CN116955800B - Cold start recommendation method and thompson sampling recommendation method - Google Patents
Cold start recommendation method and thompson sampling recommendation methodInfo
- Publication number
- CN116955800B CN116955800B CN202310675331.4A CN202310675331A CN116955800B CN 116955800 B CN116955800 B CN 116955800B CN 202310675331 A CN202310675331 A CN 202310675331A CN 116955800 B CN116955800 B CN 116955800B
- Authority
- CN
- China
- Prior art keywords
- information
- recommended
- recommendation
- target user
- preset
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0605—Pooling transaction partners, e.g. group buying or group selling
-
- 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
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Recommending goods or services
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Economics (AREA)
- Development Economics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a cold start recommendation method and a thompson sampling recommendation method, wherein the cold start recommendation method comprises the steps of obtaining information to be recommended and characteristic attributes of the information to be recommended; the method comprises the steps of carrying out dimension classification on feature attributes of information to be recommended to obtain feature information of at least two dimensions, obtaining known information of a target user, wherein the target user is a user without behavior data in a first preset time, selecting a plurality of recommendation information from the information to be recommended corresponding to the feature information of each dimension based on relevance between the known information and the feature information of the at least two dimensions, and generating a recommendation pool based on the plurality of recommendation information. According to the method and the device for recommending the information of the target user in the cold start period, the information is recommended from the recommended information corresponding to the plurality of dimensional characteristic information, so that the recommended hit rate of the information is improved, and the target user can quickly pass the cold start period.
Description
Technical Field
The invention relates to the technical field of user behavior recommendation, in particular to a cold start recommendation method and a thompson sampling recommendation method.
Background
The supply and demand platform is a public platform for users to issue supply and demand information and third-party service institutions to issue information. The existing supply and demand information matching mechanism is to recommend information to a user by classifying the information and adopting a thompson sampling recommendation method. The thompson sampling recommendation principle is that when a certain piece of information is triggered by actions such as clicking, browsing and contacting of a user, an association relation between the user and the information is established, and meanwhile recommendation exploration attempts of the same type of information are carried out according to the type of the information by updating two parameters alpha and beta of beta distributed.
In the case of a new user without historical behavior data, i.e. during a cold start of the user, an assumption of uniform distribution is currently adopted, the parameters alpha and beta of the beta distribution are set to 1, each piece of information is regarded as a probability of uniform distribution, and the recommendation probability is corrected by continuous recommendation and feedback.
However, in the initialization stage of a new user, namely in the cold start stage, the recommended information lacks priori data, and the system needs to adopt a completely random mode to carry out a large number of recommendation attempts, so that the recommendation quality is difficult to ensure, the user experience is poor, and the user loses interest and has no positive feedback, so that a vicious circle is formed.
Disclosure of Invention
In view of the foregoing, it is necessary to provide a cold start recommendation method and a thompson sampling recommendation method, which are used for solving the technical problems in the prior art that the recommendation hit rate of information is low and the user experience is poor due to the lack of user behavior data in the cold start period of the user.
In order to solve the above problems, in one aspect, the present invention provides a cold start recommendation method, including:
acquiring information to be recommended and characteristic attributes of the information to be recommended;
performing dimension classification on the characteristic attribute of the information to be recommended to obtain characteristic information of at least two dimensions;
acquiring known information of a target user, wherein the target user is a user without behavior data in a first preset time;
based on the relevance of the known information and the feature information of at least two dimensions, selecting a plurality of recommendation information from the information to be recommended corresponding to the feature information of each dimension, and generating a recommendation pool based on the plurality of recommendation information;
based on a preset recommendation algorithm, pushing information in the recommendation pool to a target user.
In some possible implementations, based on the correlation between the known information and the feature information of the at least two dimensions, selecting a plurality of recommendation information from the to-be-recommended information corresponding to the feature information of each dimension, and generating a recommendation pool based on the plurality of recommendation information includes:
Based on the relevance between the known information and any one of the at least two dimensions, selecting corresponding recommended information from the information to be recommended corresponding to the any one of the at least two dimensions;
And generating a recommendation pool based on all recommendation information corresponding to the at least two dimension characteristic information.
In some possible implementations, the feature information is any one of location information, heat value information, business information, and classification information.
In some possible implementations, if the feature information is enterprise information, the feature attribute corresponding to the enterprise information is an enterprise tag and information release time, and the known information is a target user login time;
based on the relevance between the known information and any one of the at least two dimensions, selecting corresponding recommended information from the information to be recommended corresponding to the any one dimension feature information, wherein the method comprises the following steps:
Based on the target user login time and the information release time, sequencing the information to be recommended according to the information release time sequence;
selecting at least one target enterprise label from a preset enterprise label pool;
Sequentially selecting a plurality of recommended information corresponding to the target enterprise label from the information to be recommended according to the sorting sequence based on the enterprise label;
The number of the plurality of recommended information is not more than the expected selection number of the information to be recommended corresponding to the enterprise information.
In some possible implementations, if the feature information is location information, the feature attribute corresponding to the location information is a geographic location of an information publisher, and the known information is a geographic location of a target user;
based on the relevance between the known information and any one of the at least two dimensions, selecting corresponding recommended information from the information to be recommended corresponding to the any one dimension feature information, wherein the method comprises the following steps:
determining a linear distance between a target user and an information publisher based on a preset distance algorithm, the information publisher geographic position and the target user geographic position;
Sequencing the information to be recommended from the near to the far according to the linear distance;
selecting a plurality of recommendation information from the information to be recommended according to the sequence;
The number of the plurality of recommended information is not more than the expected selection number of the information to be recommended corresponding to the position information.
In some possible implementations, if the feature information is hot value information, the hot value information includes information display amount, information click browsing amount, information click contact amount and information release time, and the known information includes target user login time;
based on the relevance between the known information and any one of the at least two dimensions, selecting corresponding recommended information from the information to be recommended corresponding to the any one dimension feature information, wherein the method comprises the following steps:
selecting information with information release time in a second preset time range from the information to be recommended based on the target user login time and the information release time, and generating a primary selection pool based on the information to be recommended with the information release time in the second preset time range;
determining a heat value of information to be recommended in the primary selection pool based on a preset heat value algorithm, the information display quantity, the information click browsing quantity and the information click contact quantity;
Sorting the information to be recommended in the primary selection pool according to the heat value;
selecting a plurality of recommendation information from the information to be recommended in the primary selection pool according to the sequence;
the number of the plurality of recommended information is not more than the estimated selection number of the information to be recommended corresponding to the heat value information.
In some possible implementations, the method further includes:
Determining the estimated selection quantity of the information to be recommended corresponding to the feature information of any single dimension based on the estimated selection total quantity of the information to be recommended and the estimated selection weight of the information to be recommended corresponding to the feature information of any single dimension;
The expected selection quantity of the information to be recommended corresponding to the feature information of any one dimension is any one of the expected selection quantity of the information to be recommended corresponding to the enterprise information, the expected selection quantity of the information to be recommended corresponding to the position information and the expected selection quantity of the information to be recommended corresponding to the heat value information.
In some possible implementations, before obtaining the known information of the target user, the method further includes:
Acquiring login and browsing information of a current user in a first preset time;
judging whether the current user logs in an account in a first preset time;
if the current user does not log in the account within the first preset time, the current user is judged to be a target user;
if the current user logs in the account within the first preset time, further judging whether the current user browses information within the preset time;
if the current user browses the information within the first preset time, the current user is judged to be a non-target user;
If the current user does not browse the information within the first preset time, the current user is judged to be the target user.
In some possible implementations, based on a preset recommendation algorithm, pushing information in the recommendation pool to the target user, and pushing information in the recommendation pool to the target user, further includes:
Judging whether the current information pushed to the target user triggers user behavior or not
If the current information triggers the user behavior, the current information is removed from the recommendation pool;
If the current information does not trigger the user behavior, further judging whether the recommended times of the current information reach preset recommended times or not;
if the current user reaches the preset recommended times, the current information is removed from the recommended pool;
if the current user does not reach the preset recommended times, the current information is put back into the recommended pool;
judging whether the quantity of the residual information in the recommendation pool is smaller than the preset information quantity or not;
if the quantity of the residual information in the recommendation pool is not less than the preset quantity of information, continuing recommendation;
And if the quantity of the residual information in the recommendation pool is smaller than the preset quantity of information, exiting the cold start mode.
On the other hand, the invention also provides a thompson sampling recommendation method, which is used for acquiring behavior data of a target user, wherein the behavior data is generated after information is recommended to the target user based on the cold start recommendation method according to any one of the above steps;
taking the behavior data of the target user as the prior probability of a preset thompson sampling algorithm;
and recommending the information to be recommended to the target user based on the preset thompson sampling algorithm.
The cold start recommendation method provided by the embodiment of the invention has the beneficial effects that firstly, the information to be recommended and the characteristic attribute of the information to be recommended are obtained, then, the characteristic attribute of the information to be recommended is subjected to dimension classification to obtain multi-dimensional characteristic information, and then, recommendation information is selected from the information to be recommended corresponding to each dimension characteristic information based on the relevance of the known information and the characteristic information of the target user, and a recommendation pool is generated to recommend the information to the target user, so that the target user in the cold start period is selectively recommended from a plurality of recommendation dimensions when the user behavior data is absent, the recommendation hit rate of the information is improved, the target user can quickly supercool the start period, and meanwhile, the multi-dimensional recommendation plays a role in uniformly recommending the information, and the information recommendation is prevented from being too concentrated.
Further, after cold start, feedback of the target user on the recommendation pool information is considered, and the fed-back behavior data is used as the prior probability of a follow-up recommendation algorithm, so that recommendation of the target user is performed through a Thompson sampling algorithm.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a cold start recommendation method according to an embodiment of the present invention;
FIG. 2 is a flowchart illustrating the step S104 of FIG. 1 according to an embodiment of the present invention;
FIG. 3 is a flowchart illustrating an embodiment of step S201 when the feature information in FIG. 2 is enterprise information according to the present invention;
FIG. 4 is a flowchart illustrating an embodiment of step S201 when the feature information in FIG. 2 is location information according to the present invention;
FIG. 5 is a flowchart illustrating an embodiment of step S201 when the characteristic information in FIG. 2 is heat value information according to the present invention;
FIG. 6 is a flowchart illustrating an embodiment of determining a target user according to the present invention;
FIG. 7 is a flowchart illustrating a recommendation pool information presentation logic according to an embodiment of the present invention
Fig. 8 is a schematic flow chart of an embodiment of a thompson sampling recommendation method provided by the invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It should be understood that the schematic drawings are not drawn to scale. A flowchart, as used in this disclosure, illustrates operations implemented according to some embodiments of the present invention. It should be appreciated that the operations of the flow diagrams may be implemented out of order and that steps without logical context may be performed in reverse order or concurrently. Moreover, one or more other operations may be added to or removed from the flow diagrams by those skilled in the art under the direction of the present disclosure. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor systems and/or microcontroller systems.
The description of "first," "second," etc. in the embodiments of the present invention is for descriptive purposes only and is not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a technical feature defining "first", "second" may include at least one such feature, either explicitly or implicitly. And/or, describes the association relation of the association objects, and indicates that three relations can exist, for example, A and/or B, and can indicate that A exists alone, A and B exist simultaneously, and B exists alone.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the invention. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.
To facilitate an understanding of the invention, some terminology will be first explained.
The user is cold started, namely recommending the possibly interested articles to the new user, the new user has little interaction data and lacks historical behavior data, so that the interest preference of the user can be difficult to capture, and the cold start problem is a very challenging problem in a recommendation system and is also a problem of high attention in the academic world in the industry.
Beta distribution is a density function of conjugate prior distribution as Bernoulli distribution and binomial distribution, and has important application in machine learning and mathematical statistics. In the probability theory, beta distribution is also called beta distribution, and refers to a set of continuous probability distributions defined in the (0, 1) interval.
The thompson sampling algorithm, also known as a random probability pairing algorithm, where the rewards of each arm are expected to follow some unknown distribution (the multi-arm gambling machine where each arm follows the bernoulli distribution is discussed primarily in the present invention), is a bezels She Siduo arm gambling machine problem algorithm, the main idea of the thompson sampling algorithm is to assume that the probability of generating benefit for the ith arm in the multi-arm gambling machine, mu i, has an a priori distribution Beta (alpha, beta), with a return of 1 if a decision is performed successfully, and a return of 0 otherwise. If decision term i has been executed and succeeded Q i (t) times, executed and failed P i (t) times, the posterior distribution of updating the ith decision variable is Beta (Q i(t)+1,Pi (t) +1), and the above result is taken as the prior distribution of decisions at the next round of iteration. Then, θ i (t) is sampled from Beta (Q i(t)+1,Pi (t) +1), so that i (t) =arg imaxθi (t) is selected, then the return value is observed, and finally if the return value is 1, Q i (t) is added with 1 on the original basis, and otherwise, the value of P i (t) is added with 1 on the original basis. In short, thompson makes deterministic selections based on the probability that each arm is the best arm. In a recommendation system, a plurality of recommendation strategies can be regarded as a multi-arm gambling machine, and each recommendation strategy has a certain probability of generating clicking or purchasing behavior of a user.
The invention provides a cold start recommendation method, which is described below.
Fig. 1 is a flow chart of an embodiment of a cold start recommendation method provided by the present invention, where, as shown in fig. 1, the cold start recommendation method includes:
S101, acquiring information to be recommended and characteristic attributes of the information to be recommended;
s102, classifying the feature attributes of the information to be recommended in dimensions to obtain feature information of at least two dimensions;
s103, acquiring known information of a target user, wherein the target user is a user without behavior data in a first preset time;
s104, selecting a plurality of recommendation information from the to-be-recommended information corresponding to the feature information of each dimension based on the relevance of the known information and the feature information of at least two dimensions, and generating a recommendation pool based on the plurality of recommendation information;
s105, pushing information in a recommendation pool to a target user based on a preset recommendation algorithm.
It should be noted that the information to be recommended refers to information ontology content issued by a user, the characteristic attribute of the information to be recommended refers to marking the information to be recommended by a platform after the user issues the information, and is used for representing some information such as time, position, publisher, click browsing amount and the like of the information to be recommended, the preset recommendation algorithm is random recommendation, the end condition of the preset recommendation algorithm is that all information in a recommendation pool is recommended and displayed, when all information in the recommendation pool aiming at a target user is displayed, the cold start is exited to enter an original system Topson sampling recommendation logic, and in the cold start recommendation process, the target user performs behavior feedback on the recommended information, including clicking, browsing, contact and the like, and the behavior data of the user can be accumulated.
It should also be noted that a cold-start recommendation pool has a copy for each target user, and once a recommendation pool copy is generated for a certain target user, any modification to the recommendation pool does not affect this copy.
Compared with the prior art, the cold start recommendation method provided by the embodiment of the invention has the advantages that firstly, the information to be recommended and the characteristic attribute of the information to be recommended are obtained, then, the characteristic attribute of the information to be recommended is subjected to dimension classification to obtain multi-dimensional characteristic information, and then, based on the relevance of the known information and the characteristic information of the target user, the recommendation information is selected from the information to be recommended corresponding to the characteristic information of each dimension, and the recommendation pool is generated to recommend the information to the target user, so that the target user in the cold start period is selectively recommended from a plurality of recommendation dimensions when the user behavior data is absent, the recommendation hit rate of the information is improved, and the target user can quickly supercool the start period.
In order to better expose long-tail information while improving hit rate of recommended information and avoid too centralized information recommended information, in some embodiments, the recommended pool has several recommended information corresponding to at least two dimensional feature information, specifically, as shown in fig. 2, step S104 includes:
S201, based on the relevance of the known information and any one of at least two dimensions, selecting corresponding recommended information from the information to be recommended corresponding to the any one dimension characteristic information;
s202, generating a recommendation pool based on all recommendation information corresponding to at least two dimension characteristic information.
It should be noted that, in other embodiments, only the recommendation information corresponding to the single-dimensional feature information may be in the recommendation pool.
In order to determine the feature information of at least two dimensions, the feature information is any one of position information, heat value information, enterprise information and classification information, and has different recommendation information selection rules aiming at the feature information of different dimensions.
Specifically, in one embodiment, if the feature information is the enterprise information, the feature attribute corresponding to the enterprise information is the enterprise tag and the information release time, and the known information includes the target user login time, as shown in fig. 3, step S201 includes:
s301, sorting information to be recommended according to the sequence of release time based on the login time and the information release time of a target user;
S302, selecting at least one target enterprise label from a preset enterprise label pool;
S303, sequentially selecting a plurality of recommended information corresponding to the target enterprise label from the information to be recommended according to the sequencing order based on the enterprise label of the information to be recommended;
the number of the plurality of recommended information is not more than the expected selection number of the information to be recommended corresponding to the enterprise information.
It should be noted that, when the number of the plurality of recommended information is smaller than the expected selection number of the information to be recommended corresponding to the enterprise information, that is, the total amount of the information to be recommended corresponding to the target enterprise tag is smaller than the expected selection number of the information to be recommended corresponding to the enterprise information, the selection is stopped after the information to be recommended corresponding to the target enterprise tag is selected.
It can be understood that, for the feature information being classified information, the recommended information selection rule is similar to the selection rule for the feature information being enterprise information, except that the feature attribute corresponding to the classified information is a classification tag and information release time.
In another embodiment, if the feature information is location information and the feature attribute corresponding to the location information is the geographic location of the information publisher, the known information includes the geographic location of the target user, as shown in fig. 4, step S201 includes:
S401, determining a linear distance between a target user and an information publisher based on a preset distance algorithm, the geographic position of the information publisher and the geographic position of the target user;
s402, sorting information to be recommended from the near to the far according to the linear distance;
s403, sequentially selecting a plurality of recommendation information from the information to be recommended according to the sequence;
the number of the plurality of recommended information is not more than the expected selection number of the information to be recommended corresponding to the position information.
It should be noted that, the preset distance algorithm calculates the linear distance between the target user and the information publisher by calculating the longitude and latitude.
In order to recommend quality of the recommended information corresponding to the location information, in some embodiments, after step S401, the method further includes selecting information with a linear distance within a preset distance range from the information to be recommended. Thus, the information is subjected to preliminary screening.
In still another embodiment, if the feature information is the popularity value information, and the feature attribute corresponding to the popularity value information is the information display amount, the information click browsing amount, the information click contact amount and the information release time, the known information includes the target user login time, as shown in fig. 5, step S201 includes:
s501, selecting information with information release time in a second preset time range from information to be recommended based on target user login time and information release time, and generating a primary selection pool based on the information to be recommended with information release time in the second preset time range;
S502, determining a heat value of information to be recommended in a primary selection pool based on a preset heat value algorithm, an information display amount, an information click browsing amount and an information click contact amount;
s503, sorting the information to be recommended in the primary selection pool according to the heat value;
s504, sequentially selecting a plurality of recommended information from the information to be recommended according to the sequence;
The number of the plurality of recommended information is not more than the expected selection number of the information to be recommended corresponding to the heat value information.
It should be noted that, the preset hotness value algorithm targets display quantity, click browsing quantity and click contact quantity, and performs linear superposition calculation according to different weights.
The method comprises the steps of classifying feature attributes of information to be recommended into a plurality of dimension feature information such as enterprise information, classification information, geographic position, heat value and the like according to dimensions, establishing data initialization depending on a pre-step, specifically, after information release, carrying out label portrait on enterprise users releasing the information according to the information such as enterprise scale, operation range, registered place, residence time, office position and the like, namely enterprise information, carrying out type classification on the released information such as office building renting, raw material purchasing or supplying, namely classification information, adding further labels such as geographic position, namely position information, namely display quantity, click browsing quantity and click contact quantity, namely heat value information, and carrying out buried point statistics on browsing, contact and residence time data of the information according to date, and meanwhile, dividing the information release time into the dimension feature information as auxiliary information.
In other embodiments, the feature attribute of the information to be recommended may be classified in other dimensions based on the actual application requirement, or further subdivided, and the feature information in at least two dimensions is not limited to enterprise information, classification information, geographic location, and heat value, but may be other dimensions related to the information portrait.
It should be noted that the information to be recommended corresponding to the feature information of each dimension is calculated independently, and the recommendation pool of the target user generated by the system finally can be automatically de-duplicated.
In order to facilitate the directional display of the information to be recommended so as to satisfy the strategic layout of the platform recommendation information, in some embodiments, the recommendation information duty ratio corresponding to the plurality of recommendation characteristic information of the recommendation pool may be adjusted, and specifically, the method further includes:
Determining the estimated selection quantity of the information to be recommended corresponding to the feature information of any single dimension based on the estimated selection total quantity of the preset information to be recommended and the estimated selection weight of the information to be recommended corresponding to the feature information of any single dimension;
The expected selection quantity of the information to be recommended corresponding to the feature information of any one dimension is any one of the expected selection quantity of the information to be recommended corresponding to the enterprise information, the expected selection quantity of the information to be recommended corresponding to the position information and the expected selection quantity of the information to be recommended corresponding to the heat value information.
It should be noted that, the total information amount in the recommendation pool and the weight of each dimension information can be set in a user-defined manner according to actual requirements, for example, according to different main pushing services of different periods of the platform, or according to different times, different customer group requirements, etc.
In order to obtain a better recommendation effect, in some embodiments of the present invention, the judgment condition of the target user is limited, and before step S101, as shown in fig. 6, the method further includes:
s601, acquiring login and browsing information of a current user in a first preset time;
s602, judging whether the current user logs in an account in a first preset time;
s603, if the current user does not log in the account within the first preset time, the current user is judged to be a target user;
s604, if the current user logs in the account within the first preset time, further judging whether the current user browses information within the preset time;
S605, if the current user browses information in a first preset time, the current user is judged to be a non-target user;
S606, if the current user does not browse the information within the first preset time, the current user is judged to be the target user.
It should be noted that, if the target user is judged, cold start recommendation is performed, and if the non-target user is judged, namely, the old user, original thompson sampling recommendation logic is performed.
In order to improve the display efficiency of the recommendation pool while ensuring the good feedback effect of the target user behavior, in some embodiments, the display and rejection conditions of the information in the recommendation pool are limited, as shown in fig. 7, in step S105, the information in the recommendation pool is pushed to the target user, and then the method further includes:
s701, judging whether current information pushed to a target user triggers user behavior or not;
s702, if the current information triggers the user behavior, the current information is removed from the recommendation pool;
S703, if the current information does not trigger the user behavior, further judging whether the recommended times of the current information reach preset recommended times;
S704, if the current user reaches the preset recommended times, the current information is removed from the recommended pool;
s705, if the current user does not reach the preset recommended times, the current information is put back into the recommended pool again;
S706, judging whether the quantity of the residual information in the recommendation pool is smaller than the preset information quantity;
s707, if the number of the residual contents is not less than the preset information number, continuing recommendation;
S708, if the quantity of the residual content is smaller than the preset information quantity, exiting the cold start mode.
The information triggering user behavior refers to clicking and browsing behaviors of a user on displayed information.
Because the cold start recommendation algorithm is recommended to be used in combination with the thompson recommendation algorithm, the embodiment of the invention also provides a thompson sampling recommendation method, which is applied to the situation that the target user exits the cold start mode, as shown in fig. 8, and comprises the following steps:
s801, acquiring behavior data of a target user, wherein the behavior data is generated after the cold start recommendation method recommends information to the target user;
s802, taking behavior data of a target user as the prior probability of a preset Thompson sampling algorithm;
S803, recommending the information to be recommended to the target user based on a preset Thompson sampling algorithm.
It should be noted that, in the cold start recommendation process, the target user may perform behavior feedback on the recommended information, including clicking, browsing, contacting, and other actions, which may be accumulated into behavior data. And the target user starts to recommend information by using the cold start recommendation method, and switches to the Topson sampling recommendation method after the information in the recommendation pool is displayed.
The cold start recommendation method and the thompson sampling recommendation method provided by the invention are described in detail, specific examples are used for describing the principle and implementation mode of the invention, the description of the examples is only used for helping to understand the method and core ideas of the invention, and meanwhile, the contents of the description are not to be construed as limiting the invention as long as the person skilled in the art can change in the specific implementation mode and application range according to the ideas of the invention.
Claims (8)
1. A cold start recommendation method, comprising:
acquiring information to be recommended and characteristic attributes of the information to be recommended;
performing dimension classification on the characteristic attribute of the information to be recommended to obtain characteristic information of at least two dimensions, wherein the characteristic information is enterprise information, and the characteristic attribute corresponding to the enterprise information is enterprise tag and information release time;
acquiring known information of a target user, wherein the target user is a user without behavior data in a first preset time, and the known information is a target user login time;
Based on the relevance of the known information and any one dimension characteristic information in the at least two dimensions, selecting corresponding recommendation information from the information to be recommended corresponding to the any one dimension characteristic information respectively;
pushing information in the recommendation pool to the target user based on a preset recommendation algorithm;
based on the relevance between the known information and any one of the at least two dimensions, selecting corresponding recommended information from the information to be recommended corresponding to the any one dimension feature information, wherein the method comprises the following steps:
Based on the target user login time and the information release time, sequencing the information to be recommended according to the information release time sequence;
selecting at least one target enterprise label from a preset enterprise label pool;
Sequentially selecting a plurality of recommended information corresponding to the target enterprise labels from the information to be recommended according to the sequencing order based on the enterprise labels of the information to be recommended;
The number of the plurality of recommended information is not more than the expected selection number of the information to be recommended corresponding to the enterprise information.
2. The method of claim 1, wherein the characteristic information is any one of location information, heat value information and classification information.
3. The cold start recommendation method according to claim 2, wherein if the feature information is location information, a feature attribute corresponding to the location information is a geographic location of an information publisher, and the known information is a geographic location of a target user;
based on the relevance between the known information and any one of the at least two dimensions, selecting corresponding recommended information from the information to be recommended corresponding to the any one dimension feature information, wherein the method comprises the following steps:
determining a linear distance between a target user and an information publisher based on a preset distance algorithm, the information publisher geographic position and the target user geographic position;
Sequencing the information to be recommended from the near to the far according to the linear distance;
selecting a plurality of recommendation information from the information to be recommended according to the sequence;
The number of the plurality of recommended information is not more than the expected selection number of the information to be recommended corresponding to the position information.
4. The cold start recommendation method according to claim 2, wherein if the feature information is hot value information, the feature attribute corresponding to the hot value information is information display amount, information click browsing amount, information click contact amount and information release time, and the known information includes target user login time;
based on the relevance between the known information and any one of the at least two dimensions, selecting corresponding recommended information from the information to be recommended corresponding to the any one dimension feature information, wherein the method comprises the following steps:
selecting information with information release time in a second preset time range from the information to be recommended based on the target user login time and the information release time, and generating a primary selection pool based on the information to be recommended with the information release time in the second preset time range;
determining a heat value of information to be recommended in the primary selection pool based on a preset heat value algorithm, the information display quantity, the information click browsing quantity and the information click contact quantity;
Sorting the information to be recommended in the primary selection pool according to the heat value;
selecting a plurality of recommendation information from the information to be recommended in the primary selection pool according to the sequence;
the number of the plurality of recommended information is not more than the estimated selection number of the information to be recommended corresponding to the heat value information.
5. A cold start recommendation method according to any one of claims 2-4, further comprising:
Determining the estimated selection quantity of the information to be recommended corresponding to the feature information of any single dimension based on the estimated selection total quantity of the information to be recommended and the estimated selection weight of the information to be recommended corresponding to the feature information of any single dimension;
The expected selection quantity of the information to be recommended corresponding to the feature information of any one dimension is any one of the expected selection quantity of the information to be recommended corresponding to the enterprise information, the expected selection quantity of the information to be recommended corresponding to the position information and the expected selection quantity of the information to be recommended corresponding to the heat value information.
6. The cold start recommendation method according to claim 1, wherein before obtaining the known information of the target user, the method further comprises:
Acquiring login and browsing information of a current user in a first preset time;
judging whether the current user logs in an account in a first preset time;
if the current user does not log in the account within the first preset time, the current user is judged to be a target user;
if the current user logs in the account within the first preset time, further judging whether the current user browses information within the preset time;
if the current user browses the information within the first preset time, the current user is judged to be a non-target user;
If the current user does not browse the information within the first preset time, the current user is judged to be the target user.
7. The cold start recommendation method according to claim 1, wherein pushing information in a recommendation pool to a target user is based on a preset recommendation algorithm, and further comprising:
Judging whether the current information pushed to the target user triggers user behavior or not;
if the current information triggers the user behavior, the current information is removed from the recommendation pool;
If the current information does not trigger the user behavior, further judging whether the recommended times of the current information reach preset recommended times or not;
if the current user reaches the preset recommended times, the current information is removed from the recommended pool;
if the current user does not reach the preset recommended times, the current information is put back into the recommended pool;
judging whether the quantity of the residual information in the recommendation pool is smaller than the preset information quantity or not;
if the quantity of the residual information in the recommendation pool is not less than the preset quantity of information, continuing recommendation;
And if the quantity of the residual information in the recommendation pool is smaller than the preset quantity of information, exiting the cold start mode.
8. A thompson sampling recommendation method, comprising:
acquiring behavior data of a target user, wherein the behavior data is generated after recommending information to the target user based on the cold start recommendation method according to any one of claims 1 to 7;
taking the behavior data of the target user as the prior probability of a preset thompson sampling algorithm;
and recommending the information to be recommended to the target user based on the preset thompson sampling algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310675331.4A CN116955800B (en) | 2023-06-07 | 2023-06-07 | Cold start recommendation method and thompson sampling recommendation method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310675331.4A CN116955800B (en) | 2023-06-07 | 2023-06-07 | Cold start recommendation method and thompson sampling recommendation method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116955800A CN116955800A (en) | 2023-10-27 |
| CN116955800B true CN116955800B (en) | 2025-12-16 |
Family
ID=88457278
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310675331.4A Active CN116955800B (en) | 2023-06-07 | 2023-06-07 | Cold start recommendation method and thompson sampling recommendation method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116955800B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112328882A (en) * | 2020-11-05 | 2021-02-05 | 中国平安人寿保险股份有限公司 | Information recommendation method and device, terminal equipment and storage medium |
| CN114385904A (en) * | 2020-10-22 | 2022-04-22 | 北京鸿享技术服务有限公司 | Information recommendation method, device, equipment and storage medium |
| CN114691972A (en) * | 2020-12-31 | 2022-07-01 | 深圳云天励飞技术股份有限公司 | Cold start recommended method, device, electronic device and storage medium |
| CN115660782A (en) * | 2022-11-02 | 2023-01-31 | 广州欢聚时代信息科技有限公司 | Scene type commodity recommendation method and device, equipment, medium and product thereof |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106802915B (en) * | 2016-12-09 | 2020-07-28 | 宁波大学 | Academic resource recommendation method based on user behaviors |
| CN107688606A (en) * | 2017-07-26 | 2018-02-13 | 北京三快在线科技有限公司 | The acquisition methods and device of a kind of recommendation information, electronic equipment |
| CN112035743B (en) * | 2020-08-28 | 2021-10-15 | 腾讯科技（深圳）有限公司 | Data recommendation method and device, computer equipment and storage medium |
| CN112115364B (en) * | 2020-09-23 | 2025-02-18 | 北京明略昭辉科技有限公司 | Recommendation method, system and computer device for cold start based on login operation |
| CN113360773B (en) * | 2021-07-07 | 2023-07-04 | 脸萌有限公司 | Recommendation method and device, storage medium and electronic equipment |
-
2023
- 2023-06-07 CN CN202310675331.4A patent/CN116955800B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114385904A (en) * | 2020-10-22 | 2022-04-22 | 北京鸿享技术服务有限公司 | Information recommendation method, device, equipment and storage medium |
| CN112328882A (en) * | 2020-11-05 | 2021-02-05 | 中国平安人寿保险股份有限公司 | Information recommendation method and device, terminal equipment and storage medium |
| CN114691972A (en) * | 2020-12-31 | 2022-07-01 | 深圳云天励飞技术股份有限公司 | Cold start recommended method, device, electronic device and storage medium |
| CN115660782A (en) * | 2022-11-02 | 2023-01-31 | 广州欢聚时代信息科技有限公司 | Scene type commodity recommendation method and device, equipment, medium and product thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116955800A (en) | 2023-10-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110266745B (en) | Information flow recommendation method, device, equipment and storage medium based on deep network | |
| CN111079015B (en) | Recommendation method and device, computer equipment and storage medium | |
| CA2757668C (en) | Systems, methods and apparatus for providing media content | |
| CN110147882B (en) | Neural network model training method, crowd diffusion method, device and equipment | |
| CN108446382B (en) | Method and apparatus for pushed information | |
| CN102999588A (en) | Method and system for recommending multimedia applications | |
| CN108521439A (en) | A kind of method and apparatus of message push | |
| CN109086439A (en) | Information recommendation method and device | |
| CN109872170A (en) | Advertising feedback data processing method, device and computer equipment | |
| CN111428885B (en) | A method for indexing users in federated learning and a federated learning device | |
| CN117150127A (en) | Using reinforcement learning to select content items | |
| Zhao et al. | MAESP: Mobility aware edge service placement in mobile edge networks | |
| CN114330752A (en) | Ranking model training method and ranking method | |
| CN109885774A (en) | Recommended method, device and the equipment of individualized content | |
| CN113688305A (en) | Information processing method and device and computer readable storage medium | |
| CN103577660A (en) | Gray scale experiment system and method | |
| CN116955800B (en) | Cold start recommendation method and thompson sampling recommendation method | |
| Krohn et al. | Modelling online comment threads from their start | |
| CN109688217B (en) | Message pushing method and device and electronic equipment | |
| CN109284932B (en) | A method and system for stranger social user evaluation based on big data | |
| CN108388652A (en) | A method and device for sending song list identification | |
| CN111275473B (en) | Content item delivery method, device, server and storage medium | |
| CN114581116B (en) | Push information screening method, push information screening device, computer equipment and storage medium | |
| CN112785328A (en) | Content pushing method and device and computer storage medium | |
| CN109960748A (en) | Method and system for prediction of individual live broadcast users watching anchors based on knowledge graph |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |