CN116308651A - Financial product recommendation method and device based on time sequence and computer equipment - Google Patents

Financial product recommendation method and device based on time sequence and computer equipment Download PDF

Info

Publication number
CN116308651A
CN116308651A CN202310240150.9A CN202310240150A CN116308651A CN 116308651 A CN116308651 A CN 116308651A CN 202310240150 A CN202310240150 A CN 202310240150A CN 116308651 A CN116308651 A CN 116308651A
Authority
CN
China
Prior art keywords
user
financial product
financial
sequence
information
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.)
Pending
Application number
CN202310240150.9A
Other languages
Chinese (zh)
Inventor
肖俊
唐超
肖晓丽
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.)
Changsha University of Science and Technology
Original Assignee
Changsha University of Science and Technology
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 Changsha University of Science and Technology filed Critical Changsha University of Science and Technology
Priority to CN202310240150.9A priority Critical patent/CN116308651A/en
Publication of CN116308651A publication Critical patent/CN116308651A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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—Item recommendations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2457—Query processing with adaptation to user needs
    • G06F16/24578—Query processing with adaptation to user needs using ranking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/06—Asset management; Financial planning or analysis
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Development Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Computational Linguistics (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Technology Law (AREA)
  • Biophysics (AREA)
  • Human Resources & Organizations (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Operations Research (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The application relates to a financial product recommendation method based on time sequence. The method comprises the following steps: and constructing a behavior undirected graph according to the financial product sequence selected in the current session process of the user. And extracting the characteristic information of the financial products in the behavior undirected graph through the GNN, introducing an attention mechanism to process the characteristic information to obtain weighted characteristic information of each financial product in the financial product sequence, and generating characteristic representation of the financial product sequence according to the weighted characteristic information. Inputting the characteristic representation, acquiring user dependency information in the characteristic representation by adopting a weighted network, forming the current session interest of the user according to the user dependency information, and accumulating the user session times to obtain the user characterization. And obtaining the recommendation score of the financial product to be recommended through the user characterization and the inner product of the feature representation of the financial product sequence to be recommended, and recommending the financial product to the client according to the sequencing of the recommendation score. The method effectively combines the session time characteristics with the financial product characteristics, thereby improving the accuracy of financial product recommendation.

Description

Financial product recommendation method and device based on time sequence and computer equipment
Technical Field
The present disclosure relates to the field of recommendation technologies, and in particular, to a method, an apparatus, and a device for recommending a financial product based on time sequence.
Background
At present, in the age of rapid development of the internet, more and more financial and investment transactions are carried out, users purchase financial products through a mobile phone terminal APP and a webpage of a financial agency, when the users browse the webpage, in the recommendation modes such as traditional collaborative filtering algorithm, partial machine learning and the like, for example, BPRM technology can sort the products interested by each user according to preference, but in the financial field, the preference of the users for the products changes along with time change along with the reduction of the aging rate, for example, deep CF technology can acquire the preference of the users for the financial products, but cannot deal with the time sequence problem, and the interests of the users cannot be well captured and changed along with time change, so that the prior art cannot well meet the recommendation of the users for the products in the financial field, and the recommendation effect deviation is larger and larger after the financial products are accumulated along with time.
Disclosure of Invention
Accordingly, in view of the above-mentioned problems, it is necessary to provide a method, an apparatus and a device for recommending financial products based on time sequence, which can improve accuracy of recommending financial products.
A method of financial product recommendation based on time sequence, the method comprising:
and constructing a behavior undirected graph according to the financial product sequence selected in the current session process of the user. Nodes in the behavioural undirected graph are vector representations of financial products.
And extracting the characteristic information of the financial products in the behavior undirected graph through the graph neural network, introducing an attention mechanism to process the characteristic information to obtain weighted characteristic information of each financial product in the financial product sequence, and generating characteristic representation of the financial product sequence according to the weighted characteristic information.
And taking the feature representation as input data, acquiring user dependency information in the feature representation by adopting a weighted network, forming the current session interest of the user according to the user dependency information, and accumulating the user session times to obtain the final user representation.
And obtaining the final recommendation score of the financial product to be recommended through the user characterization and the inner product of the feature representation of the financial product sequence to be recommended, and recommending the financial product to the client according to the ordering of the final recommendation score.
In one embodiment, the method further comprises: generating unordered financial product sequences according to financial products selected in the current session process of a user, forming an adjacent matrix of the financial product sequences through adjacent relations between adjacent financial products in the financial product sequences, and splicing the adjacent matrix and the financial products of the financial product sequences to generate a behavior undirected graph.
In one embodiment, the method further comprises: and taking the current financial product as a node in the behavior undirected graph, and forming a node vector in the behavior undirected graph with the current financial product according to the adjacent relation between the next financial product and the current financial product.
In one embodiment, the method further comprises: extracting node information of the behavior undirected graph through the graph neural network, extracting feature information of financial products in the node information, introducing an attention mechanism, and obtaining weighted feature information of each financial product by giving attention coefficients to the feature information of each financial product and performing weighting processing. And updating the characteristic information of the financial product sequence according to the set of the weighted characteristic information of the financial products to obtain the characteristic representation of the financial product sequence.
In one embodiment, the method further comprises: and extracting the time characteristics of the current session of the user by using a sine and cosine method according to the current session time of the user, and fusing the time characteristics with the weighted characteristic information of the financial product to generate a set of weighted characteristic information of the financial product.
In one embodiment, the method further comprises: inputting characteristic representation, acquiring the characteristic representation by adopting an average pooling method, constructing a weighted network model, extracting the global interest of the current session of the user, storing the user dependent information of the financial products in the financial product sequence, collecting the user dependent information, and generating the current session interest of the user. And accumulating the current session interests of the user according to the user session times, generating the global interests of the user session, and obtaining the final user characterization.
In one embodiment, the method further comprises: and normalizing the inner product result by the user characterization and the inner product represented by the characteristics of the financial product sequence to be recommended by adopting a softmax activation function, and endowing the financial product in the financial product sequence to be recommended with recommendation scores. And sorting the financial products in the sequence of financial products to be recommended according to the recommendation scores from high to low, and recommending the financial products with high recommendation score sorting to the first position of the current session interface of the user.
In one embodiment, the method further comprises: and acquiring a financial product selected in the next session process of the user as a true value, taking the financial product at the first position as a predicted value, and correcting the recommendation score of the financial product of the next session of the user according to the cross entropy of the true value and the predicted value as a loss function.
A time-sequential-based financial product recommendation device, the device comprising:
and the model construction module is used for constructing a behavior undirected graph according to the financial product sequence selected in the current session process of the user. Nodes in the behavior undirected graph are vector representations of financial products;
the feature representation generating module is used for extracting feature information of the financial products in the behavior undirected graph through the graph neural network, introducing an attention mechanism to process the feature information to obtain weighted feature information of each financial product in the financial product sequence, and generating feature representation of the financial product sequence according to the weighted feature information.
The user characterization generation module is used for taking the feature representation as input data, acquiring user dependence information in the feature representation by the weighted network, forming the current session interest of the user according to the user dependence information, accumulating the user session times and obtaining a final user characterization;
and the product recommending module is used for obtaining the final recommending score of the financial product to be recommended through the user characterization and the inner product of the characteristic representation of the financial product sequence to be recommended, and recommending the financial product to the customer according to the ordering of the final recommending score.
A computer device comprising a memory storing a computer program and a processor which when executing the computer program performs the steps of:
and constructing a behavior undirected graph according to the financial product sequence selected in the current session process of the user. Nodes in the behavioural undirected graph are vector representations of financial products.
And extracting the characteristic information of the financial products in the behavior undirected graph through the graph neural network, introducing an attention mechanism to process the characteristic information to obtain weighted characteristic information of each financial product in the financial product sequence, and generating characteristic representation of the financial product sequence according to the weighted characteristic information.
And taking the feature representation as input data, acquiring user dependency information in the feature representation by adopting a weighted network, forming the current session interest of the user according to the user dependency information, and accumulating the user session times to obtain the final user representation.
And obtaining the final recommendation score of the financial products to be recommended in the financial product sequence to be recommended through the user characterization and the inner product of the feature representation of the financial product sequence to be recommended, and recommending the financial products to the client according to the ordering of the final recommendation score.
Compared with the prior art, the financial product recommendation method, the financial product recommendation device and the computer equipment based on the time sequence have the following technical effects:
the method comprises the steps of establishing a behavior undirected graph model to store financial products and corresponding characteristic information thereof, carrying out iterative processing on the behavior undirected graph data by adopting a graph neural network, an attention mechanism and a weighted network model framework, fusing the characteristic information of the financial products with product time information corresponding to the financial product yield, obtaining the corresponding relation between the financial product yield and time, carrying out average pooling, and updating the characteristic representation of the obtained financial product sequence in real time according to the session time of the user, so that the user dependent information is positively correlated with the financial product yield, and further obtaining the final recommendation score of the financial products to be recommended in the financial product sequence to be recommended by the inner product of the user characterization and the characteristic representation of the financial product sequence to be recommended.
Drawings
FIG. 1 is a flow chart of a method for recommending financial products based on time sequence in one embodiment;
FIG. 2 is a schematic diagram of an embodiment of a behavioral undirected graph;
FIG. 3 is a schematic diagram of an adjacency matrix in one embodiment;
FIG. 4 is a flow diagram of a weighting network in one embodiment;
FIG. 5 is a block diagram of a timing-based financial product recommendation device in one embodiment;
fig. 6 is an internal structural diagram of a computer device in one embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
The financial product recommendation method based on time sequence provided by the application, as shown in fig. 1, comprises the following steps:
and 102, constructing a behavior undirected graph according to the financial product sequence selected in the current session process of the user. Nodes in the behavioural undirected graph are vector representations of financial products.
The current session of the user can be browsing financial products through a mobile phone APP or browsing financial products through a computer webpage, and according to the financial products clicked by the user within a period of time, the current session of the user is used as a financial product sequence selected by the current session of the user, and the selected financial products and the corresponding adjacent relations thereof are stored by adopting a behavior undirected graph because the sequence of the financial products selected by the user is irregular.
And 104, extracting feature information of the financial products in the behavior undirected graph through the graph neural network, and processing the feature information by introducing an attention mechanism to obtain weighted feature information of each financial product in the financial product sequence, and generating feature representation of the financial product sequence according to the weighted feature information.
The method comprises the steps of taking a behavior undirected graph as a node of a graph neural network, extracting characteristic information of financial products in the behavior undirected graph, giving attention coefficients to the node of each graph neural network through an attention introducing mechanism, obtaining weighted characteristic information of the financial products, collecting the weighted characteristic information of all the financial products to obtain weighted characteristic information of a financial product sequence, updating the characteristic information of the whole financial product sequence by using a gate-controlled graph neural network, and generating characteristic representation of the financial product sequence through learning of the graph neural network.
And 106, taking the feature representation as input data, acquiring user dependency information in the feature representation by adopting a weighted network, forming the current session interest of the user according to the user dependency information, and accumulating the user session times to obtain the final user representation.
Based on the property that the yield of the financial products changes along with the change of time, the user selects the behaviors of investment and financial products, multiple user sessions can exist on the same day, the feature expression vector of the financial product sequence corresponding to each session is obtained in an average pooling mode, the feature expression vector is used as a feature expression collection of the whole behavior undirected graph in one session, further, the time feature corresponding to each session is extracted, the day and the hour of the time feature are extracted respectively by adopting a positive selection method and a cosine method, the feature expression collection and the day and the hour corresponding to the feature expression collection are fused, and further, the feature expression of the financial product sequence corresponding to the multiple sessions of the user is obtained.
Specifically, feature representation data of the financial product sequence are input, a weighting network is adopted to model the global interest of a user, the feature representation of the financial product sequence of each session is taken as a node of the weighting network, a gating unit is added to control circulation of feature representation information among neurons, and user dependence information corresponding to the financial product in the feature representation is extracted and stored. The feature of each financial product sequence represents the local interest of the user corresponding to one session, the corresponding feature is represented as a feature representation set of a plurality of financial product sequences according to the number of times of the user session, and then the feature representation set is processed through a weighted network to obtain a set of user-dependent information, and the set of user-dependent information is used as the global interest of the user for a plurality of sessions to obtain the final user representation of the plurality of sessions.
And step 108, obtaining the final recommendation score of the financial products to be recommended in the financial product sequence to be recommended through the user characterization and the inner product of the feature representation of the financial product sequence to be recommended, and recommending the financial products to the clients according to the ordering of the final recommendation score.
And taking all financial products in a financial product library in financial software or an investment agency officer network as financial products to be recommended, updating the financial product library along with financial strategies of a dealer and user demands, wherein a characteristic information set of the financial products to be recommended is taken as characteristic representation of a financial product sequence to be recommended, carrying out inner product calculation on a characteristic representation of a user corresponding to past user selection and the characteristic representation of the recommended financial product sequence based on the selection of multiple sessions of past users, carrying out normalization processing on calculation results by using a softmax activation function, further obtaining recommendation scores of each financial product, recommending the financial products with high recommendation scores to customers as a first order of a product sequence displayed by the page when the user browses the financial product recommendation page next time through a mobile phone terminal APP or a computer terminal webpage, sequentially arranging the financial products down along the page according to the recommendation scores, adding a loss function on the basis, and recording cross entropy of the financial products selected by the current session of the user and a predicted value of the recommended financial products, thereby correcting the recommendation scores.
Compared with the prior art, the financial product recommendation method based on the time sequence has the following technical effects:
the method comprises the steps of establishing a behavior undirected graph model to store financial products and corresponding characteristic information thereof, carrying out iterative processing on the behavior undirected graph data by adopting a graph neural network, an attention mechanism and a weighted network model framework, fusing the characteristic information of the financial products with product time information corresponding to the financial product yield, obtaining the corresponding relation between the financial product yield and time, carrying out average pooling, and updating the characteristic representation of the obtained financial product sequence in real time according to the session time of the user, so that the user dependent information is positively correlated with the financial product yield, and further obtaining the final recommendation score of the financial products to be recommended in the financial product sequence to be recommended by the inner product of the user characterization and the characteristic representation of the financial product sequence to be recommended.
In one embodiment, a disordered financial product sequence is generated according to financial products selected in the current session process of a user, an adjacent matrix of the financial product sequence is formed through an adjacent relation between adjacent financial products in the financial product sequence, and the adjacent matrix is spliced with the financial products of the financial product sequence to generate a behavior undirected graph.
In one embodiment, the current financial product is used as a node in the behavior undirected graph, and a node vector in the behavior undirected graph is formed with the current financial product according to the adjacent relation between the next financial product and the current financial product.
It should be noted that, as shown in fig. 2, the financial product sequence clicked by the user is s= [ v ] 1 ,v 2 ,v 3 ,v 4 ]The financial products in the series are ordered in time sequence, i.e. the user browses v in turn 1 ,v 2 ,v 3 ,v 4 The model of the behavior undirected graph is G S ＝(V S ,E S ) Wherein V is S ＝{v 1 ,v 2 ,v 3 ,v 4 }，E S ＝{(v 1 ,v 2 ),(v 2 ,v 1 ),(v 2 ,v 3 ),(v 3 ,v 2 ),(v 3 ,v 4 ),(v 4 ,v 3 ) The node in the figure is the collection of the financial products in each session, the edges in the figure represent the adjacent relation of the two financial products in the financial product sequence, and then it can be obtained that, as shown in fig. 3, the behavior undirected graph has an adjacent matrix, in the adjacent matrix, 0 represents the uncorrelation among the nodes, and 1 represents the correlation among the nodes. Therefore, the financial product selection sequence is not positively related to the preference of the user, so the financial product selection sequence is modeled into a behavior undirected graph to represent, and analysis of the characteristic information of the financial product is focused.
In one embodiment, node information of an undirected graph is extracted through a graph neural network, feature information of financial products in the node information is extracted, an attention mechanism is introduced, and weighted feature information of each financial product is obtained by giving attention coefficients to the feature information of each financial product and performing weighted processing. And updating the characteristic information of the financial product sequence according to the set of the weighted characteristic information of the financial products to obtain the characteristic representation of the financial product sequence.
It should be noted that the feature information of the financial product includes: the name of the financial product, the annual yield, the product time corresponding to the yield and the adjacency relation between adjacent financial products in the behavior undirected graph are related locally, for example, after a user browses low-risk products, the user can purchase the low-risk products with high probability, so that the characteristic information of the financial products corresponding to each node is extracted from the behavior undirected graph data through the method and the attention mechanism of the graph neural network.
In particular, with financial products v i For example, first, initialized characteristic information x is generated for each product i The formula is:
x i ＝Embedding(v i )
then calculate the financial product v i Attention coefficient alpha of each adjacent financial product of (a) ij The formula is:
Figure SMS_1
weighting the characteristic information of the adjacent financial products according to the attention coefficient obtained by the calculation result to obtain the characteristic information of each adjacent financial product after weighting, wherein the formula is as follows:
Figure SMS_2
wherein N is i Represented as financial product v i Weighted feature information sets for adjacent financial products.
Further, a gate mechanism in the gate control graph neural network is used for obtaining an updated gate control signal z i And reset gating signal r i The formula is:
Figure SMS_3
Figure SMS_4
by means of a gating signal r i For v i Weighted feature information of neighbor products and v i Is the self-characteristic information x of (a) i To make selective forget and memory so as to obtain financial product v i Final weighted feature information x' i The formula is:
Figure SMS_5
Figure SMS_6
therefore, the local relation among the financial products can be captured by using the method of the graph neural network, meanwhile, noise financial products generated due to misoperation of a user and the like are considered in the financial product sequence, the noise financial products are filtered by using the method of increasing the attention mechanism, the influence of characteristic information of the noise financial products is eliminated based on the noise financial products, the characteristic representation accuracy of the final financial product sequence is higher, and the extracted financial products are closer to the actual interest of the user in the financial products.
