CN113343095A - Model training and information recommendation method and device - Google Patents
Model training and information recommendation method and device Download PDFInfo
- Publication number
- CN113343095A CN113343095A CN202110690817.6A CN202110690817A CN113343095A CN 113343095 A CN113343095 A CN 113343095A CN 202110690817 A CN202110690817 A CN 202110690817A CN 113343095 A CN113343095 A CN 113343095A
- Authority
- CN
- China
- Prior art keywords
- merchant
- node
- keyword
- user
- determining
- 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
Links
Images
Classifications
-
- 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
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- 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
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Finance (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Accounting & Taxation (AREA)
- Biomedical Technology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Evolutionary Biology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The specification discloses a model training and information recommendation method and device, which can determine graph data according to keywords, merchants, merchant attributes and incidence relations among the keywords, the merchants and the merchant attributes searched by a user history. Then, a keyword for searching and each merchant associated with the keyword are determined from the behavior log data of the user. And determining sample data of each training sample according to the user characteristics of the user, the node characteristics of each associated merchant node determined from the graph data and the node characteristics of the keyword node. And then, labeling each training sample according to the service execution result in the behavior log data. And finally, training the click estimation model through each training sample and the label thereof. Model training is performed by combining the associated information of each associated node in the graph data, so that the richness of sample information is improved, and the output result of the model is more accurate.
Description
Technical Field
The application relates to the technical field of data processing, in particular to a model training and information recommendation method and device.
Background
Generally, when a user searches for goods in a takeaway platform, search results are usually presented in the form of a merchant list, and the order of the presented list is crucial to guiding the user to make a deal.
In the prior art, when determining the rank of each merchant in the display list, a number of merchants related to a query keyword input by a user may be recalled. And then, according to the behavior characteristics of the user such as browsing, clicking, placing orders and the like and the statistical characteristics of each merchant such as evaluation, order quantity and the like, determining the estimated score of each merchant clicked by the user Through a pre-trained Click estimation model (CTR), and sequencing the recalled merchants according to the estimated score of each merchant. Common CTR models include Deep Cross Network (DCN) model, Deep decomposition-Machine (Deep fm) model, Wide and Deep combination (Wide & Deep) model, and the like.
However, only focusing on the query keyword and the characteristics of the merchants to perform ranking, the quality of the ranking result is often poor, and "the front ranked merchant is the farther ahead, and the back ranked merchant is the farther behind".
Disclosure of Invention
The embodiment of the specification provides a model training and information recommendation method and device, which are used for partially solving the problems in the prior art.
The embodiment of the specification adopts the following technical scheme:
the model training method provided by the specification comprises the following steps:
determining the incidence relation between each keyword searched by the user and each merchant and the incidence relation between each keyword and each merchant attribute according to historical behavior log data of the user, and determining graph data according to the incidence relation among the nodes by taking each keyword, each merchant and each merchant attribute as nodes;
determining a keyword searched in the behavior log data and each merchant associated with the keyword for each piece of behavior log data, and determining sample data of a training sample for each associated merchant according to the node characteristics of the merchant node, the node characteristics of the keyword node and the user characteristics of a user corresponding to the behavior log data; wherein the node characteristics of each node are determined based on each other node in the graph data associated with the node;
determining the label of each training sample according to the service execution result in the behavior log data;
and aiming at each training sample, taking the sample data of the training sample as input, inputting a click pre-estimation model to be trained, determining the pre-estimation score of the training sample, and adjusting model parameters in the click pre-estimation model by taking the difference between the prediction score and the mark of the training sample as a target.
Optionally, determining, according to the behavior log data of the user, an association relationship between each keyword searched by the user and each merchant, specifically including:
for each keyword, determining the incidence relation between the keyword and each merchant according to the number of orders of each merchant after each user searches the keyword in the behavior log data; wherein the order-placing times are positively correlated with the relationship weight in the association relationship;
determining the incidence relation between each keyword and each merchant attribute according to the behavior log data of the user, which specifically comprises the following steps:
and aiming at each keyword, determining the incidence relation between the keyword and each merchant attribute according to the merchant attribute corresponding to the merchant which orders after each user searches the keyword in the behavior log data.
Optionally, before determining the node characteristics of the nodes based on the graph data, the method further includes:
and adjusting the model parameters in the graph data by taking the feature similarity of the connected nodes in the graph data as a maximum and the feature similarity of the unconnected nodes in the graph data as a minimum as targets.
Optionally, determining sample data of a training sample according to the node characteristics of the merchant node, the node characteristics of the keyword node, and the user characteristics of the user corresponding to the behavior log data, specifically including:
determining the matching degree of the merchant corresponding to the keyword according to the node characteristics of the merchant node and the node characteristics of the keyword node;
and determining sample data of the training sample according to the matching degree of the merchant corresponding to the keyword, the node characteristics of the merchant node and the user characteristics of the user corresponding to the behavior log data.
The information recommendation method provided by the specification comprises the following steps:
receiving a user search keyword sent by a terminal, and determining each merchant related to the keyword;
determining the node characteristics of the keyword nodes from the pre-constructed graph data, and determining the node characteristics of the merchant nodes from the pre-constructed graph data for each associated merchant; the graph data takes various merchants, various merchant attributes and various keywords searched by user history as nodes, and is constructed according to incidence relations among the nodes, and the incidence relations among the nodes are determined based on the behavior log data of the user;
determining the estimated score of the merchant through a pre-trained click estimation model according to the node characteristics of the merchant node, the node characteristics of the keyword node and the user characteristics of the user;
and sequencing the merchants according to the estimated scores of the merchants, and sending sequencing results to the terminal for display.
Optionally, determining the estimated score of the merchant according to the node characteristics of the merchant node, the node characteristics of the keyword node, and the user characteristics of the user through a pre-trained click estimation model, specifically including:
determining the matching degree of the merchant corresponding to the keyword according to the node characteristics of the merchant node and the node characteristics of the keyword node;
inputting the matching degree of the merchant corresponding to the keywords, the node characteristics of the merchant node and the user characteristics of the user into a pre-trained click estimation model, and determining the estimation score of the merchant.
The present specification provides a model training apparatus, including:
the mapping module is configured to determine association relations between the keywords searched by the user and the merchants and between the keywords and the attributes of the merchants according to historical behavior log data of the user, determine mapping data according to the association relations among the nodes by taking the keywords, the merchants and the attributes of the merchants as nodes;
the sample determining module is configured to determine a keyword searched in the behavior log data and each merchant associated with the keyword for each piece of behavior log data, and determine sample data of a training sample for each associated merchant according to the node characteristics of the merchant node, the node characteristics of the keyword node and the user characteristics of a user corresponding to the behavior log data; wherein the node characteristics of each node are determined based on each other node in the graph data associated with the node;
the marking module is configured to determine the marking of each training sample according to the service execution result in the behavior log data;
and the ordering module is configured to take the sample data of the training sample as input aiming at each training sample, input the click prediction model to be trained, determine the prediction score of the training sample, and adjust the model parameters in the click prediction model by taking the difference between the prediction score and the mark of the training sample as a target.
An information recommendation apparatus provided in this specification includes:
the receiving module is used for receiving a user search keyword sent by a terminal and determining each merchant related to the keyword;
the first determining module is used for determining the node characteristics of the keyword nodes from the pre-constructed graph data and determining the node characteristics of the merchant nodes from the pre-constructed graph data for each associated merchant; the graph data takes various merchants, various merchant attributes and various keywords searched by user history as nodes, and is constructed according to incidence relations among the nodes, and the incidence relations among the nodes are determined based on the behavior log data of the user;
the second determination module is used for determining the estimated score of the merchant through a pre-trained click estimation model according to the node characteristics of the merchant node, the node characteristics of the keyword node and the user characteristics of the user; and the sorting module sorts the merchants according to the estimated scores of the merchants and sends the sorting results to the terminal for display.
The present specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described model training and information recommendation method.
The electronic device provided by the present specification includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the model training and information recommendation method when executing the program.
The embodiment of the specification adopts at least one technical scheme which can achieve the following beneficial effects:
in this specification, the graph data may be determined according to each keyword, merchant attribute, and an association relationship therebetween searched in the user history. Then, a keyword for searching and each merchant associated with the keyword are determined from the behavior log data of the user. And determining sample data of each training sample according to the user characteristics of the user, the node characteristics of each associated merchant node determined from the graph data and the node characteristics of the keyword node. And then, labeling each training sample according to a businessman for ordering after the user searches in the behavior log data. And finally, training the click estimation model through each training sample and the label thereof. Model training is performed by combining the associated information of each associated node in the graph data, so that the richness of sample information is improved, and the output result of the model is more accurate.
Drawings
The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the application and together with the description serve to explain the application and not to limit the application. In the drawings:
fig. 1 is a schematic flow chart of a model training method provided in an embodiment of the present disclosure;
fig. 2 is a schematic diagram of a graph network provided in an embodiment of the present disclosure;
FIG. 3 is a diagram illustrating a network according to an embodiment of the present disclosure;
FIG. 4 is a schematic diagram of a model structure provided in an embodiment of the present disclosure;
fig. 5 is a schematic flowchart of an information recommendation method provided in an embodiment of the present specification;
FIG. 6 is a schematic structural diagram of a model training apparatus according to an embodiment of the present disclosure;
fig. 7 is a schematic structural diagram of an information recommendation device provided in an embodiment of the present specification;
fig. 8 is a schematic view of an electronic device implementing a model training method or an information recommendation method according to an embodiment of the present disclosure.
Detailed Description
In order to make the objects, technical solutions and advantages of the present disclosure more apparent, the technical solutions of the present disclosure will be clearly and completely described below with reference to the specific embodiments of the present disclosure and the accompanying drawings. It should be apparent that the described embodiments are only some of the embodiments of the present application, and not all of the embodiments. All other embodiments obtained by a person skilled in the art without making any inventive step based on the embodiments in the description belong to the protection scope of the present application.
The present specification provides a model training method, and the following describes technical solutions provided in embodiments of the present application in detail with reference to the accompanying drawings.
Fig. 1 is a schematic flow chart of a model training method provided in an embodiment of the present specification, which may specifically include the following steps:
s100: determining the incidence relation between each keyword searched by the user and each merchant and the incidence relation between each keyword and each merchant attribute according to the historical behavior log data of the user, and determining graph data according to the incidence relation among the nodes by taking each keyword, each merchant and each merchant attribute as nodes.
The model training method provided by the specification can be used for training a Click prediction model, predicting Click Through Rate (CTR) of a user to each merchant, and carrying out sequencing display according to the Click Through Rate from high to low.
In order to improve the effective information utilization rate and enable the sequencing result to be more accurate, in the description, a graph network related to searching can be constructed based on the historical search data of the user and the information of related merchants, and a click estimation model is trained based on nodes in the graph network and the incidence relation among the nodes.
Specifically, historical behavior log data of the user in the service platform can be obtained first, and each keyword of the historical search of the user is determined according to the behavior log data of each user. And then, aiming at each determined keyword, determining a merchant identifier of a merchant and a merchant attribute of the merchant, which are browsed or ordered in the display list after each user searches the keyword from the behavior log data of each user. The attribute of the merchant at least comprises the category of the merchant and the attribute of the brand of the merchant. The types of the products of the merchants can be divided into food types, beverage types, hot pots and the like, and the specification does not limit the division of the types of the products of the merchants and the brands to which the merchants belong, and can be specifically set according to needs.
Then, for each keyword, according to the behavior log data, after the keyword is searched by the user, the merchants browsing or placing orders in the display list determine the association relationship between the keyword and each merchant, wherein the order placing times of the user is positively correlated with the relationship weight in the association relationship, and the more the order placing times, the greater the relationship weight between the keyword and the merchants. And determining the incidence relation between the key word and each merchant attribute according to the merchant attribute of the merchant for browsing or ordering by the user.
And finally, determining graph data according to the incidence relation among the nodes by taking the keywords, the merchants and the attributes of the merchants as the nodes.
Fig. 2 is a schematic diagram of a graph network provided in this specification, where the graph network includes nodes of three types, i.e., a keyword for search, a merchant, and a merchant attribute, and an edge connected between the nodes is determined according to a merchant that a user browses or places an order after searching for each keyword, and a merchant attribute of the merchant.
Certainly, there may be multiple merchant attributes in this specification, and taking the example that the merchant attribute includes a merchant category and a merchant brand as an example for explanation, based on multiple different merchant attributes, a graph network constructed includes nodes of four different types, that is, a searched keyword, a merchant category, and a merchant brand, where an edge connected between the nodes is determined according to a merchant that a user browses or places an order after searching for the keyword, and a merchant category and a merchant brand to which the merchant belongs.
When the incidence relation between the keyword and the merchant categories is determined, the merchant categories connected with the keyword can be determined according to the category distribution of each issued merchant after the keyword is searched by a user. For example, assuming that 9 of the ordered merchants belong to the food category merchant and 1 belongs to the beverage category merchant, it is determined that the keyword is linked to the food category.
Further, taking the searched keyword as 'milk tea' as an example, according to historical behavior log data, statistics is carried out on the fact that after the keyword is searched by the user, beverage merchants a, b and c are displayed in the display list. The merchant a and the merchant c belong to the same merchant brand 1, the merchant b belongs to the merchant brand 2, n users place orders in the merchant a, 2n users place orders in the merchant b, and no user places an order in the merchant c. The constructed graph network is shown in fig. 3, wherein the keyword 'milk tea' is connected with the merchant a and has a relation weight of n, the keyword 'milk tea' is connected with the merchant b and has a relation weight of 2n, and the keyword 'milk tea' is not connected with the merchant c. Merchant a and merchant c are connected with merchant brand 1 and merchant b is connected with merchant brand 2 respectively. Since the merchant category of the merchants a, b and c is beverage category, the keyword 'milk tea' is connected with the merchant category 'beverage category'.
S102: and determining the keyword searched in the behavior log data and each merchant associated with the keyword aiming at each piece of behavior log data, and determining sample data of a training sample aiming at each associated merchant according to the node characteristics of the merchant node, the node characteristics of the keyword node and the user characteristics of the user corresponding to the behavior log data.
S104: and determining the label of each training sample according to the service execution result in the behavior log data.
In one or more embodiments of the present disclosure, after graph data with rich information is obtained by constructing a graph network, a training sample may be determined based on the graph data, and the training sample may be used to train a click prediction model to obtain a better ranking result.
Specifically, for each piece of behavior log data acquired in history, a keyword searched by a user in the behavior log data and each recalled merchant associated with the keyword may be determined.
Because rich association information exists among nodes in a pre-constructed graph network, in order to learn the association relationship among the nodes and enable the sequencing result to be more accurate, the node characteristics of each node can be comprehensively determined according to the node information of each node and the associated node information of other nodes.
When the sample data of the training sample is determined, the keyword node and each node associated with the keyword node can be determined from the pre-constructed graph data, and the node characteristics of the keyword node are determined according to the node information of the keyword node and the node information of each other node associated with the keyword node. And aiming at each associated merchant, determining the merchant node and each node associated with the merchant node from the pre-constructed graph data, and determining the node characteristics of the merchant node according to the node information of the merchant node and the node information of each other node associated with the merchant node.
After the graph network is constructed, graph learning may be further performed by using a graph SAmple and aggregation algorithm (graph SAGE), and node characteristics of each target node are determined by sampling other nodes associated with the target node and aggregating node information of each sampled node.
When graph learning is performed, in order to reflect the mutual influence between the associated nodes, the model parameters in the graph data can be adjusted with the goal of maximizing the feature similarity of the connected nodes in the graph data and minimizing the feature similarity of the unconnected nodes in the graph data. Therefore, the loss function can be set as the difference between the feature similarity of the unconnected node and the connected node, and the model parameters are adjusted with the goal of minimizing the loss function.
And finally, determining sample data of the training sample according to the node characteristics of the merchant node, the node characteristics of the keyword node and the user characteristics of the user corresponding to the behavior log data. The user characteristics of the user can be determined based on behavior log data of the user in the service platform, and include merchants who place orders in the user history and the order placing times, the click rate of the user on various merchant categories, merchant identifications of various merchants clicked by the user within a preset time length and the like.
Further, when sample data of the training sample is determined, the matching degree of the merchant corresponding to the keyword can be determined according to the node characteristics of the merchant node and the node characteristics of the keyword node, and the matching degree of the merchant corresponding to the keyword, the node characteristics of the merchant node and the user characteristics of the user are used as the sample data of the training sample.
In this specification, after the sample data of each training sample is determined, each training sample may be labeled according to the service execution result in the behavior log data. The business execution result at least comprises various merchants browsing, clicking and placing orders after the user searches the key words. For example, the merchant that the user placed an order is labeled with a score of 100, the merchant that the user clicked on is labeled with a score of 80, and the merchant that the user browsed is labeled with a score of 10.
S106: and aiming at each training sample, taking the sample data of the training sample as input, inputting a click pre-estimation model to be trained, determining the pre-estimation score of the training sample, and adjusting model parameters in the click pre-estimation model by taking the difference between the pre-estimation score and the mark of the training sample as a target.
After each training sample and the label thereof are determined, the click pre-estimation model can be trained.
Specifically, for each training sample, the sample data of the training sample is used as input and input into the click pre-estimation model to be trained, the pre-estimation score of the training sample output by the click pre-estimation model is determined, and the model parameters in the ranking model are adjusted by taking the difference between the pre-estimation score and the labeling score of the training sample as a target. The click prediction model can adopt CTR models such as DCN, Deep FM, Wide & Deep and the like.
As shown in fig. 4, the matching degree between the merchant and the keyword may be determined based on the node characteristics of the merchant node and the node characteristics of the keyword node determined in the graph data. And then, inputting the click estimation model by taking the user characteristics of the searched user, the node characteristics of the merchant node and the matching degree between the merchant and the keyword as input, and determining the estimation score output by the model.
Based on the model training method shown in fig. 1, graph data can be determined according to keywords, businesses, business attributes and incidence relations among the keywords, the businesses and the business attributes searched by the user history. Then, a keyword for searching and each merchant associated with the keyword are determined from the behavior log data of the user. And determining sample data of each training sample according to the user characteristics of the user, the node characteristics of each associated merchant node determined from the graph data and the node characteristics of the keyword node. And then, labeling each training sample according to a businessman for ordering after the user searches in the behavior log data. And finally, training the click estimation model through each training sample and the label thereof. Model training is performed by combining the associated information of each associated node in the graph data, so that the richness of sample information is improved, and the output result of the model is more accurate.
In one embodiment of the present specification, in order to ensure that the graph data is valid for the ranking of the merchants, the validity of the graph data may be determined by comparing the influence of the training samples containing the graph data and the influence of the training samples not containing the graph data on the prediction result output by the click prediction model.
Specifically, sample data including statistical characteristics of merchants and user characteristics may be determined as a first sample, and sample data including node characteristics of merchant nodes, node characteristics of keyword nodes, and user characteristics of users may be determined as a second sample. And then training a first click prediction model and a second click prediction model based on the first sample and the second sample respectively. And finally, determining the estimation effect of the first click estimation model and the estimation effect of the second click estimation model through model evaluation indexes according to the estimation result output by the first click estimation model and the estimation result output by the second click estimation model. If the estimation effect of the second click estimation model is better than that of the first click estimation model, the graph data is shown to be effective for the ranking of the merchants, otherwise, the model parameters in the graph data are readjusted until the estimation effect of the second click estimation model is better than that of the first click estimation model. The model evaluation index may be a common Area Under Current (AUC) index, a mean average accuracy index (mAP), or the like.
The specification also provides an information recommendation method, which can be used for sequencing all merchants called back by adopting the click estimation model trained by the model training method and displaying the merchants to users according to the sequencing.
Fig. 5 is a schematic flow chart of an information recommendation method provided in an embodiment of this specification, which may specifically include the following steps:
s200: and receiving a user search keyword sent by a terminal, and determining each merchant related to the keyword.
The information recommendation method provided by the present specification may be specifically executed by a server of a service platform. After the user inputs the search keyword at the terminal, the terminal can send the keyword searched by the user to a server of the service platform so as to determine to display recommended information to the user. The keyword may be a product, a shop name, a travel location, and the like, which is not limited in this specification.
Specifically, the server may receive a keyword searched by the user and sent by the terminal, and a current location of the user. And then, according to the current position of the user and the searched keywords, recalling a plurality of merchants which are in a preset range of the current position and are associated with the keywords from all merchants.
S202: and determining the node characteristics of the keyword nodes from the pre-constructed graph data, and determining the node characteristics of the merchant nodes from the pre-constructed graph data for each associated merchant.
In this specification, after a plurality of associated merchants are recalled based on keywords searched by a user, the ranking results of the associated merchants can be determined according to the probability of ordering the associated merchants by the user, and displayed in sequence.
Specifically, the server may determine the node characteristics of the keyword node from pre-constructed graph data. And determining node characteristics of the merchant node from the pre-constructed graph data for each merchant associated with the keyword. The process of constructing the graph data may refer to step S100, which is not described herein again. In the graph data, the node characteristics of the keyword node are comprehensively determined based on the node information of the keyword node and the node information of each node associated with the keyword node; the node characteristics of the merchant node are also determined synthetically based on the node information of the merchant node and the node information of each node associated with the merchant node.
S204: and determining the estimated score of the merchant through a pre-trained click estimation model according to the node characteristics of the merchant node, the node characteristics of the keyword node and the user characteristics of the user.
In this specification, after determining each factor that affects the user to click on the merchant, the influence factors may be integrated to determine the click probability of the user on each merchant, so as to determine the display order of each merchant according to the click probability.
Specifically, for each recalled merchant, the server may input the node characteristics of the merchant node, the node characteristics of the keyword, and the user characteristics of the user into a pre-trained click prediction model, and determine the prediction score of the merchant output by the click prediction model. The estimated score of the merchant represents the probability that the user clicks the merchant after searching the keyword. The user characteristics of the user at least comprise user preference characteristics such as order placing merchants and order placing times after the user searches the keywords historically. The training process of the click prediction model may refer to the detailed processes of the above steps S100 to S106, which will not be described in this specification.
Further, in order to improve the rank of the merchants highly matched with the search content, the server may also determine the matching degree of the merchant corresponding to the keyword according to the node characteristics of the merchant node and the node characteristics of the keyword node, and input the user characteristics of the user, the node characteristics of the merchant node, and the matching degree of the merchant corresponding to the keyword into the click estimation model to determine the estimated score of the merchant. The node characteristics of the merchant node at least comprise the order quantity of the merchant, the merchant identification, the user evaluation and other information.
S206: and sequencing the merchants according to the estimated scores of the merchants, and sending sequencing results to the terminal for display.
After the estimated scores of all the merchants clicked by the user are determined, the server can sort the recalled merchants from high to low according to the estimated scores of all the merchants, and sends the sorting result to the terminal, so that the terminal can display the merchant information of all the merchants according to the sorting.
Based on the information recommendation method shown in fig. 5, each merchant associated with the keyword can be determined according to the search keyword sent by the terminal. And then determining the node characteristics of the keyword node and the associated node characteristics of each merchant node from the pre-constructed graph data, and determining the estimated score of the merchant through a pre-trained click estimation model according to the node characteristics of the merchant node, the node characteristics of the keyword node and the user characteristics of the user. And finally, sequencing the merchants according to the estimated scores of the merchants, and sending the sequencing result to the terminal for display. By combining the associated information of each associated node in the graph data, all merchants are comprehensively sequenced, so that the sequencing result is more accurate.
And for the merchants in the back of the sequence, the merchants in the front of the sequence are associated in the graph data through the attributes of the merchants, so that the matching degree with the keywords is improved, the probability of the subsequent sequence in the front is higher, and the problem that the merchants in the back of the sequence are behind the former merchant is relieved to a certain extent.
In addition, in this specification, in order to ensure timeliness of the ranking result, the graph data is periodically reconstructed based on the latest behavior log data of the user, and the click estimation model is retrained according to the reconstructed graph data.
It should be noted that the information recommendation method provided above is not limited to ranking the recalled merchants. When information such as each product, tourist attractions and the like is recalled based on keywords searched by a user, the recall objects can be sorted by adopting the method. For example, when the user searches for commodities in the e-commerce platform, the recalled commodities can be sorted by the method to determine the display sequence of the commodities.
Based on the model training method shown in fig. 1, an embodiment of the present specification further provides a schematic structural diagram of a model training apparatus, as shown in fig. 6.
Fig. 6 is a schematic structural diagram of a model training apparatus provided in an embodiment of the present disclosure, including:
the graph building module 300 is configured to determine, according to the historical behavior log data of the user, an association relationship between each keyword searched by the user and each merchant and an association relationship between each keyword and each merchant attribute, and determine graph data according to the association relationship between each node by using each keyword, each merchant and each merchant attribute as a node;
a sample determination module 302, configured to determine, for each piece of behavior log data, a keyword searched in the behavior log data, and each merchant associated with the keyword, and determine, for each associated merchant, sample data of a training sample according to a node feature of a node of the merchant, a node feature of the keyword node, and a user feature of a user corresponding to the behavior log data; wherein the node characteristics of each node are determined based on each other node in the graph data associated with the node;
a labeling module 304, configured to determine labels of the training samples according to the service execution results in the behavior log data;
the sorting module 306 is configured to, for each training sample, take sample data of the training sample as input, input the click prediction model to be trained, determine the prediction score of the training sample, and adjust model parameters in the click prediction model with a goal of minimizing a difference between the prediction score and the label of the training sample.
Optionally, the mapping module 300 is specifically configured to, for each keyword, determine an association relationship between the keyword and each merchant according to the number of orders placed by each merchant after each user searches for the keyword in the behavior log data; the order placing times are positively correlated with the relation weight in the incidence relation, and the method is used for determining the incidence relation between the keyword and each merchant attribute according to the merchant attribute corresponding to the order placing merchant after each user searches the keyword in the behavior log data aiming at each keyword
Optionally, the graph building module 300 is further configured to adjust the model parameters in the graph data with the goal of maximizing the feature similarity of the connected nodes in the graph data and minimizing the feature similarity of the unconnected nodes in the graph data.
Optionally, the sample determining module 302 is specifically configured to determine, according to the node feature of the merchant node and the node feature of the keyword node, a matching degree of the merchant corresponding to the keyword, and determine sample data of the training sample according to the matching degree of the merchant corresponding to the keyword, the node feature of the merchant node, and the user feature of the user corresponding to the behavior log data.
Based on the information recommendation method shown in fig. 5, an embodiment of the present specification further provides a schematic structural diagram of an information recommendation apparatus, as shown in fig. 7.
Fig. 7 is a schematic structural diagram of an information recommendation apparatus provided in an embodiment of this specification, including:
the receiving module 400 is used for receiving a user search keyword sent by a terminal and determining each merchant related to the keyword;
a first determining module 402, configured to determine, from pre-constructed graph data, node characteristics of the keyword node, and determine, for each associated merchant, node characteristics of the merchant node from pre-constructed graph data; the graph data takes various merchants, various merchant attributes and various keywords searched by user history as nodes, and is constructed according to incidence relations among the nodes, and the incidence relations among the nodes are determined based on the behavior log data of the user;
a second determining module 404, configured to determine an estimated score of the merchant according to the node characteristics of the merchant node, the node characteristics of the keyword node, and the user characteristics of the user through a pre-trained click estimation model;
and the sorting module 406 sorts the merchants according to the estimated scores of the merchants, and sends the sorting result to the terminal for display.
Optionally, the second determining module 404 is specifically configured to determine, according to the node feature of the merchant node and the node feature of the keyword node, a matching degree of the merchant corresponding to the keyword;
inputting the matching degree of the merchant corresponding to the keywords, the node characteristics of the merchant node and the user characteristics of the user into a pre-trained click estimation model, and determining the estimation score of the merchant.
Embodiments of the present specification further provide a computer-readable storage medium, where the storage medium stores a computer program, and the computer program may be used to execute the model training method provided in fig. 1 or the information recommendation method provided in fig. 5.
The embodiment of the present specification also proposes a schematic structural diagram of the electronic device shown in fig. 8. As shown in fig. 8, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, but may also include hardware required for other services. The processor reads a corresponding computer program from the non-volatile memory into the memory and then runs the computer program to implement the model training method provided in fig. 1 or the information recommendation method provided in fig. 5.
Of course, besides the software implementation, the present specification does not exclude other implementations, such as logic devices or a combination of software and hardware, and the like, that is, the execution subject of the following processing flow is not limited to each logic unit, and may be hardware or logic devices.
In the 90 s of the 20 th century, improvements in a technology could clearly distinguish between improvements in hardware (e.g., improvements in circuit structures such as diodes, transistors, switches, etc.) and improvements in software (improvements in process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical modules. For example, a Programmable Logic Device (PLD), such as a Field Programmable Gate Array (FPGA), is an integrated circuit whose Logic functions are determined by programming the Device by a user. A digital system is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and create a dedicated integrated circuit chip. Furthermore, nowadays, instead of manually generating an Integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development and writing, but the original code before compiling is also written by a specific Programming Language, which is called Hardware Description Language (HDL), and HDL is not only one but many, such as abel (advanced Boolean Expression Language), ahdl (alternate Language Description Language), traffic, pl (core unified Programming Language), HDCal, JHDL (Java Hardware Description Language), langue, Lola, HDL, laspam, hardbyscript Description Language (vhigh Description Language), and so on, which are currently used in the most popular languages. It will also be apparent to those skilled in the art that hardware circuitry that implements the logical method flows can be readily obtained by merely slightly programming the method flows into an integrated circuit using the hardware description languages described above.
The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic for the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller as pure computer readable program code, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be considered a hardware component, and the means included therein for performing the various functions may also be considered as a structure within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.
The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. One typical implementation device is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functions of the various elements may be implemented in the same one or more software and/or hardware implementations of the present description.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
This description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
The above description is only an example of the present specification, and is not intended to limit the present specification. Various modifications and alterations to this description will become apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification.
Claims (10)
1. A method of model training, comprising:
determining the incidence relation between each keyword searched by the user and each merchant and the incidence relation between each keyword and each merchant attribute according to historical behavior log data of the user, and determining graph data according to the incidence relation among the nodes by taking each keyword, each merchant and each merchant attribute as nodes;
determining a keyword searched in the behavior log data and each merchant associated with the keyword for each piece of behavior log data, and determining sample data of a training sample for each associated merchant according to the node characteristics of the merchant node, the node characteristics of the keyword node and the user characteristics of a user corresponding to the behavior log data; wherein the node characteristics of each node are determined based on each other node in the graph data associated with the node;
marking each training sample according to a service execution result in the behavior log data;
and aiming at each training sample, taking the sample data of the training sample as input, inputting a click pre-estimation model to be trained, determining the pre-estimation score of the training sample, and adjusting model parameters in the click pre-estimation model by taking the difference between the prediction score and the mark of the training sample as a target.
2. The method according to claim 1, wherein determining, according to the behavior log data of the user, an association between each keyword searched by the user and each merchant specifically comprises:
for each keyword, determining the incidence relation between the keyword and each merchant according to the number of orders of each merchant after each user searches the keyword in the behavior log data; wherein the order-placing times are positively correlated with the relationship weight in the association relationship;
determining the incidence relation between each keyword and each merchant attribute according to the behavior log data of the user, which specifically comprises the following steps:
and aiming at each keyword, determining the incidence relation between the keyword and each merchant attribute according to the merchant attribute corresponding to the merchant which orders after each user searches the keyword in the behavior log data.
3. The method of claim 1, wherein prior to determining node characteristics of a node based on the graph data, the method further comprises:
and adjusting the model parameters in the graph data by taking the feature similarity of the connected nodes in the graph data as a maximum and the feature similarity of the unconnected nodes in the graph data as a minimum as targets.
4. The method according to claim 1, wherein determining sample data of a training sample according to the node characteristics of the merchant node, the node characteristics of the keyword node, and the user characteristics of the user corresponding to the behavior log data specifically comprises:
determining the matching degree of the merchant corresponding to the keyword according to the node characteristics of the merchant node and the node characteristics of the keyword node;
and determining sample data of the training sample according to the matching degree of the merchant corresponding to the keyword, the node characteristics of the merchant node and the user characteristics of the user corresponding to the behavior log data.
5. An information recommendation method, comprising:
receiving a user search keyword sent by a terminal, and determining each merchant related to the keyword;
determining the node characteristics of the keyword nodes from the pre-constructed graph data, and determining the node characteristics of the merchant nodes from the pre-constructed graph data for each associated merchant; the graph data takes various merchants, various merchant attributes and various keywords searched by user history as nodes, and is constructed according to incidence relations among the nodes, and the incidence relations among the nodes are determined based on the behavior log data of the user;
determining the estimated score of the merchant through a pre-trained click estimation model according to the node characteristics of the merchant node, the node characteristics of the keyword node and the user characteristics of the user;
and sequencing the merchants according to the estimated scores of the merchants, and sending sequencing results to the terminal for display.
6. The method of claim 5, wherein determining the estimated score of the merchant according to the node characteristics of the merchant node, the node characteristics of the keyword node, and the user characteristics of the user through a pre-trained click estimation model comprises:
determining the matching degree of the merchant corresponding to the keyword according to the node characteristics of the merchant node and the node characteristics of the keyword node;
inputting the matching degree of the merchant corresponding to the keywords, the node characteristics of the merchant node and the user characteristics of the user into a pre-trained click estimation model, and determining the estimation score of the merchant.
7. A model training apparatus, comprising:
the mapping module is configured to determine association relations between the keywords searched by the user and the merchants and between the keywords and the attributes of the merchants according to historical behavior log data of the user, determine mapping data according to the association relations among the nodes by taking the keywords, the merchants and the attributes of the merchants as nodes;
the sample determining module is configured to determine a keyword searched in the behavior log data and each merchant associated with the keyword for each piece of behavior log data, and determine sample data of a training sample for each associated merchant according to the node characteristics of the merchant node, the node characteristics of the keyword node and the user characteristics of a user corresponding to the behavior log data; wherein the node characteristics of each node are determined based on each other node in the graph data associated with the node;
the marking module is configured to mark each training sample according to the service execution result in the behavior log data;
and the ordering module is configured to take the sample data of the training sample as input aiming at each training sample, input the click prediction model to be trained, determine the prediction score of the training sample, and adjust the model parameters in the click prediction model by taking the difference between the prediction score and the mark of the training sample as a target.
8. An information recommendation apparatus, comprising:
the receiving module is used for receiving a user search keyword sent by a terminal and determining each merchant related to the keyword;
the first determining module is used for determining the node characteristics of the keyword nodes from the pre-constructed graph data and determining the node characteristics of the merchant nodes from the pre-constructed graph data for each associated merchant; the graph data takes various merchants, various merchant attributes and various keywords searched by user history as nodes, and is constructed according to incidence relations among the nodes, and the incidence relations among the nodes are determined based on the behavior log data of the user;
the second determination module is used for determining the estimated score of the merchant through a pre-trained click estimation model according to the node characteristics of the merchant node, the node characteristics of the keyword node and the user characteristics of the user;
and the sorting module sorts the merchants according to the estimated scores of the merchants and sends the sorting results to the terminal for display.
9. A computer-readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of any of claims 1 to 4 or 5 to 6.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the method of any of claims 1 to 4 or 5 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110690817.6A CN113343095A (en) | 2021-06-22 | 2021-06-22 | Model training and information recommendation method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110690817.6A CN113343095A (en) | 2021-06-22 | 2021-06-22 | Model training and information recommendation method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113343095A true CN113343095A (en) | 2021-09-03 |
Family
ID=77477539
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110690817.6A Pending CN113343095A (en) | 2021-06-22 | 2021-06-22 | Model training and information recommendation method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113343095A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113887234A (en) * | 2021-09-15 | 2022-01-04 | 北京三快在线科技有限公司 | Model training and recommending method and device |
| CN117398662A (en) * | 2023-12-15 | 2024-01-16 | 苏州海易泰克机电设备有限公司 | Three-degree-of-freedom rotation training parameter control method based on physiological acquisition information |
-
2021
- 2021-06-22 CN CN202110690817.6A patent/CN113343095A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113887234A (en) * | 2021-09-15 | 2022-01-04 | 北京三快在线科技有限公司 | Model training and recommending method and device |
| CN113887234B (en) * | 2021-09-15 | 2023-01-06 | 北京三快在线科技有限公司 | Model training and recommending method and device |
| CN117398662A (en) * | 2023-12-15 | 2024-01-16 | 苏州海易泰克机电设备有限公司 | Three-degree-of-freedom rotation training parameter control method based on physiological acquisition information |
| CN117398662B (en) * | 2023-12-15 | 2024-03-12 | 苏州海易泰克机电设备有限公司 | Three-degree-of-freedom rotation training parameter control method based on physiological acquisition information |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9665622B2 (en) | Publishing product information | |
| TWI652584B (en) | Method and device for matching text information and pushing business objects | |
| CN110413877A (en) | A kind of resource recommendation method, device and electronic equipment | |
| US20170154116A1 (en) | Method and system for recommending contents based on social network | |
| CN108596695B (en) | Entity pushing method and system | |
| CN105989004A (en) | Information releasing pretreatment method and device | |
| CN107832338B (en) | Method and system for recognizing core product words | |
| CN106126544B (en) | Internet content delivery method and device | |
| JP6728178B2 (en) | Method and apparatus for processing search data | |
| CN113010640B (en) | Service execution method and device | |
| CA3059929C (en) | Text searching method, apparatus, and non-transitory computer-readable storage medium | |
| CN110599307A (en) | Commodity recommendation method and device | |
| CN113688313A (en) | Training method of prediction model, information pushing method and device | |
| CN113343095A (en) | Model training and information recommendation method and device | |
| CN114329207A (en) | Multi-service information sequencing system, method, storage medium and electronic equipment | |
| CN111061954A (en) | Search result sorting method and device and storage medium | |
| CN112733024A (en) | Information recommendation method and device | |
| CN113010809A (en) | Information recommendation method and device | |
| CN112699307A (en) | Information generation method and device | |
| CN111639269A (en) | Site recommendation method and device | |
| CN113343132B (en) | Model training method, information display method and device | |
| CN108830298A (en) | A kind of method and device of determining user characteristics label | |
| CN114331602A (en) | Model training method based on transfer learning, information recommendation method and device | |
| TWI547888B (en) | A method of recording user information and a search method and a server | |
| CN112613320A (en) | Method and device for acquiring similar sentences, storage medium and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication |