CN109508424B - Feature evolution-based streaming data recommendation method - Google Patents

Feature evolution-based streaming data recommendation method Download PDF

Info

Publication number
CN109508424B
CN109508424B CN201811542194.2A CN201811542194A CN109508424B CN 109508424 B CN109508424 B CN 109508424B CN 201811542194 A CN201811542194 A CN 201811542194A CN 109508424 B CN109508424 B CN 109508424B
Authority
CN
China
Prior art keywords
recommendation
feature set
model
data
training
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201811542194.2A
Other languages
Chinese (zh)
Other versions
CN109508424A (en
Inventor
程国艮
李欣杰
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.)
Global Tone Communication Technology Co ltd
Original Assignee
Global Tone Communication Technology 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 Global Tone Communication Technology Co ltd filed Critical Global Tone Communication Technology Co ltd
Priority to CN201811542194.2A priority Critical patent/CN109508424B/en
Publication of CN109508424A publication Critical patent/CN109508424A/en
Application granted granted Critical
Publication of CN109508424B publication Critical patent/CN109508424B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting

Landscapes

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

Abstract

The invention provides a problem of characteristic set evolution in a flow type data recommendation system, and an old recommendation model and a new recommendation model are considered simultaneously when recommendation models are alternated, so that smooth evolution of the recommendation models is achieved, and sudden reduction of recommendation performance during model switching is avoided. The model switching is expanded from one point to a time period, and in the time period of the model switching, a new characteristic set and an old characteristic set and a recommended model exist and play roles simultaneously; and combining the recommendation results obtained by the two models through the weight factor so as to obtain the final recommendation result in the time period. The recommendation system with the smoothly evolved features and models solves the problem of sudden reduction of recommendation performance caused by direct switching of feature sets and recommendation models in the original streaming data recommendation system.

Description

Feature evolution-based streaming data recommendation method
Technical Field
The invention relates to the field of data recommendation, in particular to the field of streaming data recommendation.
Background
Recommendation technology based on batch data has been developed for many years, and with the popularization of the internet of things and social networks, recommendation based on streaming data becomes a popular research direction in the field of content recommendation. Different from a common recommendation system, data only has feature vector description, in the recommendation system based on streaming data, time attributes are added to training data, and data input into a recommendation model are also in time sequence. In the streaming data recommendation system, a big data real-time processing related technology is needed, and the construction and training of a recommendation system model are different.
Because the streaming data recommendation system is trained based on time data, and the characteristics of the streaming data recommendation system, whether environmental data generated in the internet of things or interpersonal interaction data generated in a social network, change slowly with time, the streaming data recommendation system is significantly different from the batch data recommendation system. The current streaming data recommendation system mainly solves the problems of real-time recommendation and recommendation precision, such as the characteristic of huge real-time data volume in a real-time streaming data processing application software feature identification method, and provides an information organization method based on an M-tree so as to reduce the calculated amount; the distributed online recommendation method for the streaming data decomposes the recommendation task in a distributed mode, so that the quick learning and recommendation of real-time data are realized; a clustering-based streaming recommendation engine, a recommendation system and a recommendation method are combined with domain characteristics and association characteristics, so that more accurate recommendation performance is provided. However, there has been no relevant research on the problem of changing the feature set. For a dynamically changing feature set, an obvious scheme is to adopt model replacement, that is, after a period of time, new features are extracted and evaluated from the used model, so that the feature set is updated, and then a new recommended model is trained for the new feature set. This approach completely ignores the continuity of feature changes, completely abandons the previously used recommendation models, so that the recommendation performance becomes very poor at the time periods when the models alternate.
Disclosure of Invention
In view of this, the present invention provides a method for solving the problem of feature set evolution in a streaming data recommendation system, in which an old recommendation model and a new recommendation model are considered when recommendation models are alternated, so as to achieve smooth evolution of the recommendation models and avoid sudden degradation of recommendation performance during model switching.
In order to achieve the above purpose, the invention provides the following technical scheme:
performing learning training by adopting data in the first feature set in a first time period to obtain a first recommendation model and performing data recommendation;
in a transition time period of model switching, performing weighted calculation by using a first recommendation model obtained by learning and training data in a first feature set and a second recommendation model obtained by learning and training data in a second feature set through a weighting factor to obtain final data recommendation in the transition time period;
and performing learning training by adopting the data in the second feature set in a second time period to obtain a second recommendation model and performing data recommendation.
Specifically, in each round of training for the first time period, T is 1,21B and is an integer, data tag is
Figure BDA0001908434040000021
And is
Figure BDA0001908434040000022
Wherein R isd1Is S1Feature set space, d1 is S1Number of features in feature set, T1-B is the total number of training rounds;
in each round of training of the model switching transition period, T is T1-B,...,T1When, data is marked as
Figure BDA0001908434040000023
And
Figure BDA0001908434040000024
and is
Figure BDA0001908434040000025
Wherein R isd1Is S1Feature set space, d1 is S1Number of features in feature set, Rd2Is S2Feature set space, d2 is S2The number of features in the feature set, B is the total number of training rounds, and t is an integer;
in each round of training for the second period of time, T ═ T1,...,T2And is an integer, data tag is
Figure BDA0001908434040000026
And is
Figure BDA0001908434040000027
Wherein R isd2Is S2Feature set space, d2 is S2Number of features in feature set, T2-T1Is the total number of training rounds.
Definition < > denotes the inner product in Ω1∈Rd1Denotes S1The prediction model trained by the feature set is in omega2∈Rd2Denotes S2The prediction result is expressed as
Figure BDA0001908434040000031
Wherein wi,t∈Rdi，i＝1,2。
In the recommendation system based on streaming data, in each learning training process, an algorithm observes an example and gives a predicted value for the example, the predicted value is compared with a label value, and a loss value l is calculated according to the difference value, so that the model is improved.
Loss function l (w)Tx, y) are calculated from the predicted and labeled results, while the prediction model is updated by a loss function at each iteration, wherein the update function is
Figure BDA0001908434040000032
Wherein i is 1, 2.
The mapping relation between the features in the first feature set and the second feature set in the model switching transition time period is described by adopting a linear model, and the correlation matrix between the features is assumed to be M, and the duration is T1-B,...,T1Then the least square method is adopted to calculate
Figure BDA0001908434040000033
Wherein M isTRepresenting the transpose of M.
The optimal value M of the matrix M is calculated using the following formula,
Figure BDA0001908434040000034
in the transition time period of model switching, a final recommendation result is calculated by adopting a weight-based collaborative recommendation method, which specifically comprises the following steps:
pt＝α1,tf1,t+α2,tf2,twherein p istIndicates the final recommended value, αi,tIs the ith round of predicted weight values,
Figure BDA0001908434040000035
representing the second feature set S2Is mapped to a first set of features S1In (1),
Figure BDA0001908434040000036
the weighted value is updated by the following formula:
Figure BDA0001908434040000037
where i is 1,2, η is an empirical parameter and l is a loss function.