In one embodiment, according to the current session time of the user, the time characteristics of the current session of the user are extracted by using sine and cosine methods, and the time characteristics and the weighted characteristic information of the financial products are fused to generate a set of weighted characteristic information of the financial products.
It is worth to say that, each time the user browses a financial product recommendation page as a session, each session is generated at a certain time, and the profitability feature in the financial product feature information is positively correlated with the time, so that a sine and cosine method is adopted to provide time features, and the time features are extracted in a manner of sine of days and cosine of hours respectively, and the formula is expressed as follows:
D＝sin(2*Π*T/365.0)
M＝cos(2*Π*T/24.0)
wherein D represents days, M represents hours, the above obtained time features are connected with the feature representation of the financial product sequence in one session, and then the time features are mapped into a set vector B of weighted feature information of the financial product through a full connection layer, and the set vector B is used as the feature representation of the financial product sequence in one session, and the formula is as follows:
B＝(D+M+b)E
therefore, the feature information of the financial product is fused with the current corresponding user session time information, so that when the income ratio of the financial product fluctuates along with time, the latest income ratio can be matched with the financial product, and the feature information of the whole financial product sequence is updated.
In one embodiment, a feature representation is input, the feature representation is acquired by adopting an average pooling method, a weighted network model is constructed, the global interest of the current session of the user is extracted, the user dependent information of the financial products in the financial product sequence is stored, the user dependent information is collected, and the current session interest of the user is generated. And accumulating the current session interests of the user according to the user session times, generating the global interests of the user session, and obtaining the final user characterization.
It is worth noting that the weighting characteristic information x 'of the learned financial products is utilized' i And obtaining the characteristic representation of a single financial product sequence, and further obtaining the characteristic representations of a plurality of financial product sequences in an accumulated manner according to the total number of user sessions. The financial product sequence browsed by the user is B i Wherein comprises |B i The weighted characteristic information corresponding to the financial products is x j ，(j＝1,2,3……，|B i I), obtaining a characteristic representation b of the financial product sequence of each session by adopting an average pooling method, wherein the formula is as follows:
Figure SMS_7
through the above averaging and pooling operation, a feature vector B is obtained in each session, where the feature vector represents the feature set of the whole behavior undirected graph in one session, and further, by introducing a weighted network, a gating unit is added to control the flow of information between neurons, so that the dependency information of the user on the financial product sequence can be saved, as shown in fig. 4, first, a feature representation b= { B of the whole financial product sequence corresponding to the given total number t of user sessions is input 1 ,B 2 ,......B t After the weighted network processing, the final user representation H is obtained, and the formula is expressed as follows:
QKV＝WB t
Figure SMS_8
therefore, through the combination of the average pooling method and the weighted network architecture, the long-time browsing behavior of the user and the preference of the financial products can be recorded and reserved, the financial product sequence selected by the user at a certain time or a certain time point is not limited, the characteristic information of the financial products favored by the user can be more comprehensively tracked and analyzed for a long time, and meanwhile, the recommendation of the financial products is more and more accurate along with the change of the preference of the user along with the accumulation of time.
In one embodiment, the inner product result is normalized by the user characterization and the inner product represented by the features of the sequence of financial products to be recommended, using a softmax activation function, and a recommendation score is assigned to the financial products in the sequence of financial products to be recommended. And sorting the financial products in the sequence of financial products to be recommended according to the recommendation scores from high to low, and recommending the financial products with high recommendation score sorting to the first position of the current session interface of the user.
It is worth noting that, based on the final user characterization obtained above, a push score is calculated for each commodity to be recommended, the commodity to be recommended is ordered according to the recommendation score, the commodity ordered before is recommended to the user, specifically, the user characterization and the feature representation of the financial product sequence to be recommended are subjected to inner product calculation, and the initial recommendation score Z of each financial product to be recommended is obtained, wherein the formula is:
Z＝HX
wherein, X is the characteristic information of the financial products to be recommended, H is the final user characterization, the initial recommendation score Z is normalized by a softmax activation function to obtain the final recommendation score Y of each financial product to be recommended, and the formula is expressed as follows:
Y＝softmax(Z)
in the recommending process, the financial products to be recommended are ranked from high to low according to the final recommending score Y, then the financial products with highest ranking are displayed at the first position of a browsing page of a user, and the like, according to the recommending score, the information of the products to be recommended is arranged from top to bottom in the browsing page, and the arrangement result is recommended to the user.
In one embodiment, the financial product selected in the next session of the user is obtained as a true value, the financial product at the first position is used as a predicted value, and the recommendation score of the financial product in the next session of the user is corrected according to the cross entropy of the true value and the predicted value as a loss function.
It is worth noting that the formula for the loss function is:
Figure SMS_9
wherein, the liquid crystal display device comprises a liquid crystal display device,
Figure SMS_10
for the predicted value of the financial product recommended to the user, y i And browsing the page generated by the predicted value for the user, and selecting the true value of the financial product.
It should be understood that, although the steps in the flowcharts of fig. 1-4 are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in fig. 1-4 may include multiple sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, nor do the order in which the sub-steps or stages are performed necessarily occur sequentially, but may be performed alternately or alternately with at least a portion of the sub-steps or stages of other steps or steps.
In one embodiment, as shown in fig. 5, there is provided a financial product recommendation device based on time sequence, including: the system comprises a model construction module, a characteristic representation generation module, a user representation generation module and a product recommendation module, wherein:
and the model construction module is used for constructing a behavior undirected graph according to the financial product sequence selected in the current session process of the user. Nodes in the behavior undirected graph are vector representations of financial products;
the feature representation generating module is used for extracting feature information of the financial products in the behavior undirected graph through the graph neural network, introducing an attention mechanism to process the feature information to obtain weighted feature information of each financial product in the financial product sequence, and generating feature representation of the financial product sequence according to the weighted feature information.
The user characterization generation module is used for taking the characteristic representation as input data, acquiring user dependence information in the characteristic representation by adopting a weighted network, forming the current session interest of the user according to the user dependence information, accumulating the user session times and obtaining a final user characterization;
the product recommending module is used for obtaining the final recommending score of the financial product to be recommended in the financial product sequence to be recommended through the user characterization and the inner product of the feature representation of the financial product sequence to be recommended, and recommending the financial product to the customer according to the ordering of the final recommending score.
In one embodiment, a disordered series of financial products is generated according to the financial products selected in the current session process of the user, an adjacency matrix of the series of financial products is formed through adjacency relations between adjacent financial products in the series of financial products, and the adjacency matrix is spliced with the financial products of the series of financial products to generate the behavioural undirected graph.
In one embodiment, the node vector in the behavior undirected graph is formed with the current financial product according to the adjacent relation between the next financial product and the current financial product by taking the current financial product as the node in the behavior undirected graph.
In one embodiment, node information of the behavior undirected graph is extracted through a graph neural network, feature information of financial products in the node information and relations among the node information are extracted, an attention mechanism is introduced, and weighted feature information of each financial product is obtained by giving attention coefficients to the feature information of each financial product and performing weighted processing. And updating the characteristic information of the financial product sequence according to the set of the weighted characteristic information of the financial products to obtain the characteristic representation of the financial product sequence.
In one embodiment, the time characteristics of the current session of the user are extracted by using sine and cosine methods according to the current session time of the user, and the time characteristics and the weighted characteristic information of the financial product are fused to generate a set of weighted characteristic information of the financial product.
In one embodiment, a feature representation is obtained by inputting the feature representation and adopting an average pooling method, a weighted network model is constructed, the global interest of the current session of the user is extracted, the user dependent information of the financial products in the financial product sequence is stored, the user dependent information is collected, and the current session interest of the user is generated. And accumulating the current session interests of the user according to the user session times, generating the global interests of the user session, and obtaining the final user characterization.
In one embodiment, the recommendation score is assigned to the financial products in the sequence of financial products to be recommended by normalizing the inner product results by the user characterization and the inner product represented by the characteristics of the sequence of financial products to be recommended using a softmax activation function. And sorting the financial products in the sequence of financial products to be recommended according to the recommendation scores from high to low, and recommending the financial products with high recommendation score sorting to the first position of the current session interface of the user.
In one embodiment, the recommendation score of the financial product of the next session of the user is corrected by acquiring the financial product selected by the user in the next session as a true value, the financial product of the first location as a predicted value, and the cross entropy of the true value and the predicted value as a loss function.
For specific limitations on the time-series-based financial product recommendation device, reference may be made to the above limitations on the time-series-based financial product recommendation method, and the detailed description thereof will be omitted. The above-described respective modules in the time-series-based financial product recommendation apparatus may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
It will be appreciated by those skilled in the art that the structure shown in fig. 5 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided, which may be a terminal, and the internal structure of which may be as shown in fig. 6. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program, when executed by a processor, implements a time-series-based financial product recommendation method. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on the shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.
In one embodiment, a computer device is provided comprising a memory storing a computer program and a processor that when executing the computer program performs the steps of:
and constructing a behavior undirected graph according to the financial product sequence selected in the current session process of the user. Nodes in the behavioural undirected graph are vector representations of financial products.
Feature information and relations among the financial products in the behavior undirected graph are extracted through the graph neural network, a attention mechanism is introduced to process the feature information, weighted feature information of each financial product in the financial product sequence is obtained, and feature representation of the financial product sequence is generated according to the weighted feature information.
And taking the feature representation as input data, acquiring user dependency information in the feature representation by adopting a weighted network, forming the current session interest of the user according to the user dependency information, and accumulating the user session times to obtain the final user representation.
And obtaining the final recommendation score of the financial products to be recommended in the financial product sequence to be recommended through the user characterization and the inner product of the feature representation of the financial product sequence to be recommended, and recommending the financial products to the client according to the ordering of the final recommendation score.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the various embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples merely represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the invention. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application is to be determined by the claims appended hereto.

Claims (10)

1. A financial product recommendation method, a financial product recommendation device and computer equipment based on time sequence are characterized in that the method comprises the following steps:
constructing a behavior undirected graph according to a financial product sequence selected in the current session process of a user; the nodes in the behavior undirected graph are vector representations of financial products;
extracting feature information of the financial products in the behavior undirected graph through a graph neural network, and processing the feature information by an attention introducing mechanism to obtain weighted feature information of each financial product in the financial product sequence, and generating feature representation of the financial product sequence according to the weighted feature information;
the characteristic representation is used as input data, user dependence information in the characteristic representation is obtained by adopting a weighting network, the current session interests of the user are formed according to the user dependence information, the number of times of user session is accumulated, and the final user representation is obtained;
and obtaining the final recommendation score of the financial products to be recommended in the financial product sequence to be recommended through the user characterization and the inner product of the feature representation of the financial product sequence to be recommended, and recommending the financial products to the clients according to the ordering of the final recommendation score.
2. The method of claim 1, wherein constructing the behavioral undirected graph from the sequence of financial products selected during the user's current session comprises:
generating unordered financial product sequences according to financial products selected in the current session process of a user, forming an adjacent matrix of the financial product sequences through adjacent relations between adjacent financial products in the financial product sequences, and splicing the adjacent matrix and the financial products of the financial product sequences to generate a behavior undirected graph.
3. The method of claim 2, wherein the nodes in the behavioral undirected graph are vector representations of financial products, comprising:
and taking the current financial product as a node in the behavior undirected graph, and forming a node vector in the behavior undirected graph with the current financial product according to the adjacent relation between the next financial product and the current financial product.
4. The method of claim 3, wherein the characteristic information of the financial product comprises: product name, product yield, and product adjacency;
extracting feature information of the financial products in the behavior undirected graph through a graph neural network, and processing the feature information by an attention introducing mechanism to obtain weighted feature information of each financial product in the financial product sequence, and generating feature representation of the financial product sequence according to the weighted feature information, wherein the feature representation comprises the following steps:
extracting node information of the behavior undirected graph through a graph neural network, extracting feature information of financial products in the node information, introducing an attention mechanism, and obtaining weighted feature information of each financial product by giving attention coefficients to the feature information of each financial product and performing weighting treatment;
and updating the characteristic information of the financial product sequence according to the set of the weighted characteristic information of the financial product to obtain the characteristic representation of the financial product sequence.
5. The method of claim 4, comprising, prior to the step of updating the characteristic information of the sequence of financial products based on the set of weighted characteristic information of the financial products, obtaining a characteristic representation of the sequence of financial products:
and extracting the time characteristics of the current session of the user by using a sine and cosine method according to the current session time of the user, and fusing the time characteristics with the weighted characteristic information of the financial product to generate a set of the weighted characteristic information of the financial product.
6. The method of claim 5, wherein inputting the feature representation, obtaining user-dependent information in the feature representation using a weighted network, composing a user's current session interests based on the user-dependent information, and accumulating the number of user sessions to obtain a final user representation, comprising:
inputting the characteristic representation, acquiring the characteristic representation by adopting an average pooling method, constructing a weighted network model, extracting the global interest of the current session of the user, storing the user dependent information of the financial products in the financial product sequence, and collecting the user dependent information to generate the current session interest of the user;
and accumulating the current session interests of the user according to the times of the user session, generating the global interests of the user session, and obtaining the final user characterization.
7. The method of claim 6, wherein obtaining a final recommendation score for the financial product to be recommended in the sequence of financial products to be recommended by the user characterization and an inner product of the characterization of the sequence of financial products to be recommended, and recommending the financial product to the customer according to the ranking of the final recommendation score, comprises:
normalizing the inner product result by the user characterization and the inner product represented by the characteristics of the financial product sequence to be recommended by adopting a softmax activation function, and endowing the financial product in the financial product sequence to be recommended with recommendation scores;
and sorting the financial products in the sequence of the financial products to be recommended according to the recommendation score from high to low, and recommending the financial products with high recommendation score sorting to the first position of the current session interface of the user.
8. The method of claim 7, further comprising, after the step of ranking the financial products in the sequence of financial products to be recommended from high to low according to the recommendation score, recommending a financial product with a high recommendation score to the first location of the user's next session interface:
and acquiring a financial product selected in the next session process of the user as a true value, taking the financial product at the first position as a predicted value, and correcting the recommendation score of the financial product of the next session of the user according to the cross entropy of the true value and the predicted value as a loss function.
9. A time-sequential-based financial product recommendation device, the device comprising:
the model construction module is used for constructing a behavior undirected graph according to the financial product sequence selected in the current session process of the user; the nodes in the behavior undirected graph are vector representations of financial products;
the characteristic representation generation module is used for extracting characteristic information of the financial products in the behavior undirected graph through the graph neural network, introducing an attention mechanism to process the characteristic information to obtain weighted characteristic information of each financial product in the financial product sequence, and generating characteristic representation of the financial product sequence according to the weighted characteristic information;
the user characterization generation module is used for taking the characteristic representation as input data, acquiring user dependence information in the characteristic representation by adopting a weighting network, forming the current session interest of the user according to the user dependence information, accumulating the number of times of user session and obtaining a final user characterization;
and the product recommending module is used for obtaining the final recommending score of the financial product to be recommended in the financial product sequence to be recommended through the user characterization and the inner product represented by the characteristics of the financial product sequence to be recommended, and recommending the financial product to the customer according to the ordering of the final recommending score.
10. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 8 when the computer program is executed.
CN202310240150.9A 2023-03-09 2023-03-09 Financial product recommendation method and device based on time sequence and computer equipment Pending CN116308651A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310240150.9A CN116308651A (en) 2023-03-09 2023-03-09 Financial product recommendation method and device based on time sequence and computer equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310240150.9A CN116308651A (en) 2023-03-09 2023-03-09 Financial product recommendation method and device based on time sequence and computer equipment

Publications (1)

Publication Number Publication Date
CN116308651A true CN116308651A (en) 2023-06-23

Family

ID=86837449

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310240150.9A Pending CN116308651A (en) 2023-03-09 2023-03-09 Financial product recommendation method and device based on time sequence and computer equipment

Country Status (1)

Country Link
CN (1) CN116308651A (en)

Similar Documents

Publication Publication Date Title
CN110717098B (en) Meta-path-based context-aware user modeling method and sequence recommendation method
Lima et al. Domain knowledge integration in data mining using decision tables: case studies in churn prediction
CN111291264B (en) Access object prediction method and device based on machine learning and computer equipment
US8412665B2 (en) Action prediction and identification temporal user behavior
CN109582876B (en) Tourist industry user portrait construction method and device and computer equipment
CN109783730A (en) Products Show method, apparatus, computer equipment and storage medium
CN114372573B (en) User portrait information recognition method and device, computer equipment and storage medium
CN111815415A (en) Commodity recommendation method, system and equipment
CN110633421B (en) Feature extraction, recommendation, and prediction methods, devices, media, and apparatuses
CN110008397B (en) Recommendation model training method and device
CN112905876B (en) Information pushing method and device based on deep learning and computer equipment
CN112035611B (en) Target user recommendation method, device, computer equipment and storage medium
CN112418978B (en) Product recommendation method, device, equipment and medium
CN113873330B (en) Video recommendation method and device, computer equipment and storage medium
CN112288279A (en) Business risk assessment method and device based on natural language processing and linear regression
CN114371946B (en) Information push method and information push server based on cloud computing and big data
CN115630221A (en) Terminal application interface display data processing method and device and computer equipment
CN114693409A (en) Product matching method, device, computer equipment, storage medium and program product
CN110807693A (en) Album recommendation method, device, equipment and storage medium
CN112685639A (en) Activity recommendation method and device, computer equipment and storage medium
CN110209944B (en) Stock analyst recommendation method and device, computer equipment and storage medium
CN116501979A (en) Information recommendation method, information recommendation device, computer equipment and computer readable storage medium
CN114529399A (en) User data processing method, device, computer equipment and storage medium
CN116308651A (en) Financial product recommendation method and device based on time sequence and computer equipment
CN112559905A (en) Conversation recommendation method based on dual-mode attention mechanism and social similarity

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