  • This application relates to the field of big data analysis, and in particular, to a product recommendation method, device, device, and storage medium based on user behavior.
  • the traditional product recommendation method is to place a large number of product advertisements on Internet websites, or recommend the same current main product on the homepage recommendation position of some product platforms, that is, recommend the same product to different users.
  • the effect of this recommendation method is very unsatisfactory; at the same time, useless products Pushing will also consume a lot of energy and time from the user, and may even arouse the user's resentment, thereby affecting the promotion of the product.
  • the main purpose of the present application is to provide a product recommendation method, device, equipment and storage medium based on user behavior, aiming to implement targeted product recommendation and improve the recommendation effect.
  • the present application provides a product recommendation method based on user behavior.
  • the product recommendation method based on user behavior includes:
  • the present application also provides a device for recommending products based on user behavior.
  • the device for recommending products based on user behavior includes:
  • a data acquisition module used to acquire historical behavior data of the user to be recommended, and determine the corresponding historical product set according to the historical behavior data
  • An interest determination module configured to analyze the historical behavior data based on a preset hidden Markov model, obtain an attitude probability of the potential attitude of the user to be recommended to each historical product in the historical product set, and according to the attitude Probability to determine the real interest products of the user to be recommended in the historical product set;
  • the information pushing module is used to obtain recommended product information of recommended products having an association relationship with the real interest product, and push the recommended product information to a user terminal corresponding to the user to be recommended based on a preset recommendation rule.
  • the present application also provides a product recommendation device based on user behavior
  • the product recommendation device based on user behavior includes a processor, a memory, and is stored on the memory and can be used by the processor Executed computer-readable instructions, where the computer-readable instructions are executed by the processor to implement the steps of the product recommendation method based on user behavior as described above.
  • the present application also provides a storage medium that stores computer readable instructions, where the computer readable instructions are executed by a processor to implement a user behavior-based product as described above Recommended method steps.
  • This application analyzes the real interest of the user based on the historical behavior data of the user, and then obtains and pushes the recommended products associated with the product based on the real interest, so that the product recommendation results meet the actual needs of the user, thereby improving the recommendation effect; at the same time, analyzing the user's real
  • the interest time is based on the hidden Markov model of the user's behavior, in order to estimate the user's attitude probability when performing various historical behaviors, and then estimate the user's product attitude, and then determine the real interest based on the product attitude, which can be to a certain extent
  • non-realistic interest behavior data noise
  • FIG. 1 is a schematic diagram of a hardware structure of a product recommendation device based on user behavior involved in an embodiment of the present application
  • FIG. 2 is a schematic flowchart of a first embodiment of a product recommendation method based on user behavior in this application.
  • the product recommendation method based on user behavior involved in the embodiments of the present application is mainly applied to a product recommendation device based on user behavior.
  • the product recommendation device may be implemented by a device with a data processing function such as a personal computer (PC), a server, or the like.
  • FIG. 1 is a schematic diagram of a hardware structure of a product recommendation device based on user behavior involved in an embodiment of the present application.
  • the product recommendation device may include a processor 1001 (for example, a central processing unit, CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005.
  • the communication bus 1002 is used to realize the connection and communication between these components;
  • the user interface 1003 may include a display (Display), an input unit such as a keyboard (Keyboard);
  • the network interface 1004 may optionally include a standard wired interface, a wireless interface (Such as wireless fidelity WIreless-FIdelity, WI-FI interface);
  • the memory 1005 can be a high-speed random access memory (random access memory, RAM), or a stable memory (non-volatile memory), such as disk memory, memory 1005 may optionally be a storage device independent of the foregoing processor 1001.
  • RAM random access memory
  • non-volatile memory such as disk memory
  • memory 1005 may optionally be a storage device independent of the foregoing processor 1001.
  • the memory 1005 in FIG. 1 as a computer-readable storage medium may include an operating system, a network communication module, and computer-readable instructions.
  • the network communication module is mainly used to connect to a database and perform data communication with the database; and the processor 1001 can call computer-readable instructions stored in the memory 1005 and perform product recommendation based on user behavior provided by embodiments of the present application method.
  • the embodiments of the present application provide a product recommendation method based on user behavior.
  • FIG. 2 is a schematic flowchart of a first embodiment of a product recommendation method based on user behavior in this application.
  • the product recommendation method based on user behavior includes the following steps:
  • Step S10 Obtain the historical behavior data of the user to be recommended, and determine the corresponding historical product set according to the historical behavior data;
  • This embodiment proposes a product recommendation method based on user behavior, analyzes the user's real interest based on the user's historical behavior data, and then obtains and pushes the recommended products associated with the product based on the real interest product, so that the product recommendation result matches the user's Actual needs, so as to improve the recommendation effect.
  • the product recommendation method based on user behavior in this embodiment is implemented by a product recommendation device based on user behavior.
  • the device uses a recommendation server as an example for description; and for the product, it may be financial products such as stocks, funds, and insurance. However, for other digital products, daily necessities, etc., in this embodiment, financial products are used as examples.
  • the recommendation server In order to implement targeted product recommendation, the recommendation server first needs to obtain the historical behavior data of the user to be recommended.
  • Historical behavior data are the historical operation behavior of the user to be recommended on some products, and their behavior types include but are not limited to browsing, searching, and clicking Products, collections (following), sharing, purchasing, etc.; of course, the historical behavior data also includes the occurrence time of each historical behavior, and the behavior object products of each historical behavior, and the collection of behavior object products of each historical behavior can be called Historical product set; for example, the user to be recommended browsed three financial products A, B, and C on x, y, and z, 2018, of which B was browsed by searching, and purchased product B, followed product C, of which The historical product set includes A, B, and C financial products.
  • the statistical software development kit SDK (Software Development Kit) can be embedded in the terminal application app in advance, and when the user to be recommended installs the terminal application app on their own user terminal, the SDK requests Obtain the operation behavior information of the user to be recommended; when the user to be recommended agrees, the user terminal will record the user's search, browsing and other behaviors according to the internal statistical logic of the SDK, combined with the identity information of the user to be recommended (such as terminal IP address, user Account, etc.) Generate corresponding historical behavior data, and then send the historical behavior data to the recommendation server.
  • the recommendation server can obtain the historical behavior data of the user to be recommended through the statistical SDK installed in the application of the user terminal.
  • the buried point logic on the script of the relevant product website for statistics.
  • the website server of the website will obtain the current visit. Identify the user terminal's identity information and record its behavior to generate historical behavior data, etc., and send the historical behavior data to the recommendation server; and the relevant statistical conditions of the buried point logic can be set according to the actual situation, for example, It means that the page stays longer than the preset time, performs specific operations on a specific element of the page (such as clicks, favorites, etc.), and uses specific keywords for retrieval.
  • the historical behavior data of the user to be recommended can also be collected by a third-party organization, and the recommendation server obtains these historical behavior data from the third-party organization.
  • Step S20 Analyze the historical behavior data based on a preset hidden Markov model, obtain the attitude probability of the potential attitude of the user to be recommended to each historical product in the historical product set, and according to the attitude probability The historical products collectively determine the products of real interest of the user to be recommended;
  • the historical behavior of the user to be recommended can reflect the interests of the user to be recommended to a certain extent, so when the recommendation server obtains historical behavior data and historical product sets, it will analyze the historical behavior data to determine the user Products of interest.
  • the historical behaviors of the users to be recommended are not necessarily operations performed by the users to be recommended driven by their own interests (or needs). It is possible that the actions of the users to be recommended are influenced by other factors; For example, the user to be recommended enters the browsing interface of financial products from other event promotion windows, and this historical behavior is an unconscious behavior; for example, the user to be recommended to retrieve a certain financial is not a purchase, but to search for a friend; These historical behaviors must not fully reflect the true interests of users.
  • Attitude probabilities such as wait-and-see probabilities and curiosity probabilities
  • Attitude probabilities such as wait-and-see probabilities and curiosity probabilities
  • Attitude probabilities of various historical products that have been involved are used to characterize the interest (or likelihood of interest) of the users to be recommended in these historical products, and further determine the users to be recommended according to the attitude probabilities Products of real interest; where the higher the probability of an attitude of a historical product (eg, the probability of watching is greater than the probability of curiosity), the more interested the user of the historical product is to be considered, (or the user of the historical product to be recommended Is more likely to be interested).
  • Hidden Markov Model is a statistical model. When an observation sequence and related model parameters are given, the hidden state probability corresponding to the observation sequence can be estimated based on a certain algorithm.
  • the historical behavior data acquired by the recommendation server can be converted into a corresponding observation sequence, and the true potential attitude (interest level) of the user to be recommended to the historical product when performing the historical behavior can correspond to the observation sequence
  • the probability value of is characterized, so in this embodiment, the degree of interest of the user to be recommended in the historical product can be predicted by estimating the potential attitude probability corresponding to the observation sequence.
  • the types of attitudes of the potential attitudes of the users to be recommended to each historical product include at least two types, for example, two types of wait-and-see attitudes (close attention) and curiosity attitudes. Of course, more types of attitudes may be included, and step S20 includes :
  • the recommendation server may first classify the behavior objects (that is, the historical products targeted by the behavior) to obtain various types of product behavior data corresponding to each historical product.
  • the product behavior data of product M includes: M products were browsed at z1 in 2017 x1, y1, and M products were purchased at z2 in 2017, x1, y2; product behavior data of product N is 2017 x1, y1 I browsed the N product on the day of z1, and browsed the N product on the day of x1, January 3, 2017.
  • the behavior data of each historical product of the user to be recommended can be obtained.
  • the recommendation server When obtaining product category behavior data, the recommendation server will perform statistical analysis on each product category data to determine the behavior type corresponding to each historical product, that is, determine the behavior history of the user to be recommended on each historical product;
  • the behavior time of the type behavior obtains the observation behavior sequence O corresponding to each historical product, and the earlier the historical behavior is, the higher the ranking is in the observation behavior sequence O, wherein each observation behavior sequence O includes at least one behavior.
  • the behavior time for paying attention to M products is T1
  • the behavior time for buying purchases for M products is T2
  • the observation behavior sequence O corresponding to the M product is (attention, purchase).
  • the preset behavior probability matrix B Based on the preset behavior probability matrix B, the preset attitude transition matrix A, the observed behavior sequence O, and the preset Viterbi algorithm, respectively obtain the attitude probability of the potential attitude of the user to be recommended to the historical products, and according to the The attitude probability determines the real interest product of the user to be recommended in the historical product set.
  • a behavior probability matrix B (confusion matrix) can be set in advance, and different rows represent the probability of each observed behavior under different potential attitudes, such as behavior probability matrix B
  • the values in the first row of the above behavior probability matrix B are the attitude behavior probability of the user browsing and paying attention (collection) when the user is curious about the product, and the values in the second row are the browsing, paying attention (collection) of the user when the product is in the wait-and-see attitude.
  • Probability of attitude and behavior for example, the value in the first column of the first row indicates that the behavior probability of the user browsing when the product is in a curious attitude is 0.7.
  • the initial probability of attitude of each potential attitude will also be set to characterize the attitude possibility of the user when they first contact the product.
  • the initial probability of attitude to curiosity and wait-and-see attitude are both set to 0.5;
  • the probability can also be recorded in the behavior probability matrix B. It is worth noting that both the probabilities of attitude and behavior and the initial probabilities of attitude can be set according to the actual situation.
  • an attitude change matrix A may also be preset to represent the probability of attitude change between different potential attitudes (from one potential attitude to another); it is worth noting that the attitude In the transition matrix A, the probability of keeping the same latent attitude unchanged when performing two kinds of behaviors consecutively (that is, to "hold the attitude S1 unchanged” is considered to be “convert the attitude S1 to the attitude S1").
  • the 0.4 value in the first row and first column of the above attitude change matrix A is the probability that the user maintains the curious attitude when the user continuously performs two actions on the product, and the 0.6 value in the first row and first column is the user's continuous two actions on the product
  • the above-mentioned attitude change probability can also be set according to the actual situation.
  • the recommendation server when obtaining the observed behavior sequence for each historical product, the recommendation server will obtain the preset behavior probability matrix B and the preset attitude transition matrix A, and then based on the preset behavior probability matrix B and the preset attitude transition
  • the matrix A, the observed behavior sequence O, and the preset Viterbi algorithm respectively obtain attitude probabilities of the potential attitudes of the user to be recommended to each historical product.
  • the Viterbi algorithm is a dynamic programming algorithm used to find the hidden state sequence that is most likely to produce the observed event sequence.
  • the attitude probability of each potential attitude can be estimated based on the Viterbi algorithm.
  • the process of estimating the attitude probability of each potential attitude through the Viterbi algorithm is described by taking a binary observation behavior sequence (the observation behavior sequence includes two behaviors) as an example, and the potential of the user to be recommended
  • the attitude may be set to include the first attitude S1 and the second attitude S2.
  • the sequence O of the observation behavior of the user to be recommended for a product includes two sequence behaviors, which are called the first sequence behavior and the second sequence behavior in sequence.
  • the first sequence behavior belongs to the first type behavior O L (such as browsing )
  • the first sequence of behavior belongs to the first type of behavior OL (such as browsing)
  • the preset behavior probability matrix B includes an attitude initial probability P(S1) of the user to be recommended initially in the first attitude S1, and an attitude 1 of the first type of behavior O L under the first attitude S1 of the user to be recommended Behavior Probability P(O L
  • the initial probability of the second attitude of S2 is P(S2), the user to be recommended is in the second attitude S2.
  • the probability of the second attitude of the first type of behavior OL is P(O L
  • the preset attitude transition matrix A it includes an attitude retention probability P(S1
  • the recommendation server When estimating, the recommendation server will first based on the initial probability of one attitude P(S1), the initial probability of two attitudes P(S2), the probability of one attitude and one behavior P(O L
  • the recommendation server When a sequence of attitude probabilities P S1
  • t 1 and a sequence of two attitude probabilities P S2
  • t 1 , the recommendation server will use a sequence of one attitude probabilities P S1
  • t 1 and a sequence of two attitude probabilities P S2
  • t 1 , one attitude holding probability P(S1
  • the two can be compared in size, and according to the size relationship between the two, it can be determined whether the historical product is a product of real interest for the user to be recommended.
  • the first attitude S1 is a wait-and-see (positive) attitude
  • the second attitude S2 is a curious (negative) attitude.
  • the probability of the second sequence one attitude corresponding to the first attitude S1 is greater than the probability of the second sequence two attitudes corresponding to the second attitude S2
  • the historical product can be judged to be the product of real interest of the user to be recommended; otherwise, if the probability of the second sequence-one attitude corresponding to the first attitude S1 is less than or equal to The second sequence of two attitude probabilities corresponding to the second attitude S2, then the historical product is considered not to be the product of real interest of the user to be recommended.
  • the observed behavior sequence can include more than three (here "above” includes the number, the same below), and the potential attitude of the user to be recommended can also include more than three kinds of
  • the calculation of the attitude probability of the user to be recommended to each historical product is also similar to the above embodiment, and will not be repeated here.
  • Step S30 Obtain recommended product information of a recommended product having an association relationship with the real interest product, and push the recommended product information to a user terminal corresponding to the user to be recommended based on a preset recommendation rule.
  • the recommendation server when the recommendation server determines the real interest product of the user to be recommended, it can perform a targeted product query according to the real interest product, obtain the recommended product having an association relationship with the real interest product, and obtain the recommended product Recommend product information.
  • the association relationship may be embodied in different types of real interest products. For example, for financial products such as stocks, funds, and insurance, the relationship can be similar in amount range, operating organization, and risk level. For digital products, the relationship can be the same function, similar price, and the same brand.
  • the recommendation server when the recommendation server obtains the recommended product information of the recommended product, it can push the recommended product information to the user terminal of the recommended user according to a preset recommendation rule.
  • the preset recommendation rule may include provisions such as recommendation time, pushing frequency, and pushing data amount.
  • the step of obtaining recommended product information of a recommended product having an association relationship with the real interest product includes:
  • the recommendation server may first determine the fund risk type of the real interest product (fund), such as conservative, robust, aggressive, etc., among which different fund risk types Corresponding to different risk levels. At the same time, the recommendation server will also obtain the stock holding information of the real interest product (fund), the stock holding information includes the stock name, the industry of each stock issuer, the market value of each stock holding, etc.; then it can be based on the stock holding information Determine the heavy stocks of the product of real interest, where the heavy stocks are the highest stocks in the market. When determining the heavy stock, the recommendation server will also determine the industry type of the heavy stock (ie, the industry to which the stock issuer belongs).
  • fund fund
  • the recommendation server will also obtain the stock holding information of the real interest product (fund), the stock holding information includes the stock name, the industry of each stock issuer, the market value of each stock holding, etc.; then it can be based on the stock holding information Determine the heavy stocks of the product of real interest, where the heavy stocks are the highest stocks in the market.
  • the recommendation server will
  • the recommendation server when determining the industry type of heavy stocks, the recommendation server will query the optional stocks of the industry type and obtain the stock price change information of these optional stocks within a preset period, and then determine these based on these stock price change information
  • Optional stock type of stock risk type For example, taking "one week" as a cycle, if a stock's stock price extreme value fluctuation range is less than 5%, then the stock's stock risk type is conservative; if the stock's stock price extreme value fluctuation range is 5% to 10% In between, the stock risk type of the stock is robust; if the extreme value fluctuation range of a stock is greater than 10%, the stock risk type of the stock is aggressive.
  • an optional stock having the same (or similar) risk level as the real interest product may be determined according to the stock risk type and fund risk type, for example, both are conservative Type, the same is robust type, etc.; the optional stock is the recommended stock associated with the real interest product, and the recommended stock is determined as the recommended product.
  • the recommendation server when determining the recommended product, can obtain the recommended product information of the recommended product for pushing to the user to be recommended.
  • the above determines the heavy stocks from the interest funds of the users to be recommended, and then selects the recommended stocks from the industry of the heavy stocks, thereby achieving the recommendation of products similar to the interest products in the industry field, and also helps reduce the interest fund operating agencies
  • the daily operation of the company will adversely affect the recommended stocks; and the selection of recommended stocks will also consider the user's risk tolerance level to improve the fit of the recommended products to the user's interests.
  • the step of obtaining recommended product information of the recommended product having an association relationship with the real interest product includes:
  • the recommendation server may first determine the holder of the stock, where the holding of the stock may include fund institutions, individuals, companies, etc.; then the recommendation server may Among the holders, the highest fund holding the stock with the highest amount is determined, that is, the fund with the largest share of the stock, and the operating institution of the highest fund is determined.
  • the recommendation server When determining the operating agency of the highest fund of the stock, the recommendation server will query all optional (purchasable) funds operated by the operating agency, and use these optional funds as recommended products.
  • the recommendation server when determining the recommended product, can obtain the recommended product information of the recommended product for pushing to the user to be recommended.
  • the above recommends the fund products of fund operating institutions that hold high amounts of stocks of interest, from the perspective of the operator to let users understand other fund products related to the stock, so that users can easily obtain the products they need and improve the recommendation effect.
  • the browsing habits of the user to be recommended can also be analyzed based on historical behavior data, and then targeted product recommendations can be made according to the user's browsing habits.
  • the recommendation server may analyze the historical behavior data of the recommended user to obtain the high-frequency browsing period of the user to be recommended; for example, the user to be recommended has 5 days in the past 7 days at 12 noon to 12:30, Browsing the product from 22 pm to 22:20 pm, the high frequency browsing period of the target user can be considered as 12 pm to 12:30 pm and 22 pm to 22:20 pm.
  • the recommendation server will also determine the period duration of each high-frequency browsing period.
  • the recommendation server When it is detected that the current time is in the high-frequency browsing period, the recommendation server will determine the amount of recommended information according to the period of the current high-frequency browsing period; where the relationship between the recommended information amount and the period of time may be a preset rule Set up, for example, 10 minutes corresponds to 1 product, 20 minutes corresponds to 3 products, etc.
  • the recommended information can also be characterized according to the type of product information, for example, 10 minutes corresponds to the product name and introduction, The 20-minute duration corresponds to the detailed introduction of the product.
  • the recommendation server may push the corresponding recommended product information to the user terminal of the user to be recommended according to the recommended information amount. In the above manner, the time and amount of information recommended by the product can be closer to the browsing habits of the user to be recommended, reducing the situation of user annoyance caused by invalid push, which is beneficial to improving the recommendation effect.
  • the real interest of the user is analyzed based on the historical behavior data of the user, and then the recommended products associated with the real interest product are obtained and pushed to make the product recommendation result meet the actual needs of the user, thereby improving the recommendation effect; at the same time, analyzing the user
  • the real interest is based on the hidden Markov model of the user's behavior, in order to estimate the user's attitude probability of each historical behavior, and then estimate the user's product attitude, and then determine the real interest based on the product attitude, so that it can be To a certain extent, it will reduce the adverse impact of non-realistic interest behavior data (noise) caused by user unconscious behavior browsing, advertising, marketing activities and other factors on user interest analysis, and improve the accuracy of interest analysis.
  • non-realistic interest behavior data noise
  • the recommended product information includes a manual service link, and after step S30, it also includes:
  • manual consultation service may also be provided for the user to be recommended.
  • the recommended product information pushed by the recommendation server includes a manual service link; after the recommended user browses the pushed recommended product information through the user terminal, if he needs to consult with the customer service staff manually, he can click the user terminal
  • the manual service link triggers the corresponding manual service request; the user terminal sends the manual service request to the recommendation server according to the operation of the user to be recommended.
  • the recommendation server When the recommendation server receives the manual service request, it will first query the corresponding manual customer service terminal (the terminal of the business person responsible for the credit product, product manager, etc.) according to the recommended product information, and send the corresponding to the manual customer service terminal Service task information; where the service task information can include the IP address, account name, phone number, etc. of the user terminal, so that the customer service staff can contact the user to be recommended through the manual customer service terminal, provide manual service for the target user, and improve the target User service experience.
  • the service task information can include the IP address, account name, phone number, etc. of the user terminal, so that the customer service staff can contact the user to be recommended through the manual customer service terminal, provide manual service for the target user, and improve the target User service experience.
  • an embodiment of the present application further provides a device for product recommendation based on user behavior.
  • the device for product recommendation based on user behavior includes:
  • a data acquisition module used to acquire historical behavior data of the user to be recommended, and determine the corresponding historical product set according to the historical behavior data
  • An interest determination module configured to analyze the historical behavior data based on a preset hidden Markov model, obtain an attitude probability of the potential attitude of the user to be recommended to each historical product in the historical product set, and according to the attitude Probability to determine the real interest products of the user to be recommended in the historical product set;
  • the information pushing module is used to obtain recommended product information of recommended products having an association relationship with the real interest product, and push the recommended product information to a user terminal corresponding to the user to be recommended based on a preset recommendation rule.
  • each virtual function module of the above-mentioned user behavior-based product recommendation device is stored in the memory 1005 of the user behavior-based product recommendation device shown in FIG. 1 and is used to implement all functions of computer-readable instructions; each module is used by the processor 1001 During execution, it is possible to obtain the user's historical behavior data, analyze the user's interest products from these historical behavior data, and perform related product push functions according to the interest product.
  • At least two types of attitudes of the potential attitudes of the users to be recommended to the historical products include:
  • the interest determination module includes:
  • a data classification unit configured to classify the historical behavior data according to the historical products corresponding to the historical behavior data to obtain the product behavior data corresponding to each historical product;
  • a sequence obtaining unit configured to separately count behavior types in the product behavior data, and obtain the observed behavior sequence O of each historical product according to the behavior time of each type of behavior;
  • Probability acquisition unit for acquiring the potential attitudes of the users to be recommended to the historical products based on the preset behavior probability matrix B, the preset attitude transition matrix A, the observed behavior sequence O, and the preset Viterbi algorithm, respectively Attitude probability, and determine the real interest product of the user to be recommended in the historical product set according to the attitude probability.
  • the potential attitude of the user to be recommended to the historical products includes a first attitude S1 and a second attitude S2, and the observed behavior sequence O includes a first sequence behavior and a second sequence behavior, wherein the first The sequence behavior belongs to the first type behavior O L , and the second sequence behavior belongs to the second type behavior O D ;
  • the preset behavior probability matrix B includes an initial attitude attitude of the user to be recommended initially in the first attitude S1 Probability P(S1), an attitude-behavior probability P(O L
  • S1) of the type behavior O D also includes the two-attitude initial probability P(S2) of the user to be recommended initially in the second attitude S2 and the second attitude S2 the first type of behavior for the two attitude O L a behavior probability P (O L
  • the first estimation subunit is used for according to the initial probability of one attitude P(S1), the initial probability of two attitudes P(S2), the probability of one attitude-behavior P( OL
  • the second estimation subunit is used for according to the one-sequence one-attitude probability P S1
  • t 1 , the one-sequence two-attitude probability P S2
  • t 1 , the one-attitude maintenance probability P(S1
  • the second-sequence attitude probability of each potential attitude when performing the second-sequence behavior, the second formula group is:
  • the probability comparison subunit is used to compare the two-sequence one-attitude probabilities with the two-sequence two-attitude probabilities, and determine whether each historical product is a real interest product of the user to be recommended according to the size relationship between the two.
  • the product type of the real interest product is a fund
  • the interest determination module 20 includes:
  • the first determining unit determines the fund risk type of the real interest product and the heavy stocks held by the real interest product, and determines the industry type of the heavy stocks;
  • the second determining unit is used to query the optional stocks corresponding to the industry type, and determine the stock risk type of the optional stocks according to the stock price changes of the optional stocks in a preset period;
  • a third determining unit configured to determine recommended stocks associated with the real interest product among the selectable stocks according to the stock risk type and the fund risk type, and determine the recommended stock as a recommended product;
  • the first obtaining unit is configured to obtain recommended product information of the recommended product.
  • the product type of the real interest product is stock
  • the interest determination module includes:
  • the fourth determination unit is used to determine the highest amount of funds holding the real interest product and determine the operating institution of the highest amount of funds
  • the fifth determining unit is used to query the optional funds operated by the operating agency and determine the optional funds as recommended products;
  • the second obtaining unit is configured to obtain recommended product information of the recommended product.
  • the information pushing module includes:
  • a period acquisition unit configured to acquire the high-frequency browsing period of the user to be recommended according to the historical behavior data, and determine the period duration of the high-frequency browsing period
  • the information pushing unit is used to determine the amount of recommended information according to the time period of the current high-frequency browsing period when the current time is in the high-frequency browsing period, and to the user terminal of the user to be recommended according to the recommended information amount Push the recommended product information.
  • the product recommendation information includes a manual service link
  • the product recommendation device based on user behavior further includes:
  • the task sending module is configured to, when receiving a manual service request sent by the user terminal based on the manual service link, query the corresponding manual customer service terminal according to the recommended product information, and send the corresponding service to the manual customer service terminal Task information.
  • each module in the device for recommending products based on user behavior corresponds to the steps in the embodiment of the method for recommending products based on user behavior, and the functions and implementation processes thereof will not be described here one by one.
  • an embodiment of the present application further provides a storage medium, and the computer-readable storage medium may be a non-volatile readable storage medium.
  • the storage medium of the present application stores computer-readable instructions, where the computer-readable instructions are executed by the processor to implement the steps of the product recommendation method based on user behavior as described above.
  • the computer-readable instructions are executed by the processor to implement the steps of the product recommendation method based on user behavior as described above.
  • the method implemented when the computer-readable instructions are executed reference may be made to various embodiments of the product recommendation method based on user behavior in this application, and details are not described herein again.

Landscapes

  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本申请涉及大数据分析领域，提供一种基于用户行为的产品推荐方法、装置、设备及存储介质，该方法包括：获取待推荐用户的历史行为数据，并确定根据所述历史行为数据确定对应的历史产品集；基于预设隐马尔可夫模型对所述历史行为数据进行分析，获取所述待推荐用户对所述历史产品集中各历史产品的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品；获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息，并基于预设推荐规则推送所述推荐产品信息。本申请可获取用户历史行为数据，从这些历史行为数据中分析用户的兴趣产品，并根据该兴趣产品进行关联产品推送，提升了产品推荐的效果。

Description

基于用户行为的产品推荐方法、装置、设备及存储介质
本申请要求于2019年1月4日提交中国专利局、申请号为201910014123.3、发明名称为“基于用户行为的产品推荐方法、装置、设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在申请中。
技术领域
本申请涉及大数据分析领域，尤其涉及一种基于用户行为的产品推荐方法、装置、设备及存储介质。
背景技术
随着网络技术的发展，网络成为用户获取产品的主要平台之一。如何有效的向网络用户推荐产品，是产品提供商重点关注的问题。传统的产品推荐方法，是通过在互联网网站上大量投放产品广告，或是在某些产品平台的首页推荐位置推荐相同的当前主推产品，即向不同的用户推荐相同的产品。然而这种“广撒网”式宣传行为，由于针对对象以及推荐的产品不明确，且不同的用户对同一产品的关注度不同，导致这种推荐方法的效果很不理想；同时，无用的产品推送还会耗费用户比较多的精力和时间，甚至于还会引起用户的反感，从而影响了产品的推广。
发明内容
本申请的主要目的在于提供一种基于用户行为的产品推荐方法、装置、设备及存储介质，旨在实现针对性的进行产品推荐，提升推荐效果。
为实现上述目的，本申请提供一种基于用户行为的产品推荐方法，所述基于用户行为的产品推荐方法包括：
获取待推荐用户的历史行为数据，并根据所述历史行为数据确定对应的历史产品集；
基于预设隐马尔可夫模型对所述历史行为数据进行分析，获取所述待推荐用户对所述历史产品集中各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品；
获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息，并基于预设推荐规则向所述待推荐用户对应的用户终端推送所述推荐产品信息。
此外，为实现上述目的，本申请还提供一种基于用户行为的产品推荐装置，所述基于用户行为的产品推荐装置包括：
数据获取模块，用于获取待推荐用户的历史行为数据，并根据所述历史行为数据确定对应的历史产品集；
兴趣确定模块，用于基于预设隐马尔可夫模型对所述历史行为数据进行分析，获取所述待推荐用户对所述历史产品集中各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品；
信息推送模块，用于获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息，并基于预设推荐规则向所述待推荐用户对应的用户终端推送所述推荐产品信息。
此外，为实现上述目的，本申请还提供一种基于用户行为的产品推荐设备，所述基于用户行为的产品推荐设备包括处理器、存储器、以及存储在所述存储器上并可被所述处理器执行的计算机可读指令，其中所述计算机可读指令被所述处理器执行时，实现如上述的基于用户行为的产品推荐方法的步骤。
此外，为实现上述目的，本申请还提供一种存储介质，所述存储介质上存储有计算机可读指令，其中所述计算机可读指令被处理器执行时，实现如上述的基于用户行为的产品推荐方法的步骤。
本申请基于用户的历史行为数据分析其真实兴趣，再根据真实兴趣产品获取与之关联的推荐产品并进行推送，使得产品推荐结果符合用户的实际需要，从而提升推荐效果；同时，在分析用户真实兴趣时是基于隐马尔可夫模型对用户的行为进行，以估算用户进行各历史行为时的态度概率，进而估测用户的产品态度，再根据产品态度进行真实兴趣的确定，从而可在一定程度上减少用户无意识行为浏览、广告、营销活动等因素引起的非真实兴趣行为数据(噪声)对用户兴趣分析造成的不利影响，提高兴趣分析的准确性。
附图说明
图1为本申请实施例方案中涉及的基于用户行为的产品推荐设备的硬件结构示意图；
图2为本申请基于用户行为的产品推荐方法第一实施例的流程示意图。
本申请目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
具体实施方式
应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请实施例涉及的基于用户行为的产品推荐方法主要应用于基于用户行为的产品推荐设备，该产品推荐设备可以是个人计算机(personal computer，PC)、服务器等具有数据处理 功能的设备实现的。
参照图1，图1为本申请实施例方案中涉及的基于用户行为的产品推荐设备的硬件结构示意图。本申请实施例中，该产品推荐设备可以包括处理器1001(例如中央处理器Central Processing Unit，CPU)，通信总线1002，用户接口1003，网络接口1004，存储器1005。其中，通信总线1002用于实现这些组件之间的连接通信；用户接口1003可以包括显示屏(Display)、输入单元比如键盘(Keyboard)；网络接口1004可选的可以包括标准的有线接口、无线接口(如无线保真WIreless-FIdelity，WI-FI接口)；存储器1005可以是高速随机存取存储器(random access memory，RAM)，也可以是稳定的存储器(non-volatile memory)，例如磁盘存储器，存储器1005可选的还可以是独立于前述处理器1001的存储装置。本领域技术人员可以理解，图1中示出的硬件结构并不构成对本申请的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。图1中作为一种计算机可读存储介质的存储器1005可以包括操作系统、网络通信模块以及计算机可读指令。在图1中，网络通信模块主要用于连接数据库，与数据库进行数据通信；而处理器1001可以调用存储器1005中存储的计算机可读指令，并执行本申请实施例提供的基于用户行为的产品推荐方法。
本申请实施例提供了一种基于用户行为的产品推荐方法。
参照图2，图2为本申请基于用户行为的产品推荐方法第一实施例的流程示意图。
本实施例中，所述基于用户行为的产品推荐方法包括以下步骤：
步骤S10，获取待推荐用户的历史行为数据，并根据所述历史行为数据确定对应的历史产品集；
本实施例提出了一种基于用户行为的产品推荐方法，根据用户的历史行为数据分析用户的真实兴趣，再根据真实兴趣产品获取与之关联的推荐产品并进行推送，使得产品推荐结果符合用户的实际需要，从而提升推荐效果。本实施例基于用户行为的产品推荐方法是由基于用户行为的产品推荐设备实现的，该设备以推荐服务器为例进行说明；而对于该产品，则可以是股票、基金、保险等金融产品，还可是其它数码产品、日用品等，本实施例中以金融产品为例进行说明。推荐服务器为了实现针对性的产品推荐，首先需要获取待推荐用户的历史行为数据，这些历史行为数据为待推荐用户在对一些产品的历史操作行为，其行为类型包括但不限于浏览、搜索、点击产品、收藏(关注)、分享、购买等；当然，历史行为数据中还包括有各历史行为的发生时间，以及各历史行为的行为对象产品，而各历史行为的行为对象产品的集合可称为历史产品集；例如待推荐用户在2018年x月y日z时浏览了A、B、C三款金融产品，其中B是通过搜索的方式浏览，且购买了B产品，关注了C产品，其中历史产品集包括A、B、C三款金融产品。
对于上述的历史行为数据获取，可以预先在终端应用app中内嵌统计软件开发工具包SDK(Software Development Kit)，当待推荐用户在自己的用户终端上安装该终端应用app时，通过该SDK请求获取待用户的操作行为信息；当待推荐用户同意时，用户终端将根据SDK的内在统计逻辑对用户的搜索、浏览等行为进行记录，并结合待推荐用户的身份信息(例如终端IP地址、用户账户等)生成对应的历史行为数据，然后将该历史行为数据发送至推荐服务器中，此时，推荐服务器即可通过安装在用户终端的应用中的统计SDK获取到待推荐用户的历史行为数据。此外，还可以是在相关产品网站的脚本上进行设置埋点逻辑进行统计，当待推荐用户通过用户终端访问产品网站时，若满足某一统计条件，则该网站的网站服务器将获取当前访问的用户终端的身份信息、并记录其行为，从而生成历史行为数据等，并将这些历史行为数据发送至推荐服务器；而对于埋点逻辑的相关统计条件，则可以是根据实际情况进行设置，例如可以是页面停留时长超过预设时间、对页面的某个特定元素执行特定的操作(如点击、收藏等)、检索使用了特定关键字等。当然，在实际中，待推荐用户的历史行为数据也可以是由第三方机构进行收集，推荐服务器则是从该第三方机构获取这些历史行为数据。
步骤S20，基于预设隐马尔可夫模型对所述历史行为数据进行分析，获取所述待推荐用户对所述历史产品集中各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品；
本实施例中，对于待推荐用户的历史行为，在一定程度可以反映待推荐用户兴趣所在，因此推荐服务器在得到历史行为数据和历史产品集时，将根据该历史行为数据进行分析，以确定用户的兴趣产品。但值得说明的是，待推荐用户的历史行为不一定均为待推荐用户在其自身的兴趣(或需求)驱使下进行的操作，有可能是受到其它因素影响才促使待推荐用户进行的行为；例如待推荐用户是从其它的活动促销窗口进入到金融产品的浏览界面，该项历史行为是无意识的行为；又例如待推荐用户检索某个金融并非自己购买，而是为了给朋友进行检索；对于这些历史行为并一定不能完全反映用户的真实兴趣所在。同时，对于不同类型的历史行为，其对待推荐用户兴趣点的反映力度也不一定相同；例如待推荐用户购买了B产品，关注了C产品，浏览了A产品(未对A产品进行其它点击、关注等特别操作)，在这种情况下，待推荐用户对A、B、C三种产品的兴趣度可认为是不同的，其对B产品最感兴趣，其次分别为C产品和A产品。因此，本实施例在根据待推荐用户的历史行为数据分析其兴趣产品时，将基于预设隐马尔可夫模型、结合历史行为数据的行为类型和行为顺序进行分析和估算，得出待推荐用户曾经涉及到的各历史产品的态度概率(如观望概率、好奇概率)，用以表征待推荐用户对这些历史产品的兴趣程度(或兴趣可能性)，并进一步根据该态度概率确定待推荐用户的真实兴 趣产品；其中，某个历史产品的某个态度概率越高(如观望概率大于好奇概率)，则可认为待推荐用户对该历史产品的越感兴趣，(或待推荐用户对该历史产品的感兴趣的可能性越高)。
对于隐马尔可夫模型(Hidden MarkovModel，HMM)是一种统计模型，在给定观察序列和相关模型参数时，可以基于一定的算法来估算该观察序列所对应的隐含状态概率。本实施例中，推荐服务器所获取的历史行为数据，可转化为对应的观察序列，而待推荐用户在进行历史行为时对历史产品的真实潜在态度(兴趣程度)，则可通过该观察序列对应的概率值进行表征，因此本实施例中可通过估算该观察序列所对应的潜在态度概率，来预测待推荐用户对历史产品的兴趣程度。
具体的，待推荐用户对各历史产品的潜在态度的态度类型，至少包括两种，例如包括观望态度(密切关注)和好奇态度两种，当然可以包括更多的态度类型，所述步骤S20包括：
根据所述历史行为数据所对应的历史产品对所述历史行为数据进行分类，得到各历史产品对应的产品类行为数据；
对于步骤S10获得的历史行为数据，推荐服务器可先根据其行为对象(也即该行为针对的历史产品)进行分类，从而得到各历史产品对应的各类产品类行为数据。例如，产品M的产品类行为数据包括：2017年x1月y1日z1时浏览了M产品、2017年x1月y2日z2时购买了M产品；N产品的产品类行为数据为2017年x1月y1日z1时浏览了N产品，2017年x1月y3日z3时浏览了N产品。通过上述处理，可得到待推荐用户针对每种历史产品的行为数据。
分别统计所述产品类行为数据中的行为类型，并根据各类型行为的行为时间获取所述各历史产品的观察行为序列O；
在得到产品类行为数据时，推荐服务器将对各产品类数据进行统计分析，确定各历史产品所对应的行为类型，也即确定待推荐用户对各历史产品进行过的行为历史；然后将根据各类型行为的行为时间获取各历史产品对应的观察行为序列O，越早进行的历史行为，其在观察行为序列O中的排序越靠前，其中每个观察行为序列O中至少包括一种行为。例如，对于M产品的产品类行为数据，包括关注(收藏)行为、购买行为，其中对M产品进行关注行为的行为时间为T1，对M产品进行购买行为的行为时间为T2，T1先于T2；则M产品对应的观察行为序列O为(关注、购买)。通过上述处理，可得到待推荐用户针对每种历史产品的行为特征。
基于预设行为概率矩阵B、预设态度转变矩阵A、所述观察行为序列O和预设维特比算法分别获取所述待推荐用户对所述各历史产品的潜在态度的态度概率，并根据所述态度概率 在所述历史产品集中确定所述待推荐用户的真实兴趣产品。
本实施例中，考虑到待推荐用户怀着不同的潜在态度对产品进行操作时，其可能进行的操作类型并不一定相同，而该差异性可以用一概率进行表征，如用户对产品为好奇态度时，进行浏览行为的概率0.7，进行关注行为的概率0.3，又例如当用户对产品处于观望(密切关注)态度时，浏览的概率0.2，关注的概率0.8。根据上述不同态度下的行为概率，可预先设置一行为概率矩阵B(混淆矩阵)，不同的行代表不同的潜在态度下进行各观察行为的概率，例如行为概率矩阵B
Figure PCTCN2019092103-appb-000001
上述行为概率矩阵B第一行数值分别为用户对产品处于好奇态度时进行浏览、关注(收藏)的态度行为概率，第二行数值分别为用户对产品处于观望态度时进行浏览、关注(收藏)的态度行为概率；如第一行第一列数值，表示用户对产品处于好奇态度时进行浏览的行为概率为0.7。同时，本实施例中还将设置各个潜在态度的态度初始概率，用以表征用户初次接触该产品时的态度可能性，例如对于好奇和观望态度的态度初始概率均设置为0.5；对于该态度初始概率，也可记录在该行为概率矩阵B中。值得说明的是，对于上述态度行为概率和态度初始概率，均可以是根据实际情况进行设置。
此外，在实际中，用户对某一产品连续进行两种行为时，其潜在态度可能是会发生变化的，例如用户对M产品依次进行了浏览和购买操作，其潜在态度则是由好奇态度转变为观望态度。对此，本实施例中还可预先设置一态度转变矩阵A，用以代表不同潜在态度之间的态度转变概率(由一种潜在态度转变为另一种态度)；值得说明的是，该态度转变矩阵A中，也可一同记录在连续进行两种行为时保持同一潜在态度不变的概率(也即将“保持态度S1不变”认为是“将态度S1转变为态度S1“)。例如态度转变矩阵A
Figure PCTCN2019092103-appb-000002
上述态度转变矩阵A第一行第一列的0.4数值为用户对产品连续进行两种行为时保持所述好奇态度的概率，第一行第一列的0.6数值为用户对产品连续进行两种行为时由好奇态度转变为观望态度的概率；第二行第一列的0.5数值为用户对产品连续进行两种行为时由观望态度转变为好奇态度的概率，第二行第二列的0.5数值为用户对产品连续进行两种行为时保持所述好奇态度的概率。当然，上述态度转变概率也可以是根据实际情况进行设置。
本实施例中，在得到针对每种历史产品的观察行为序列时，推荐服务器将获取预设行为概 率矩阵B、预设态度转变矩阵A，然后基于该预设行为概率矩阵B、预设态度转变矩阵A、观察行为序列O和预设维特比算法分别获取所述待推荐用户对各历史产品的各潜在态度的态度概率。其中，维特比算法是一种动态规划算法，用于寻找最有可能产生观测事件序列的隐含状态序列，本实施例中可基于维特比算法估算各潜在态度的态度概率。
为介绍方便，本实施例中，通过维特比算法估算各潜在态度的态度概率的过程，以二元观察行为序列(观察行为序列中包括两种行为)为例进行说明，而待推荐用户的潜在态度可设置为包括第一态度S1和第二态度S2两种。其中待推荐用户对某一产品的观察行为序列O包括两个序列行为，依其排序依次称为第一序列行为和第二序列行为，其中第一序列行为属于第一类型行为O L(如浏览)，第一序列行为属于第一类型行为O L(如浏览)，该观察行为序列也可表示为O＝(O L,O G)。而对于预设行为概率矩阵B，则包括待推荐用户初始处于第一态度S1的一态度初始概率P(S1)、待推荐用户处于第一态度S1下进行第一类型行为O L的一态度一行为概率P(O L|S1)、待推荐用户处于第一态度S1下进行第二类型行为O D的一态度二行为概率P(O D|S1)，还包括待推荐用户初始处于第二态度S2的二态度初始概率P(S2)、待推荐用户处于第二态度S2下进行第一类型行为O L的二态度一行为概率P(O L|S2)、待推荐用户处于所述第二态度S2下进行所述第二类型行为O D的二态度二行为概率P(O D|S2)。对于预设态度转变矩阵A，则包括待推荐用户连续进行两种行为时保持第一态度S1的一态度保持概率P(S1|S1)、待推荐用户连续进行两种行为时由第一态度S1转变为第二态度S2的一态度转变概率P(S2|S1)，还包括待推荐用户连续进行两种行为时保持第二态度S2的二态度保持概率P(S2|S2)、待推荐用户连续进行两种行为时由第二态度S2转变为第一态度S1的二态度转变概率P(S1|S2)。
在估算时，推荐服务器首先将根据一态度初始概率P(S1)、二态度初始概率P(S2)、一态度一行为概率P(O L|S1)、二态度一行为概率P(O L|S2)和第一公式组分别估算所述待推荐用户进行第一序列行为时各潜在态度的一序列态度概率，所述第一公式组为：
Figure PCTCN2019092103-appb-000003
其中，P S1|t＝1为待推荐用户进行第一序列行为时处于第一态度S1的一序列一态度概率，P S2|t＝1为待推荐用户进行第一序列行为时处于所述第二态度S2的一序列二态度概率。
在得到一序列态度概率P S1|t＝1和一序列二态度概率P S2|t＝1时，推荐服务器将根据一序列一态度概率P S1|t＝1、一序列二态度概率P S2|t＝1、一态度保持概率P(S1|S1)、一态度转变概率 P(S2|S1)、二态度保持概率P(S2|S2)、二态度转变概率P(S1|S2)和第二公式组分别估算所述待推荐用户进行所述第二序列行为时各潜在态度的二序列态度概率，所述第二公式组为：
Figure PCTCN2019092103-appb-000004
其中，P S1|t＝2为待推荐用户进行第二序列行为时处于第一态度S1的二序列一态度概率，P S2|t＝2为待推荐用户进行第二序列行为时处于第二态度S2的二序列二态度概率。
在计算得到二序列一态度概率和二序列二态度概率时，可将两者进行大小比较，并根据两者大小关系判断所述该历史产品是否为待推荐用户的真实兴趣产品。例如，该第一态度S1为观望(积极)态度，第二态度S2为好奇(消极)态度，若第一态度S1对应的二序列一态度概率大于第二态度S2对应的二序列二态度概率，则可认为待推荐用户对该历史产品感兴趣的可能性更大，则可判断该历史产品为待推荐用户的真实兴趣产品；反之，若第一态度S1对应的二序列一态度概率小于或等于第二态度S2对应的二序列二态度概率，则认为该历史产品不是待推荐用户的真实兴趣产品。值得说明的是，在实际中，观察行为序列可以包括三个以上(此处“以上”包括本数，下同)的序列行为，而待推荐用户的潜在态度也可以是包括三种以上，而对于待推荐用户对各历史产品的态度概率的计算也如上述实施例类似，此处不再赘述。
步骤S30，获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息，并基于预设推荐规则向所述待推荐用户对应的用户终端推送所述推荐产品信息。
本实施例中，推荐服务器确定待推荐用户的真实兴趣产品时，即可根据该真实兴趣产品进行针对性的产品查询，获取与该真实兴趣产品具有关联关系的推荐产品，并获取该推荐产品的推荐产品信息。其中，对于该关联关系，在不同类型的真实兴趣产品中可以是有不同的体现。例如，对于股票、基金、保险等金融产品，该关联关系可以金额范围类似、运营机构相同、风险等级类似等；对于数码产品，该关联关系可以是功能相同、价位相似、品牌相同等。本实施例中，推荐服务器在获取到推荐产品产品的推荐产品信息时，即可根据一预设推荐规则向推荐用户的用户终端推送该推荐产品信息。其中，该预设推荐规则可以包括推荐时间、推送频率、推送数据量等内容的规定。
可选地，当真实兴趣产品的产品类型为基金时，所述获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息步骤包括：
确定所述真实兴趣产品的基金风险类型和所述真实兴趣产品持有的重仓股票，并确定所述重仓股票的行业类型；
本实施例中，当真实兴趣产品的产品类型为基金时，推荐服务器可先确定该真实兴趣产品(基金)的基金风险类型，如保守型、稳健型、进取型等，其中不同的基金风险类型对应不同的风险等级。同时推荐服务器还将获取该真实兴趣产品(基金)的股票持有信息，该股票持有信息包括股票名称、各股票发行者所属行业、各股票持有市值等；然后可根据该股票持有信息确定该真实兴趣产品的重仓股票，其中该重仓股票为持有市场最高的股票。在确定重仓股票时，推荐服务器还将确定该重仓股票的行业类型(即股票发行者所属行业)。
查询所述行业类型对应的可选股票，并根据所述可选股票在预设周期的股价变化确定所述可选股票的股票风险类型；
本实施例中，当确定重仓股票的行业类型时，推荐服务器将查询该行业类型的可选股票，并获取这些可选股票在预设周期内的股价变化信息，然后根据这些股价变化信息确定这些可选股票类型的股票风险类型。例如，以“一周”为周期，若某一股票的股价极值波动范围在小于5％，则该股票的股票风险类型为保守型；若该股票的股价极值波动范围在5％到10％之间，则该股票的股票风险类型为稳健型；若某一股票的股价极值波动范围在大于10％，则该股票的股票风险类型为进取型等。
根据所述股票风险类型和所述基金风险类型在所述可选股票中确定与所述真实兴趣产品关联的推荐股票，并将所述推荐股票确定为推荐产品；
本实施例中，当确定各可选股票的股票风险类型时，可根据股票风险类型和基金风险类型确定与所述真实兴趣产品具有相同(或相似)风险等级的可选股票，例如同为保守型、同为稳健型等；该可选股票即为与所述真实兴趣产品关联的推荐股票，并将该推荐股票确定为推荐产品。
获取所述推荐产品的推荐产品信息。
本实施例中，在确定推荐产品时，推荐服务器即可获取该推荐产品的推荐产品信息，用以向待推荐用户进行推送。
以上通过从待推荐用户的兴趣基金中确定重仓股票，再从该重仓股票的所属行业中选择推荐股票，从而实现了推荐与兴趣产品的行业领域相似的产品，还有利于降低兴趣基金的运营机构的日常运营操作对推荐股票造成的不利影响；而且在选择推荐股票时还将会考虑用户的风险承受等级，提高推荐产品与用户的兴趣贴合度。
可选地，当真实兴趣产品的产品类型为股票时，所述获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息步骤包括：
确定最高额持有所述真实兴趣产品的最高额基金，并确定所述最高额基金的运营机构；
本实施例中，当真实兴趣产品的产品类型为股票时，推荐服务器可先确定该股票的持有者，其中该股票的持有可能包括基金机构、个人、公司等；然后推荐服务器可在这些持有者中确定最高额持有该股票的最高额基金，也即拥有该股票的份额最多基金，并确定该最高额基金的运营机构。
查询所述运营机构运营的可选基金，并将所述可选基金确定为推荐产品；
在确定该股票的最高额基金的运营机构时，推荐服务器将查询该运营机构所运营的所有可选(可购买)基金，并将这些可选基金作为推荐产品。
获取所述推荐产品的推荐产品信息。
本实施例中，在确定推荐产品时，推荐服务器即可获取该推荐产品的推荐产品信息，用以向待推荐用户进行推送。
以上通过推荐高额持有兴趣股票的基金运营机构的基金产品，以运营方的角度让用户了解与该股票的其它基金产品，方便用户获取到其需求产品，提升推荐效果。
本实施例中，为了进一步提高产品推荐的效果，还可以根据历史行为数据分析待推荐用户的浏览习惯，进而根据用户的浏览习惯进行针对性的产品推荐。具体的，推荐服务器可以对待推荐用户的历史行为数据进行分析，从而获取所述待推荐用户的高频浏览时段；例如待推荐用户在过去7天内有5天在中午12点至12点30分、晚上22点至22点20分浏览产品，则可认为目标用户的高频浏览时段为中午12点至12点30分、晚上22点至22点20分。当然，对于不同的高频浏览时段，对应的时段时长可能不同，待推荐用户所能查看的信息量也不同，因此，推荐服务器还将确定各高频浏览时段的时段时长。
当检测到当前时间处于高频浏览时段时，推荐服务器将根据目前所处的高频浏览时段的时段时长确定推荐信息量；其中对于该推荐信息量与时段时长的关系，可以是预置一规则进行设置，例如10分钟的时长对应1个产品，20分钟的时长对应3个产品等，当然该推荐信息量也可以是根据产品信息的类型进行表征，例如10分钟的时长对应产品名称和简介，20分钟的时长对应该产品的详细介绍等。当确定推荐信息量时，推荐服务器即可根据该推荐信息量向待推荐用户的用户终端推送对应的推荐产品信息。通过以上方式，可使得产品推荐的时间和信息量能够更贴近待推荐用户的浏览习惯，减少无效推送导致用户反感的情况，有利于提升推荐效果。
本实施例基于用户的历史行为数据分析其真实兴趣，再根据真实兴趣产品获取与之关联的推荐产品并进行推送，使得产品推荐结果符合用户的实际需要，从而提升推荐效果；同时，在分析用户真实兴趣时是基于隐马尔可夫模型对用户的行为进行，以估算用户进行各历史行为 时的态度概率，进而估测用户的产品态度，再根据产品态度进行真实兴趣的确定，从而可在一定程度上减少用户无意识行为浏览、广告、营销活动等因素引起的非真实兴趣行为数据(噪声)对用户兴趣分析造成的不利影响，提高兴趣分析的准确性。
基于上述图2所示实施例，提出本发明基于用户行为的产品推荐方法的第二实施例。本实施例中，所述推荐产品信息包括人工服务链接，步骤S30之后还包括：
在接收到所述用户终端基于所述人工服务链接发送的人工服务请求时，根据所述推荐产品信息查询对应的人工客服端，并向所述人工客服端发送对应的服务任务信息。
本实施例中，考虑到待推荐用户在浏览了推送的推荐产品信息后，可能会有疑问，为了方便待推荐用户进行咨询，本实施例中还可以为待推荐用户提供人工咨询服务。具体的，推荐服务器所推送的推荐产品信息中包括有人工服务链接；待推荐用户在通过用户终端浏览了推送的推荐产品信息后，若需要向客服人员进行人工咨询，则可通过用户终端点击该人工服务链接，从而触发对于的人工服务请求；用户终端根据待推荐用户的操作将该人工服务请求发送至推荐服务器。推荐服务器在接收到该人工服务请求时，首先将根据该推荐产品信息查询对应的人工客服端(负责该信贷产品的业务人员、产品经理等人的终端)，并向所述人工客服端发送对应的服务任务信息；其中该服务任务信息可以包括用户终端的IP地址、账户名称、电话号码等，从而使得客服人员能够通过人工客服端与待推荐用户进行联系，为目标用户提供人工服务，提升目标用户的服务体验。
此外，本申请实施例还提供一种基于用户行为的产品推荐装置，所述基于用户行为的产品推荐装置包括：
数据获取模块，用于获取待推荐用户的历史行为数据，并根据所述历史行为数据确定对应的历史产品集；
兴趣确定模块，用于基于预设隐马尔可夫模型对所述历史行为数据进行分析，获取所述待推荐用户对所述历史产品集中各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品；
信息推送模块，用于获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息，并基于预设推荐规则向所述待推荐用户对应的用户终端推送所述推荐产品信息。
其中，上述基于用户行为的产品推荐装置的各虚拟功能模块存储于图1所示基于用户行为的产品推荐设备的存储器1005中，用于实现计算机可读指令的所有功能；各模块被处理器1001执行时，可实现获取用户历史行为数据，并从这些历史行为数据中分析用户的兴趣产品，并根据该兴趣产品进行关联产品推送的功能。
进一步的，所述待推荐用户对所述各历史产品的潜在态度的态度类型至少包括两种，
所述兴趣确定模块包括：
数据分类单元，用于根据所述历史行为数据所对应的历史产品对所述历史行为数据进行分类，得到各历史产品对应的产品类行为数据；
序列获取单元，用于分别统计所述产品类行为数据中的行为类型，并根据各类型行为的行为时间获取所述各历史产品的观察行为序列O；
概率获取单元，用于基于预设行为概率矩阵B、预设态度转变矩阵A、所述观察行为序列O和预设维特比算法分别获取所述待推荐用户对所述各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品。
进一步的，所述待推荐用户对所述各历史产品的潜在态度包括第一态度S1和第二态度S2，所述观察行为序列O包括第一序列行为和第二序列行为，其中所述第一序列行为属于第一类型行为O L，所述第二序列行为属于第二类型行为O D；所述预设行为概率矩阵B包括所述待推荐用户初始处于所述第一态度S1的一态度初始概率P(S1)、处于所述第一态度S1下进行所述第一类型行为O L的一态度一行为概率P(O L|S1)、处于所述第一态度S1下进行所述第二类型行为O D的一态度二行为概率P(O D|S1)，还包括所述待推荐用户初始处于所述第二态度S2的二态度初始概率P(S2)、处于所述第二态度S2下进行所述第一类型行为O L的二态度一行为概率P(O L|S2)、处于所述第二态度S2下进行所述第二类型行为O D的二态度二行为概率P(O D|S2)；所述预设态度转变矩阵A包括所述待推荐用户连续进行两种行为时保持所述第一态度S1的一态度保持概率P(S1|S1)、所述待推荐用户连续进行两种行为时由所述第一态度S1转变为所述第二态度S2的一态度转变概率P(S2|S1)，还包括所述待推荐用户连续进行两种行为时保持所述第二态度S2的二态度保持概率P(S2|S2)、所述待推荐用户连续进行两种行为时由所述第二态度S2转变为所述第一态度S1的二态度转变概率P(S1|S2)；所述概率获取单元包括：
第一估算子单元，用于根据所述一态度初始概率P(S1)、所述二态度初始概率P(S2)、所述一态度一行为概率P(O L|S1)、所述二态度一行为概率P(O L|S2)和第一公式组分别估算所述待推荐用户进行所述第一序列行为时各潜在态度的一序列态度概率，所述第一公式组为：
Figure PCTCN2019092103-appb-000005
其中，P S1|t＝1为所述待推荐用户进行所述第一序列行为时处于所述第一态度S1的一序列 一态度概率，P S2|t＝1为所述待推荐用户进行所述第一序列行为时处于所述第二态度S2的一序列二态度概率；
第二估算子单元，用于根据所述一序列一态度概率P S1|t＝1、所述一序列二态度概率P S2|t＝1、所述一态度保持概率P(S1|S1)、所述一态度转变概率P(S2|S1)、所述二态度保持概率P(S2|S2)、所述二态度转变概率P(S1|S2)和第二公式组分别估算所述待推荐用户进行所述第二序列行为时各潜在态度的二序列态度概率，所述第二公式组为：
Figure PCTCN2019092103-appb-000006
其中，P S1|t＝2为所述待推荐用户进行所述第二序列行为时处于所述第一态度S1的二序列一态度概率，P S2|t＝2为所述待推荐用户进行所述第二序列行为时处于所述第二态度S2的二序列二态度概率；
概率比较子单元，用于将所述二序列一态度概率和所述二序列二态度概率进行大小比较，根据两者大小关系判断所述各历史产品是否为所述待推荐用户的真实兴趣产品。
进一步的，所述真实兴趣产品的产品类型为基金，所述兴趣确定模块20包括：
第一确定单元，确定所述真实兴趣产品的基金风险类型和所述真实兴趣产品持有的重仓股票，并确定所述重仓股票的行业类型；
第二确定单元，用于查询所述行业类型对应的可选股票，并根据所述可选股票在预设周期的股价变化确定所述可选股票的股票风险类型；
第三确定单元，用于根据所述股票风险类型和所述基金风险类型在所述可选股票中确定与所述真实兴趣产品关联的推荐股票，并将所述推荐股票确定为推荐产品；
第一获取单元，用于获取所述推荐产品的推荐产品信息。
进一步的，所述真实兴趣产品的产品类型为股票，所述兴趣确定模块包括：
第四确定单元，用于确定最高额持有所述真实兴趣产品的最高额基金，并确定所述最高额基金的运营机构；
第五确定单元，用于查询所述运营机构运营的可选基金，并将所述可选基金确定为推荐产品；
第二获取单元，用于获取所述推荐产品的推荐产品信息。
进一步的，所述信息推送模块包括：
时段获取单元，用于根据所述历史行为数据获取所述待推荐用户的高频浏览时段，并确 定所述高频浏览时段的时段时长；
信息推送单元，用于当当前时间处于所述高频浏览时段时，根据当前所处高频浏览时段的时段时长确定推荐信息量，并根据所述推荐信息量向所述待推荐用户的用户终端推送所述推荐产品信息。
进一步的，所述推荐产品信息包括人工服务链接，所述基于用户行为的产品推荐装置还包括：
任务发送模块，用于在接收到所述用户终端基于所述人工服务链接发送的人工服务请求时，根据所述推荐产品信息查询对应的人工客服端，并向所述人工客服端发送对应的服务任务信息。
其中，上述基于用户行为的产品推荐装置中各个模块的功能实现与上述基于用户行为的产品推荐方法实施例中各步骤相对应，其功能和实现过程在此处不再一一赘述。
此外，本申请实施例还提供一种存储介质，所述计算机可读存储介质可以为非易失性可读存储介质。本申请存储介质上存储有计算机可读指令，其中所述计算机可读指令被处理器执行时，实现如上述的基于用户行为的产品推荐方法的步骤。其中，计算机可读指令被执行时所实现的方法可参照本申请基于用户行为的产品推荐方法的各个实施例，此处不再赘述。
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。以上实施例并非限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种基于用户行为的产品推荐方法，其特征在于，所述基于用户行为的产品推荐方法包括：
    获取待推荐用户的历史行为数据，并根据所述历史行为数据确定对应的历史产品集；
    基于预设隐马尔可夫模型对所述历史行为数据进行分析，获取所述待推荐用户对所述历史产品集中各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品；
    获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息，并基于预设推荐规则向所述待推荐用户对应的用户终端推送所述推荐产品信息。
  2. 如权利要求1所述的基于用户行为的产品推荐方法，其特征在于，所述基于预设隐马尔可夫模型对所述历史行为数据进行分析，获取所述待推荐用户对所述历史产品集中各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品的步骤包括：
    根据所述历史行为数据所对应的历史产品对所述历史行为数据进行分类，得到各历史产品对应的产品类行为数据；
    分别统计所述产品类行为数据中的行为类型，并根据各类型行为的行为时间获取所述各历史产品的观察行为序列O；
    基于预设行为概率矩阵B、预设态度转变矩阵A、所述观察行为序列O和预设维特比算法分别获取所述待推荐用户对所述各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品。
  3. 如权利要求2所述的基于用户行为的产品推荐方法，其特征在于，所述待推荐用户对所述各历史产品的潜在态度包括第一态度S1和第二态度S2，
    所述观察行为序列O包括第一序列行为和第二序列行为，其中所述第一序列行为属于第一类型行为O L，所述第二序列行为属于第二类型行为O D；
    所述预设行为概率矩阵B包括所述待推荐用户初始处于所述第一态度S1的一态度初始概率P(S1)、处于所述第一态度S1下进行所述第一类型行为O L的一态度一行为概率P(O L|S1)、处于所述第一态度S1下进行所述第二类型行为O D的一态度二行为概率P(O D|S1)，还包括所述待推荐用户初始处于所述第二态度S2的二态度初始概率P(S2)、处于所述第二态度S2下进行所述第一类型行为O L的二态度一行为概率P(O L|S2)、处于所述 第二态度S2下进行所述第二类型行为O D的二态度二行为概率P(O D|S2)；
    所述预设态度转变矩阵A包括所述待推荐用户连续进行两种行为时保持所述第一态度S1的一态度保持概率P(S1|S1)、所述待推荐用户连续进行两种行为时由所述第一态度S1转变为所述第二态度S2的一态度转变概率P(S2|S1)，还包括所述待推荐用户连续进行两种行为时保持所述第二态度S2的二态度保持概率P(S2|S2)、所述待推荐用户连续进行两种行为时由所述第二态度S2转变为所述第一态度S1的二态度转变概率P(S1|S2)；
    所述基于预设行为概率矩阵B、预设态度转变矩阵A、所述观察行为序列O和预设维特比算法分别获取所述待推荐用户对所述各历史产品的潜在态度的态度概率，并根据h所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品的步骤包括：
    根据所述一态度初始概率P(S1)、所述二态度初始概率P(S2)、所述一态度一行为概率P(O L|S1)、所述二态度一行为概率P(O L|S2)和第一公式组分别估算所述待推荐用户进行所述第一序列行为时各潜在态度的一序列态度概率，所述第一公式组为：
    Figure PCTCN2019092103-appb-100001
    其中，P S1|t＝1为所述待推荐用户进行所述第一序列行为时处于所述第一态度S1的一序列一态度概率，P S2|t＝1为所述待推荐用户进行所述第一序列行为时处于所述第二态度S2的一序列二态度概率；
    根据所述一序列一态度概率P S1|t＝1、所述一序列二态度概率P S2|t＝1、所述一态度保持概率P(S1|S1)、所述一态度转变概率P(S2|S1)、所述二态度保持概率P(S2|S2)、所述二态度转变概率P(S1|S2)和第二公式组分别估算所述待推荐用户进行所述第二序列行为时各潜在态度的二序列态度概率，所述第二公式组为：
    Figure PCTCN2019092103-appb-100002
    其中，P S1|t＝2为所述待推荐用户进行所述第二序列行为时处于所述第一态度S1的二序列一态度概率，P S2|t＝2为所述待推荐用户进行所述第二序列行为时处于所述第二态度S2的二序列二态度概率；
    将所述二序列一态度概率和所述二序列二态度概率进行大小比较，根据两者大小关系判断所述各历史产品是否为所述待推荐用户的真实兴趣产品。
  4. 如权利要求1所述的基于用户行为的产品推荐方法，其特征在于，所述真实兴趣产品的产品类型为基金，
    所述获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息步骤包括：
    确定所述真实兴趣产品的基金风险类型和所述真实兴趣产品持有的重仓股票，并确定所述重仓股票的行业类型；
    查询所述行业类型对应的可选股票，并根据所述可选股票在预设周期的股价变化确定所述可选股票的股票风险类型；
    根据所述股票风险类型和所述基金风险类型在所述可选股票中确定与所述真实兴趣产品关联的推荐股票，并将所述推荐股票确定为推荐产品；
    获取所述推荐产品的推荐产品信息。
  5. 如权利要求1所述的基于用户行为的产品推荐方法，其特征在于，所述真实兴趣产品的产品类型为股票，
    所述获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息步骤包括：
    确定最高额持有所述真实兴趣产品的最高额基金，并确定所述最高额基金的运营机构；
    查询所述运营机构运营的可选基金，并将所述可选基金确定为推荐产品；
    获取所述推荐产品的推荐产品信息。
  6. 如权利要求1所述的基于用户行为的产品推荐方法，其特征在于，所述基于预设推荐规则向所述待推荐用户对应的用户终端推送所述推荐产品信息的步骤包括：
    根据所述历史行为数据获取所述待推荐用户的高频浏览时段，并确定所述高频浏览时段的时段时长；
    当当前时间处于所述高频浏览时段时，根据当前所处高频浏览时段的时段时长确定推荐信息量，并根据所述推荐信息量向所述待推荐用户的用户终端推送所述推荐产品信息。
  7. 如权利要求1所述的基于用户行为的产品推荐方法，其特征在于，所述推荐产品信息包括人工服务链接，
    所述获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息，并基于预设推荐规则向所述待推荐用户对应的用户终端推送所述推荐产品信息的步骤之后，还包括：
    在接收到所述用户终端基于所述人工服务链接发送的人工服务请求时，根据所述推荐产品信息查询对应的人工客服端，并向所述人工客服端发送对应的服务任务信息。
  8. 一种基于用户行为的产品推荐装置，其特征在于，所述基于用户行为的产品推荐装置包括：
    数据获取模块，用于获取待推荐用户的历史行为数据，并根据所述历史行为数据确定对应的历史产品集；
    兴趣确定模块，用于基于预设隐马尔可夫模型对所述历史行为数据进行分析，获取所述待推荐用户对所述历史产品集中各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品；
    信息推送模块，用于获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息，并基于预设推荐规则向所述待推荐用户对应的用户终端推送所述推荐产品信息。
  9. 如权利要求8所述的产品推荐装置，其特征在于，所述待推荐用户对所述各历史产品的潜在态度的态度类型至少包括两种，所述兴趣确定模块包括：
    数据分类单元，用于根据所述历史行为数据所对应的历史产品对所述历史行为数据进行分类，得到各历史产品对应的产品类行为数据；
    序列获取单元，用于分别统计所述产品类行为数据中的行为类型，并根据各类型行为的行为时间获取所述各历史产品的观察行为序列O；
    概率获取单元，用于基于预设行为概率矩阵B、预设态度转变矩阵A、所述观察行为序列O和预设维特比算法分别获取所述待推荐用户对所述各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品。
  10. 如权利要求9所述的产品推荐装置，其特征在于，所述待推荐用户对所述各历史产品的潜在态度包括第一态度S1和第二态度S2，所述观察行为序列O包括第一序列行为和第二序列行为，其中所述第一序列行为属于第一类型行为O L，所述第二序列行为属于第二类型行为O D；所述预设行为概率矩阵B包括所述待推荐用户初始处于所述第一态度S1的一态度初始概率P(S1)、处于所述第一态度S1下进行所述第一类型行为O L的一态度一行为概率P(O L|S1)、处于所述第一态度S1下进行所述第二类型行为O D的一态度二行为概率P(O D|S1)，还包括所述待推荐用户初始处于所述第二态度S2的二态度初始概率P(S2)、处于所述第二态度S2下进行所述第一类型行为O L的二态度一行为概率P(O L|S2)、处于所述第二态度S2下进行所述第二类型行为O D的二态度二行为概率P(O D|S2)；
    所述预设态度转变矩阵A包括所述待推荐用户连续进行两种行为时保持所述第一态度S1的一态度保持概率P(S1|S1)、所述待推荐用户连续进行两种行为时由所述第一态度S1转变为所述第二态度S2的一态度转变概率P(S2|S1)，还包括所述待推荐用户连续进行两种行为时保持所述第二态度S2的二态度保持概率P(S2|S2)、所述待推荐用户连续进行两种行为时由所述第二态度S2转变为所述第一态度S1的二态度转变概率P(S1|S2)；
    所述概率获取单元包括：
    第一估算子单元，用于根据所述一态度初始概率P(S1)、所述二态度初始概率P(S2)、所述一态度一行为概率P(O L|S1)、所述二态度一行为概率P(O L|S2)和第一公式组分别估算所述待推荐用户进行所述第一序列行为时各潜在态度的一序列态度概率，所述第一公式组为：
    Figure PCTCN2019092103-appb-100003
    其中，P S1|t＝1为所述待推荐用户进行所述第一序列行为时处于所述第一态度S1的一序列一态度概率，P S2|t＝1为所述待推荐用户进行所述第一序列行为时处于所述第二态度S2的一序列二态度概率；
    第二估算子单元，用于根据所述一序列一态度概率P S1|t＝1、所述一序列二态度概率P S2|t＝1、所述一态度保持概率P(S1|S1)、所述一态度转变概率P(S2|S1)、所述二态度保持概率P(S2|S2)、所述二态度转变概率P(S1|S2)和第二公式组分别估算所述待推荐用户进行所述第二序列行为时各潜在态度的二序列态度概率，所述第二公式组为：
    Figure PCTCN2019092103-appb-100004
    其中，P S1|t＝2为所述待推荐用户进行所述第二序列行为时处于所述第一态度S1的二序列一态度概率，P S2|t＝2为所述待推荐用户进行所述第二序列行为时处于所述第二态度S2的二序列二态度概率；
    概率比较子单元，用于将所述二序列一态度概率和所述二序列二态度概率进行大小比较，根据两者大小关系判断所述各历史产品是否为所述待推荐用户的真实兴趣产品。
  11. 如权利要求8所述的产品推荐装置，其特征在于，所述真实兴趣产品的产品类型为基金，所述兴趣确定模块包括：
    第一确定单元，确定所述真实兴趣产品的基金风险类型和所述真实兴趣产品持有的重仓股票，并确定所述重仓股票的行业类型；
    第二确定单元，用于查询所述行业类型对应的可选股票，并根据所述可选股票在预设周期的股价变化确定所述可选股票的股票风险类型；
    第三确定单元，用于根据所述股票风险类型和所述基金风险类型在所述可选股票中确定与所述真实兴趣产品关联的推荐股票，并将所述推荐股票确定为推荐产品；
    第一获取单元，用于获取所述推荐产品的推荐产品信息。
  12. 如权利要求8所述的产品推荐装置，其特征在于，所述真实兴趣产品的产品类型为股票，所述兴趣确定模块包括：
    第四确定单元，用于确定最高额持有所述真实兴趣产品的最高额基金，并确定所述最高额基金的运营机构；
    第五确定单元，用于查询所述运营机构运营的可选基金，并将所述可选基金确定为推荐产品；
    第二获取单元，用于获取所述推荐产品的推荐产品信息。
  13. 如权利要求8所述的产品推荐装置，其特征在于，所述信息推送模块包括：
    时段获取单元，用于根据所述历史行为数据获取所述待推荐用户的高频浏览时段，并确定所述高频浏览时段的时段时长；
    信息推送单元，用于当当前时间处于所述高频浏览时段时，根据当前所处高频浏览时段的时段时长确定推荐信息量，并根据所述推荐信息量向所述待推荐用户的用户终端推送所述推荐产品信息。
  14. 如权利要求8所述的产品推荐装置，其特征在于，所述推荐产品信息包括人工服务链接，所述基于用户行为的产品推荐还包括：
    任务发送模块，用于在接收到所述用户终端基于所述人工服务链接发送的人工服务请求时，根据所述推荐产品信息查询对应的人工客服端，并向所述人工客服端发送对应的服务任务信息。
  15. 一种基于用户行为的产品推荐设备，其特征在于，所述基于用户行为的产品推荐设备包括处理器、存储器、以及存储在所述存储器上并可被所述处理器执行的计算机可读指令，其中所述计算机可读指令被所述处理器执行时，实现如下步骤：
    获取待推荐用户的历史行为数据，并根据所述历史行为数据确定对应的历史产品集；
    基于预设隐马尔可夫模型对所述历史行为数据进行分析，获取所述待推荐用户对所述历史产品集中各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品；
    获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息，并基于预设推荐规则向所述待推荐用户对应的用户终端推送所述推荐产品信息。
  16. 如权利要求15所述的基于用户行为的产品推荐设备，其特征在于，所述基于预设隐马尔可夫模型对所述历史行为数据进行分析，获取所述待推荐用户对所述历史产品集中各历史 产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品的步骤包括：
    根据所述历史行为数据所对应的历史产品对所述历史行为数据进行分类，得到各历史产品对应的产品类行为数据；
    分别统计所述产品类行为数据中的行为类型，并根据各类型行为的行为时间获取所述各历史产品的观察行为序列O；
    基于预设行为概率矩阵B、预设态度转变矩阵A、所述观察行为序列O和预设维特比算法分别获取所述待推荐用户对所述各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品。
  17. 如权利要求15所述的基于用户行为的产品推荐设备，其特征在于，所述真实兴趣产品的产品类型为基金，
    所述获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息步骤包括：
    确定所述真实兴趣产品的基金风险类型和所述真实兴趣产品持有的重仓股票，并确定所述重仓股票的行业类型；
    查询所述行业类型对应的可选股票，并根据所述可选股票在预设周期的股价变化确定所述可选股票的股票风险类型；
    根据所述股票风险类型和所述基金风险类型在所述可选股票中确定与所述真实兴趣产品关联的推荐股票，并将所述推荐股票确定为推荐产品；
    获取所述推荐产品的推荐产品信息。
  18. 一种存储介质，其特征在于，所述存储介质上存储有计算机可读指令，其中所述计算机可读指令被处理器执行时，实现如下步骤：
    获取待推荐用户的历史行为数据，并根据所述历史行为数据确定对应的历史产品集；
    基于预设隐马尔可夫模型对所述历史行为数据进行分析，获取所述待推荐用户对所述历史产品集中各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品；
    获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息，并基于预设推荐规则向所述待推荐用户对应的用户终端推送所述推荐产品信息。
  19. 如权利要求18所述的存储介质，其特征在于，所述基于预设隐马尔可夫模型对所述历史行为数据进行分析，获取所述待推荐用户对所述历史产品集中各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品的步骤 包括：
    根据所述历史行为数据所对应的历史产品对所述历史行为数据进行分类，得到各历史产品对应的产品类行为数据；
    分别统计所述产品类行为数据中的行为类型，并根据各类型行为的行为时间获取所述各历史产品的观察行为序列O；
    基于预设行为概率矩阵B、预设态度转变矩阵A、所述观察行为序列O和预设维特比算法分别获取所述待推荐用户对所述各历史产品的潜在态度的态度概率，并根据所述态度概率在所述历史产品集中确定所述待推荐用户的真实兴趣产品。
  20. 如权利要求18所述的存储介质，其特征在于，所述真实兴趣产品的产品类型为基金，
    所述获取与所述真实兴趣产品具有关联关系的推荐产品的推荐产品信息步骤包括：
    确定所述真实兴趣产品的基金风险类型和所述真实兴趣产品持有的重仓股票，并确定所述重仓股票的行业类型；
    查询所述行业类型对应的可选股票，并根据所述可选股票在预设周期的股价变化确定所述可选股票的股票风险类型；
    根据所述股票风险类型和所述基金风险类型在所述可选股票中确定与所述真实兴趣产品关联的推荐股票，并将所述推荐股票确定为推荐产品；
    获取所述推荐产品的推荐产品信息。
PCT/CN2019/092103 2019-01-04 2019-06-20 基于用户行为的产品推荐方法、装置、设备及存储介质 Ceased WO2020140400A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910014123.3 2019-01-04
CN201910014123.3A CN109741146B (zh) 2019-01-04 2019-01-04 基于用户行为的产品推荐方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2020140400A1 true WO2020140400A1 (zh) 2020-07-09

Family

ID=66363783

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/092103 Ceased WO2020140400A1 (zh) 2019-01-04 2019-06-20 基于用户行为的产品推荐方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN109741146B (zh)
WO (1) WO2020140400A1 (zh)

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111931065A (zh) * 2020-09-03 2020-11-13 猪八戒股份有限公司 基于lstm模型的商机推荐方法、系统、电子设备及介质
CN111951050A (zh) * 2020-08-14 2020-11-17 中国工商银行股份有限公司 理财产品推荐方法及装置
CN112150199A (zh) * 2020-09-21 2020-12-29 北京小米松果电子有限公司 垂类特征的确定方法及装置、存储介质
CN112199584A (zh) * 2020-09-23 2021-01-08 深圳市其乐游戏科技有限公司 个性化推荐方法及终端设备、推荐设备、存储介质
CN112612955A (zh) * 2020-12-18 2021-04-06 中国工商银行股份有限公司 基于深度学习的产品推送方法及系统
CN113656690A (zh) * 2021-08-16 2021-11-16 中国平安财产保险股份有限公司 产品推荐方法、装置、电子设备及可读存储介质
CN113742570A (zh) * 2020-12-01 2021-12-03 北京沃东天骏信息技术有限公司 一种商品搜索方法、终端设备及服务器
TWI755020B (zh) * 2020-08-13 2022-02-11 財團法人亞洲大學 應用程式監控分析系統
CN114817692A (zh) * 2021-01-20 2022-07-29 腾讯科技（深圳）有限公司 确定推荐对象的方法、装置和设备及计算机存储介质
CN115935090A (zh) * 2023-03-10 2023-04-07 北京锐服信科技有限公司 一种基于时间分片的数据查询方法及系统
CN116758322A (zh) * 2022-03-03 2023-09-15 清华大学 一种数据处理方法、装置、计算机设备及存储介质
CN116956170A (zh) * 2022-03-30 2023-10-27 中国移动通信集团山西有限公司 行为预测方法、装置、电子设备和存储介质
CN117495432A (zh) * 2023-11-22 2024-02-02 北京亿家老小科技有限公司 一种多时间尺度投票式用户潜在需求分析方法及系统
CN118297670A (zh) * 2024-04-02 2024-07-05 深圳市久芯网科技有限公司 一种基于大数据的电子元器件产品推广平台及方法
CN118967253A (zh) * 2024-07-19 2024-11-15 北京起创科技有限公司 一种算法推荐服务的数据管理方法及系统
TWI895753B (zh) * 2023-06-20 2025-09-01 韓商韓領有限公司 商品購買選項提供方法和系統
CN121303711A (zh) * 2025-10-13 2026-01-09 北京瑞星智腾科技有限公司 基于供需预测的农产品交易智能匹配与调度系统

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109741146B (zh) * 2019-01-04 2022-06-28 平安科技（深圳）有限公司 基于用户行为的产品推荐方法、装置、设备及存储介质
CN112307310A (zh) * 2019-07-29 2021-02-02 中国移动通信有限公司研究院 一种推荐方法及装置、设备、存储介质
CN112422594B (zh) * 2019-08-20 2022-05-31 中移(苏州)软件技术有限公司 一种信息推送方法、服务器及计算机可读存储介质
CN112989278B (zh) * 2019-12-12 2025-02-21 北京沃东天骏信息技术有限公司 确定状态数据的方法和装置
CN111090822A (zh) * 2019-12-18 2020-05-01 浙江口碑网络技术有限公司 业务对象的推送方法及装置
CN111008897B (zh) * 2019-12-23 2023-08-25 集奥聚合(北京)人工智能科技有限公司 一种基于雷达技术的银行卡拒件导流方法
CN111429293A (zh) * 2020-04-21 2020-07-17 重庆新致金服信息技术有限公司 一种保险产品的推荐系统及推荐方法
CN112540819B (zh) * 2020-12-08 2023-02-24 杭州讯酷科技有限公司 一种根据查询页面自动生成推荐详细页面及表单页面的方法
CN113420649B (zh) * 2021-06-22 2025-08-19 康键信息技术（深圳）有限公司 用户行为分析方法、装置、设备及存储介质
CN113516539A (zh) * 2021-07-29 2021-10-19 中移（杭州）信息技术有限公司 商品推荐方法、装置及计算机可读存储介质
CN113610580B (zh) * 2021-08-10 2023-09-19 平安科技（深圳）有限公司 产品推荐方法、装置、电子设备及可读存储介质
CN115357809B (zh) * 2022-07-22 2026-02-03 东软睿驰汽车技术(大连)有限公司 策略推荐方法、装置、设备及存储介质
CN116595200A (zh) * 2023-05-16 2023-08-15 深圳市酷开网络科技股份有限公司 媒资推荐方法、设备及计算机可读存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2012059057A (ja) * 2010-09-09 2012-03-22 Nippon Telegr & Teleph Corp <Ntt> 視聴確率算出方法及び装置及びプログラム及び視聴プラン推薦方法及び装置及びプログラム
CN102982131A (zh) * 2012-11-16 2013-03-20 杭州东信北邮信息技术有限公司 一种基于马尔科夫链的图书推荐方法
CN104182449A (zh) * 2013-05-20 2014-12-03 Tcl集团股份有限公司 基于用户兴趣建模的个性化视频推荐系统和方法
CN109741146A (zh) * 2019-01-04 2019-05-10 平安科技（深圳）有限公司 基于用户行为的产品推荐方法、装置、设备及存储介质

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11042898B2 (en) * 2014-03-18 2021-06-22 Staples, Inc. Clickstream purchase prediction using Hidden Markov Models
CN104008203B (zh) * 2014-06-17 2018-04-17 浙江工商大学 一种融入本体情境的用户兴趣挖掘方法
CN107369091B (zh) * 2016-05-12 2021-02-05 创新先进技术有限公司 产品推荐方法、装置及理财产品推荐方法
CN107944643A (zh) * 2017-12-22 2018-04-20 上海斐讯数据通信技术有限公司 一种基于隐马尔科夫模型的购买转化率预测方法及系统
CN108681961A (zh) * 2018-05-24 2018-10-19 平安普惠企业管理有限公司 信贷产品推介方法、装置、设备及计算机可读存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2012059057A (ja) * 2010-09-09 2012-03-22 Nippon Telegr & Teleph Corp <Ntt> 視聴確率算出方法及び装置及びプログラム及び視聴プラン推薦方法及び装置及びプログラム
CN102982131A (zh) * 2012-11-16 2013-03-20 杭州东信北邮信息技术有限公司 一种基于马尔科夫链的图书推荐方法
CN104182449A (zh) * 2013-05-20 2014-12-03 Tcl集团股份有限公司 基于用户兴趣建模的个性化视频推荐系统和方法
CN109741146A (zh) * 2019-01-04 2019-05-10 平安科技（深圳）有限公司 基于用户行为的产品推荐方法、装置、设备及存储介质

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI755020B (zh) * 2020-08-13 2022-02-11 財團法人亞洲大學 應用程式監控分析系統
CN111951050A (zh) * 2020-08-14 2020-11-17 中国工商银行股份有限公司 理财产品推荐方法及装置
CN111951050B (zh) * 2020-08-14 2023-08-11 中国工商银行股份有限公司 理财产品推荐方法及装置
CN111931065A (zh) * 2020-09-03 2020-11-13 猪八戒股份有限公司 基于lstm模型的商机推荐方法、系统、电子设备及介质
CN112150199A (zh) * 2020-09-21 2020-12-29 北京小米松果电子有限公司 垂类特征的确定方法及装置、存储介质
CN112199584A (zh) * 2020-09-23 2021-01-08 深圳市其乐游戏科技有限公司 个性化推荐方法及终端设备、推荐设备、存储介质
CN113742570A (zh) * 2020-12-01 2021-12-03 北京沃东天骏信息技术有限公司 一种商品搜索方法、终端设备及服务器
CN112612955A (zh) * 2020-12-18 2021-04-06 中国工商银行股份有限公司 基于深度学习的产品推送方法及系统
CN114817692A (zh) * 2021-01-20 2022-07-29 腾讯科技（深圳）有限公司 确定推荐对象的方法、装置和设备及计算机存储介质
CN113656690B (zh) * 2021-08-16 2023-06-02 中国平安财产保险股份有限公司 产品推荐方法、装置、电子设备及可读存储介质
CN113656690A (zh) * 2021-08-16 2021-11-16 中国平安财产保险股份有限公司 产品推荐方法、装置、电子设备及可读存储介质
CN116758322A (zh) * 2022-03-03 2023-09-15 清华大学 一种数据处理方法、装置、计算机设备及存储介质
CN116956170A (zh) * 2022-03-30 2023-10-27 中国移动通信集团山西有限公司 行为预测方法、装置、电子设备和存储介质
CN115935090A (zh) * 2023-03-10 2023-04-07 北京锐服信科技有限公司 一种基于时间分片的数据查询方法及系统
TWI895753B (zh) * 2023-06-20 2025-09-01 韓商韓領有限公司 商品購買選項提供方法和系統
CN117495432A (zh) * 2023-11-22 2024-02-02 北京亿家老小科技有限公司 一种多时间尺度投票式用户潜在需求分析方法及系统
CN117495432B (zh) * 2023-11-22 2024-05-28 北京亿家老小科技有限公司 一种多时间尺度投票式用户潜在需求分析方法及系统
CN118297670A (zh) * 2024-04-02 2024-07-05 深圳市久芯网科技有限公司 一种基于大数据的电子元器件产品推广平台及方法
CN118967253A (zh) * 2024-07-19 2024-11-15 北京起创科技有限公司 一种算法推荐服务的数据管理方法及系统
CN121303711A (zh) * 2025-10-13 2026-01-09 北京瑞星智腾科技有限公司 基于供需预测的农产品交易智能匹配与调度系统

Also Published As

Publication number Publication date
CN109741146A (zh) 2019-05-10
CN109741146B (zh) 2022-06-28

Similar Documents

Publication Publication Date Title
WO2020140400A1 (zh) 基于用户行为的产品推荐方法、装置、设备及存储介质
WO2020140399A1 (zh) 基于用户行为的产品推荐方法、装置、设备及存储介质
US11315155B2 (en) Method and system for exposing data used in ranking search results
US7594189B1 (en) Systems and methods for statistically selecting content items to be used in a dynamically-generated display
CN103827906B (zh) 用于选择、过滤或呈现可用的销售渠道的方法和系统
US9922333B2 (en) Automated multivariate behavioral prediction
US20080288347A1 (en) Advertising keyword selection based on real-time data
US20220036391A1 (en) Auto-segmentation
US8510309B2 (en) Selection and delivery of invitational content based on prediction of user interest
US9916589B2 (en) Advertisement selection using multivariate behavioral model
US20200019644A1 (en) Automated Assignment Of User Profile Values According To User Behavior
CN112001754A (zh) 用户画像生成方法、装置、设备及计算机可读介质
US20120054627A1 (en) Selection and delivery of invitational content based on prediction of user intent
EP3326070A1 (en) Cross-screen measurement accuracy in advertising performance
WO2017019643A1 (en) Targeting tv advertising slots based on consumer online behavior
US20190080348A1 (en) Exploration for search advertising
CN101203852A (zh) 自动广告投放
CN111242661A (zh) 优惠券发放方法和装置以及计算机系统和介质
US20170098240A1 (en) Systems and methods for establishing and utilizing a hierarchical bayesian framework for ad click through rate prediction
WO2020221022A1 (zh) 业务对象推荐方法
US20210192549A1 (en) Generating analytics tools using a personalized market share
CN112015970A (zh) 产品推荐方法、相关设备及计算机存储介质
CN119558901A (zh) 一种营销策略执行方法和相关设备
CN118886986A (zh) 产品推荐方法、装置、设备及存储介质
CN114549125A (zh) 物品推荐方法及装置、电子设备和计算机可读存储介质

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: 19907964

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: 19907964

Country of ref document: EP

Kind code of ref document: A1