Compared with the prior art, the embodiment of the invention has the beneficial effects that: in the training process of the basic flow data recommendation model, the association between new and old features before and after feature replacement is not considered, so that the old recommendation model is completely abandoned. However, in a real-world scenario, the characteristics of streaming data are slowly changing, and there is a necessary link between new and old characteristics, so this complete alternative wastes the old recommendation model that could otherwise be used to assist in the prediction of new data. According to the method, the mapping from the new features to the old features is established according to the relation between the new features and the old features, so that the new data can be expressed according to the new and old features, and the streaming data can be recommended better by using the new and old recommendation models.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
FIG. 1 is a schematic diagram of a recommendation feature and a recommendation model based on feature evolution according to the present invention.
Fig. 2 is a schematic diagram of a recommendation model training algorithm provided in the embodiment of the present invention.
Fig. 3 is a schematic diagram of a collaborative recommendation algorithm according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of a comparison result between the collaborative recommendation model and the general real-time recommendation algorithm according to the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The terminology used in the embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the examples of the present invention and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms as well, and "a" and "an" generally include at least two, but do not exclude at least one, unless the context clearly dictates otherwise.
It should be understood that the term "and/or" as used herein is merely one type of association that describes an associated object, meaning that three relationships may exist, e.g., a and/or B may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" herein generally indicates that the former and latter related objects are in an "or" relationship.
It should be understood that although the terms first, second, third, etc. may be used to describe … … in embodiments of the present invention, these … … should not be limited to these terms. These terms are used only to distinguish … …. For example, the first … … can also be referred to as the second … … and similarly the second … … can also be referred to as the first … … without departing from the scope of embodiments of the present invention.
The words "if", as used herein, may be interpreted as "at … …" or "at … …" or "in response to a determination" or "in response to a detection", depending on the context. Similarly, the phrases "if determined" or "if detected (a stated condition or event)" may be interpreted as "when determined" or "in response to a determination" or "when detected (a stated condition or event)" or "in response to a detection (a stated condition or event)", depending on the context.
It is also noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a good or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such good or system. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a commodity or system that includes the element.
In addition, the sequence of steps in each method embodiment described below is only an example and is not strictly limited.
In the process of researching the invention, the inventor finds that the prior art has problems: for a dynamically changing feature set, an obvious scheme is to adopt model replacement, that is, after a period of time, new features are extracted and evaluated from the used model, so that the feature set is updated, and then a new recommended model is trained for the new feature set. This approach completely ignores the continuity of feature changes, completely abandons the previously used recommendation models, so that the recommendation performance becomes very poor at the time periods when the models alternate.
Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
Aiming at the problem of characteristic set evolution in a streaming data recommendation system, the invention considers the old recommendation model and the new recommendation model simultaneously when the recommendation models are alternated, thereby achieving smooth evolution of the recommendation models and avoiding sudden drop of recommendation performance during model switching.
As shown in fig. 1, in the present invention, model switching is extended from one point to a time period B, and in the model switching time period B, two new and old feature sets and a recommended model exist and function simultaneously. Through the weighting factor, the recommendation results obtained by the two models can be combined, so that the final recommendation result in the time period B is obtained. The recommendation system for the smooth evolution of the features and the models overcomes the problem of sudden reduction of recommendation performance caused by direct switching of the feature sets and the recommendation models in the original streaming data recommendation system, and simultaneously keeps the new and old feature sets and recommendation models in a period of switching time B, so that when the new model is used alone, a certain amount of training data is accumulated, and the smooth evolution and transition of the feature sets and the recommendation models are achieved.
In a recommendation system based on streaming data, in each learning process, an algorithm observes an instance and gives an expected value for the instance, compares the expected value with a tag value, and calculates loss from the difference, thereby improving the model. As shown in fig. 1, the data recording method is as follows:
in each round of training for the first period of time, T1, 21B and is an integer, data tag is
Figure BDA0001908434040000061
And is
Figure BDA0001908434040000062
Wherein R isd1Is S1Feature set space, d1 is S1Number of features in feature set, T1-B is the total number of training rounds;
in each round of training of the model switching transition period, T is T1-B,...,T1When, data is marked as
Figure BDA0001908434040000071
And
Figure BDA0001908434040000072
and is
Figure BDA0001908434040000073
Wherein R isd1Is S1Feature set space, d1 is S1Number of features in feature set, Rd2Is S2Feature set space, d2 is S2The number of features in the feature set, B is the total number of training rounds, and t is an integer;
in each round of training for the second period of time, T ═ T1,...,T2And is an integer, data tag is
Figure BDA0001908434040000074
And is
Figure BDA0001908434040000075
Wherein R isd2Is S2Feature set space, d2 is S2Number of features in feature set, T2-T1Is the total number of training rounds.
Definition < > denotes the inner product in Ω1∈Rd1Denotes S1The prediction model trained by the feature set is in omega2∈Rd2Denotes S2The prediction result is expressed as
Figure BDA0001908434040000076
Wherein wi,t∈Rdi，i＝1,2。
Loss function l (w)Tx, y) are calculated by the prediction result and the marking result, the loss function can be calculated by logistic or square loss, and the prediction model is updated by the loss function in each iteration, wherein the updating function is
Figure BDA0001908434040000077
Wherein i is 1, 2. The specific training process is shown in fig. 2.
In the training process of the basic flow data recommendation model, the association between new and old features before and after feature replacement is not considered, so that the old recommendation model is completely abandoned. However, in a real-world scenario, the characteristics of streaming data are slowly changing, and there is a necessary link between new and old characteristics, so this complete alternative wastes the old recommendation model that could otherwise be used to assist in the prediction of new data. According to the method, the mapping from the new features to the old features is established according to the relation between the new features and the old features, so that the new data can be expressed according to the new and old features, and the streaming data can be recommended better by using the new and old recommendation models.
Suppose that a mapping relationship ψ exists between two sets of new and old features: rd2-→Rd1Due to characteristic cuttingThe change time does not last for a long time, so that the relation between the new set of characteristics and the old set of characteristics is not suitable to be expressed by complex nonlinear mapping, and the patent uses a linear model to describe the mapping relation between the two sets of characteristic sets. During the feature overlap period B, assume that the correlation matrix between features is M and the duration is T1-B,...,T1Then M can be calculated by the least squares method:
Figure BDA0001908434040000081
wherein M isTA transpose matrix representing M;
to obtain the optimum value M of M, the following formula may be used:
Figure BDA0001908434040000082
with the correlation matrix, a new feature set S can be obtained2Of (3) to the old feature set S1For example, we are at S2In which a data is observed
Figure BDA00019084340400000812
An S can be recovered from the characteristic correlation matrix1Data under a feature set
Figure BDA0001908434040000083
Reapplication of S1Recommendation model under feature set
Figure BDA0001908434040000084
The recommended results under the old model can be obtained. Thus, referring to FIG. 1, the usage of the recommendation model is:
t∈0,1,...,T1b, apply feature set S1And a recommendation model
Figure BDA0001908434040000085
t∈T1-B,...,T1Applying feature set S1、S2And a recommendation model
Figure BDA0001908434040000086
At the same time, a feature set S is learned from data during a model common period1To S2The mapping relation M between the two;
t∈T1,...,T2applying feature set S2And a recommendation model
Figure BDA0001908434040000087
Meanwhile, according to the learned mapping relation M, the observation data is mapped to an old feature set, and an old recommendation model is utilized
Figure BDA0001908434040000088
And performing auxiliary calculation.
To be at T1,......,T2In the period, two old and new recommendation models are utilized for collaborative recommendation, a weight-based collaborative recommendation method is adopted in the patent, and the recommendation value at the time t is as follows: p is a radical oft＝α1,tf1,t+α2,tf2,tWherein p istIndicates the final recommended value, αi,tIs the ith round of predicted weight values,
Figure BDA0001908434040000089
representing the second feature set S2Is mapped to a first set of features S1In (1),
Figure BDA00019084340400000810
according to the loss function of each training round, the weight value can be updated according to the following formula:
Figure BDA00019084340400000811
where i is 1,2, η is an empirical parameter and l is a loss function.
From the above equation, it can be seen that if a model has a large loss function in the previous training round, its weight value will decrease exponentially in the next recommendation round. The collaborative recommendation algorithm training process is shown in fig. 3.
In the algorithm based on the new and old recommendation models, psi is initialized according to the prior eigenvector correlation matrix, and T is used1Pre-temporal recommendation model initialization
Figure BDA0001908434040000091
Initialization αi,t0.5. According to the previous recommendation model
Figure BDA0001908434040000092
To pair
Figure BDA0001908434040000093
And
Figure BDA0001908434040000094
initialization is performed. At T1+1 to T2From the labeled values of the sampled data, a loss function for each recommendation model may be calculated, from which a weight for each recommendation model may be calculated, where the empirical parameter is η ═ 8(ln2)/T2)1/2. Through the iteration of the steps, T can be finally obtained1+1 to T2And the recommendation models are coordinated in the time period, so that the recommendation model in the previous stage is utilized to assist the recommendation of the current time node, and the recommendation performance is improved.
The method introduces the collaborative feature recommendation model based on feature evolution, avoids the problem that the recommendation performance is sharply reduced during model switching in stream data recommendation by fully utilizing the old feature set and the recommendation model, and achieves the optimization of the recommendation performance of stream data recommendation. In order to verify the performance of the algorithm, a feature evolution collaborative recommendation model (FE) is compared with a general real-time recommendation algorithm ROGD, a data set is based on 9 comprehensive data sets and 8 ROT data sets, the total number of the data sets is 30, and in order to reduce the calculation cost of the recommendation algorithm, 10 optimal features are selected for training the recommendation model each time. The comparison of the recommendation accuracy is shown in fig. 4 (where the item with the highest recommendation accuracy is marked black). According to the comparison result, the collaborative feature recommendation model based on feature evolution can effectively improve the recommendation accuracy.
For a clearer explanation of this patent, an example is given. As shown in FIG. 1, the time division is 0 → T1-B、T1-B→T1、T1→T2Three stages, each stage feature set is S1、S1+S2、S2，S1And S2The corresponding recommendation model is w1And w2；
At 0 → T1During the time period B, the data are collected by the characteristics S1Expressed in a recommendation model w1Recommending;
at T1-B→T1During the time period, the data is collected by the characteristics S1And S2Expressed in a recommendation model w1、w2Performing collaborative recommendation, and simultaneously learning S according to data in the time period1To S2The mapping relation M;
at T1→T2During the time period, the data is collected by the characteristics S2Representing and mapping the data to a feature set S according to the mapping relation M in the step1Using the model w as well1And w2And carrying out collaborative recommendation.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.

Claims (2)

1. An Internet time series data recommendation method based on feature evolution is characterized in that:
training by adopting data in the first feature set in a first time period to obtain a first recommendation model and carrying out data recommendation; wherein, in each round of training of the first time period, T is 1,21B and is an integer, data tag is
Figure FDA0002542343280000011
And is
Figure FDA0002542343280000012
Wherein R isd1Is S1Feature set space, d1 is S1Number of features in feature set, T1-B is the total number of training rounds for the first time period; in omega1∈Rd1Denotes S1A prediction model trained by the feature set;
in a transition time period of model switching, performing weighted calculation by using a first recommendation model obtained by training data in a first feature set and a second recommendation model obtained by training data in a second feature set through a weighting factor to obtain final data recommendation in the transition time period; in each round of training of the model switching transition period, T is T1-B,...,T1When, data is marked as
Figure FDA0002542343280000013
And
Figure FDA0002542343280000014
and is
Figure FDA0002542343280000015
Wherein R isd1Is S1Feature set space, d1 is S1Number of features in feature set, Rd2Is S2Feature set space, d2 is S2The number of features in the feature set, B is the total number of training rounds in the transition time period, and t is an integer;
training by adopting data in the second feature set in a second time period to obtain a second recommendation model and carrying out data recommendation; in each round of training for the second period of time, T ═ T1,...,T2And is an integer, data tag is
Figure FDA0002542343280000016
And is
Figure FDA0002542343280000017
Wherein R isd2Is S2Feature set space, d2 is S2Number of features in feature set, T2-T1Is the total number of training rounds in the second time period; in omega2∈Rd2Denotes S2A prediction model trained by the feature set;
the mapping relation between the features in the first feature set and the second feature set in the model switching transition time period is described by adopting a linear model, and the correlation matrix between the features is assumed to be M, and the duration is T1-B,...,T1Then the least square method is adopted to calculate
Figure FDA0002542343280000018
Wherein M isTA transpose matrix representing M; the optimum value M of the matrix M,
Figure FDA0002542343280000021
in the transition time period of model switching, a final recommendation result is calculated by adopting a weight-based collaborative recommendation method, which specifically comprises the following steps:
pt＝α1,tf1,t+α2,tf2,twherein p istIndicates the final recommended value, αi,tIs the ith round of predicted weight values,
Figure FDA0002542343280000022
Figure FDA0002542343280000023
representing the second feature set S2Is mapped to a first set of features S1In (1),
Figure FDA0002542343280000024
the weighted value is updated by the following formula:
Figure FDA0002542343280000025
where i is 1,2, η is an empirical parameter and l is a loss function;
the prediction result is expressed as
Figure FDA0002542343280000026
Wherein wi,t∈RdiIs represented by RdiThe recommended model of (1), (2), and (i) represents an inner product.
2. The method of claim 1, wherein: in each learning and training process, an algorithm observes an example and provides a predicted value aiming at the example, the predicted value is compared with a label value, and a loss value l is calculated according to a difference value, so that the model is improved; loss function l (w)Tx, y) are calculated from the predicted and labeled results, while the prediction model is updated by a loss function at each iteration, wherein the update function is
Figure FDA0002542343280000027
Wherein i is 1,2, ytThe labeled values in the t-th round of training are shown.
CN201811542194.2A 2018-12-17 2018-12-17 Feature evolution-based streaming data recommendation method Active CN109508424B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811542194.2A CN109508424B (en) 2018-12-17 2018-12-17 Feature evolution-based streaming data recommendation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811542194.2A CN109508424B (en) 2018-12-17 2018-12-17 Feature evolution-based streaming data recommendation method

Publications (2)

Publication Number Publication Date
CN109508424A CN109508424A (en) 2019-03-22
CN109508424B true CN109508424B (en) 2020-09-08

Family

ID=65752731

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811542194.2A Active CN109508424B (en) 2018-12-17 2018-12-17 Feature evolution-based streaming data recommendation method

Country Status (1)

Country Link
CN (1) CN109508424B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113297481A (en) * 2021-05-08 2021-08-24 武汉卓尔数字传媒科技有限公司 Information pushing method, information pushing device and server based on streaming data processing
CN113821542B (en) * 2021-11-23 2022-02-11 四川新网银行股份有限公司 Automatic significant feature recommendation system and method

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108198041A (en) * 2018-01-23 2018-06-22 重庆裕嘉隆大数据产业有限公司 A kind of Method of Commodity Recommendation based on time attenuator, terminal and medium

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20040193938A1 (en) * 2003-03-28 2004-09-30 Shah Rasiklal Punjalal Complex system serviceability method and apparatus

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108198041A (en) * 2018-01-23 2018-06-22 重庆裕嘉隆大数据产业有限公司 A kind of Method of Commodity Recommendation based on time attenuator, terminal and medium

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
多模型预测控制的平滑切换;李少远 等;《上海交通大学学报》;19991130;第33卷(第11期);1345-1347 *

Also Published As

Publication number Publication date
CN109508424A (en) 2019-03-22

Similar Documents

Publication Publication Date Title
CN108875916B (en) Advertisement click rate prediction method based on GRU neural network
CN108875776B (en) Model training method and device, service recommendation method and device, and electronic device
CN112364976B (en) User preference prediction method based on session recommendation system
JP6990371B2 (en) Predictor, parameter set production method and program
CN110033097B (en) Method and device for determining association relation between user and article based on multiple data fields
CN103399858A (en) Socialization collaborative filtering recommendation method based on trust
CN114510652B (en) Social collaborative filtering recommendation method based on federal learning
CN111797321A (en) Personalized knowledge recommendation method and system for different scenes
US20220253722A1 (en) Recommendation system with adaptive thresholds for neighborhood selection
CN112395466A (en) Fraud node identification method based on graph embedded representation and recurrent neural network
Suzuki et al. Stacked denoising autoencoder-based deep collaborative filtering using the change of similarity
CN109508424B (en) Feature evolution-based streaming data recommendation method
CN113628059B (en) Associated user identification method and device based on multi-layer diagram attention network
CN111046188A (en) User preference degree determining method and device, electronic equipment and readable storage medium
CN113010683B (en) Entity relationship identification method and system based on improved graph attention network
CN110781401A (en) Top-n project recommendation method based on collaborative autoregressive flow
CN114708479B (en) Self-adaptive defense method based on graph structure and characteristics
CN107240100B (en) Image segmentation method and system based on genetic algorithm
CN113610610B (en) Session recommendation method and system based on graph neural network and comment similarity
CN103942298A (en) Recommendation method and system based on linear regression
CN116975686A (en) Method for training student model, behavior prediction method and device
CN112507245A (en) Social network friend recommendation method based on graph neural network
CN105654345A (en) Bayesian network-based advertisement click-through rate prediction method and device
CN116541592A (en) Vector generation method, information recommendation method, device, equipment and medium
CN110544124A (en) waste mobile phone pricing method based on fuzzy neural network

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