CN113626683B

CN113626683B - CTR (control parameter) estimation processing method and device, electronic equipment and storage medium

Info

Publication number: CN113626683B
Application number: CN202110745575.6A
Authority: CN
Inventors: 陈珊; 王泽华; 王兴星
Original assignee: Beijing Sankuai Online Technology Co Ltd
Current assignee: Beijing Sankuai Online Technology Co Ltd
Priority date: 2021-06-30
Filing date: 2021-06-30
Publication date: 2023-05-30
Anticipated expiration: 2041-06-30
Also published as: CN113626683A

Abstract

The invention provides a CTR (control parameter) estimation processing method, a CTR estimation processing device, electronic equipment and a storage medium. The method comprises the following steps: aiming at any target POI, acquiring a historical CTR of the target POI under each query word condition according to a historical query record, acquiring N query words with the highest historical CTR of the target POI, and constructing a first tag sequence of the target POI; according to the CTR related parameters and real-time query words of a target user, a CTR estimated value of the target user for the target POI is obtained through a CTR estimated model; wherein the CTR related parameter comprises at least a first tag sequence of the target POI. Thereby, the beneficial effect of improving the accuracy of the CTR estimated result is achieved.

Description

CTR (control parameter) estimation processing method and device, electronic equipment and storage medium

Technical Field

The present invention relates to the field of internet technologies, and in particular, to a CTR estimation processing method, a device, an electronic device, and a storage medium.

Background

In application scenarios such as searching, in order to better improve the matching degree between the search feedback result and the search requirement of the user, it is necessary to accurately estimate the CTR (Click-Through-Rate) of the user for each target POI (target poinfintelest/target POInt ofInformation, interest point/information point). How to model the matching of a user and a target POI becomes one of the key research directions of searching.

Taking a search scenario in a take-out scenario as an example, advertisement presentation in the take-out search scenario mainly refers to a merchant (target POI) and a menu (SPU) unit. Because the user is more prone to clicking on the merchant area, click data of the user in the dish dimension is sparse; and the dish information can be aggregated to a merchant unit, so that the existing scheme is mainly modeled and matched from two aspects of Query (Query word) and merchant, historical behavior information of a user and merchant.

However, in practical applications, the emphasis of different merchants is different due to excessive merchant information (such as products, dishes, brands, etc.). The existing information mode for constructing the merchant can mix various types of information together, does not measure which type of information is used for accurately representing the merchant, and can bring more noise. In addition, the re-purchase phenomenon is obvious in the take-away scene, meanwhile, the user preference is diversified in the take-away scene, the information construction sequences such as order extraction products or target POI ID are not enough at present, the user information is not sufficiently and accurately expressed, and further the accuracy of CTR estimated results can be affected.

Disclosure of Invention

The embodiment of the invention provides a CTR (control parameter) estimation processing method, a CTR estimation processing device, electronic equipment and a storage medium, which are used for solving the problem that the accuracy of a CTR estimation result is easily affected in the related technology.

In order to solve the technical problems, the invention is realized as follows:

in a first aspect, an embodiment of the present invention provides a CTR estimation processing method, including:

aiming at any target POI, acquiring a historical CTR of the target POI under each query word condition according to a historical query record, acquiring N query words with the highest historical CTR of the target POI, and constructing a first tag sequence of the target POI;

according to the CTR related parameters and real-time query words of a target user, a CTR estimated value of the target user for the target POI is obtained through a CTR estimated model;

wherein the CTR related parameter comprises at least a first tag sequence of the target POI.

Optionally, the step of obtaining, according to the real-time query word and the CTR related parameter of the target user, a CTR estimated value of the target user for the target POI through a CTR estimated model includes:

aiming at each POI contained in each historical order of the target user, acquiring a historical CTR of the POI under each query word condition according to a historical query record, acquiring M highest query words in the query words corresponding to the target user, and constructing a second tag sequence of the target user;

According to the CTR related parameters and the real-time query words of the target user, a CTR estimated value of the target user for the target POI is obtained through the CTR estimated model;

wherein the CTR-related parameter includes at least the first tag sequence and the second tag sequence.

Optionally, the step of obtaining the history CTR of the POI under each query term according to the history query record, obtaining M query terms with highest CTR among the query terms corresponding to the target user, and constructing the second tag sequence of the target user includes:

aiming at each POI contained in each historical order of the target user, acquiring a historical CTR of the POI under the condition of each query word according to a historical query record;

aiming at any POI contained in any historical order, acquiring the weight of each query word in the historical order based on the historical CTR of the POI in each query word, and ensuring that the weight and the value of the query word in the same historical order are specified values;

adding weights of the same query word under each historical order to obtain a CTR of the query word corresponding to the target user;

and obtaining M query words with highest CTR in the query words corresponding to the target user, and constructing a second tag sequence of the target user.

Optionally, the CTR estimation model is a combination of a dual-sequence modeling sub-network model and a CTR estimation sub-network model, and the step of obtaining, by using the CTR estimation model, a CTR estimated value of the target user for the target POI according to the CTR related parameter and a real-time query word of the target user includes:

acquiring a feature vector of the first tag sequence and the second tag sequence after crossing through the dual-sequence modeling sub-network model;

according to the feature vector, the real-time query word of the target user and other CTR related parameters, a CTR estimated value of the target user for the target POI under the condition of the real-time query word is obtained through a CTR estimated sub-network model;

wherein the other CTR-related parameters are other parameters than the first tag sequence and the second tag sequence in the CTR-related parameters.

Optionally, the dual-sequence modeling sub-network model comprises an embedded layer, an attention mechanism layer and a pooling layer which are sequentially cascaded;

the step of obtaining the feature vector of the first tag sequence and the second tag sequence after crossing through the dual-sequence modeling sub-network model comprises the following steps:

Acquiring a first matrix representation of the real-time query word through a first embedding layer, and simultaneously acquiring a second matrix representation of the first tag sequence and a third matrix representation of the second tag sequence through a second embedding layer respectively;

the input matrix of the attention mechanism layer is obtained by replacing the input of the K matrix in the attention mechanism layer at the user side through the first matrix representation;

and obtaining the output vector of the attention mechanism layer, inputting the output vector into the pooling layer, and obtaining the feature vector after crossing through the pooling layer.

Optionally, the embedded layer includes any one of a Word embedding layer, a Position Embeddding layer, a Segment Embedding layer, or a combination of any plurality of the same; the attention mechanism layer comprises a multi-head attention mechanism; the pooling layer includes an average pooling layer.

In a second aspect, an embodiment of the present invention provides another CTR estimation processing method, including:

Aiming at any target POI, according to CTR related parameters and real-time query words of the target user, a CTR estimated value of the target user aiming at the target POI is obtained through a CTR estimated model;

wherein the CTR-related parameter comprises at least the second tag sequence.

Optionally, the step of obtaining, for any target POI, a CTR estimated value of the target user for the target POI through a CTR estimation model according to a CTR related parameter and a real-time query word of the target user includes:

according to the CTR related parameters and real-time query words of a target user, a CTR estimated value of the target user for the target POI is obtained through the CTR estimated model;

In a third aspect, an embodiment of the present invention provides a CTR estimation processing apparatus, including:

the first tag sequence construction module is used for acquiring the historical CTR of any target POI under the condition of each query word according to the historical query record, acquiring N query words with the highest historical CTR of the target POI, and constructing a first tag sequence of the target POI;

the first CTR estimation module is used for acquiring a CTR estimated value of the target user aiming at the target POI through a CTR estimation model according to CTR related parameters and real-time query words of the target user;

Optionally, the first CTR estimation module includes:

a second tag sequence construction sub-module, configured to obtain, for each POI included in each historical order of the target user, a historical CTR of the POI under each query word condition according to a historical query record, obtain M query words with highest CTR among query words corresponding to the target user, and construct a second tag sequence of the target user;

the first CTR estimation sub-module is used for acquiring a CTR estimated value of the target user aiming at the target POI through the CTR estimation model according to CTR related parameters and real-time query words of the target user;

Optionally, the second tag sequence constructs a sub-module, specifically for:

Optionally, the CTR estimation model is a combination of a dual-sequence modeling sub-network model and a CTR estimation sub-network model, and the first CTR estimation sub-module includes:

the information crossing unit is used for obtaining the feature vector of the first tag sequence and the second tag sequence after crossing through the double-sequence modeling sub-network model;

the CTR estimation unit is used for acquiring a CTR estimated value of the target user aiming at the target POI under the condition of the real-time query word through a CTR estimation sub-network model according to the feature vector, the real-time query word of the target user and other CTR related parameters;

The information crossing unit is specifically configured to:

In a fourth aspect, an embodiment of the present invention provides another CTR estimate processing apparatus, including:

the second tag sequence construction module is used for acquiring a historical CTR of each POI under the condition of each query word according to a historical query record aiming at each POI contained in each historical order of the target user, acquiring M query words with highest CTR in the query words corresponding to the target user, and constructing a second tag sequence of the target user;

The second CTR estimation module is used for acquiring a CTR estimated value of any target POI of the target user through a CTR estimation model according to the CTR related parameters and the real-time query words of the target user;

wherein the CTR-related parameter comprises at least the second tag sequence.

Optionally, the second CTR estimation module includes:

the first tag sequence construction submodule is used for aiming at any target POI, acquiring a historical CTR of the target POI under each query word condition according to a historical query record, acquiring N query words with the highest historical CTR of the target POI, and constructing a first tag sequence of the target POI;

the second CTR estimation sub-module is used for acquiring a CTR estimated value of the target user aiming at the target POI through the CTR estimation model according to CTR related parameters and real-time query words of the target user;

Optionally, the CTR estimation model is a combination of a dual-sequence modeling sub-network model and a CTR estimation sub-network model, and the second CTR estimation sub-module is specifically configured to:

In a fifth aspect, an embodiment of the present invention further provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, which when executed by the processor, performs the steps of the CTR estimation processing method according to the first and second aspects.

In a sixth aspect, an embodiment of the present invention further provides a computer readable storage medium, where a computer program is stored on the computer readable storage medium, where the computer program when executed by a processor implements the steps of the CTR estimation processing method according to the first aspect and the second aspect.

In the embodiment of the invention, in the first aspect, the tag sequence of the POI is constructed by introducing the historical click Query sequence of the POI so as to accurately describe the POI, thereby improving the accuracy of the CTR estimated result of the target user for the POI; in the second aspect, a label sequence of the user is constructed by introducing query words of POIs corresponding to the user orders so as to enrich the information expression of the user and accurately describe the user demands, and the accuracy of CTR estimated results of the target user on the POIs can be improved to a certain extent; in the third aspect, the improved CTR estimation model structure is utilized to model the matching between the tag sequence of the POI and the tag sequence of the user, so that the purpose of simultaneously modeling the matching between the instant Query of the user and the historical behavior information of the user and the candidate POI is achieved, and the accuracy of the CTR estimation result of the target user for the POI is further improved.

The foregoing description is only an overview of the present invention, and is intended to be implemented in accordance with the teachings of the present invention in order that the same may be more clearly understood and to make the same and other objects, features and advantages of the present invention more readily apparent.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments of the present invention will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flowchart illustrating a CTR estimation processing method according to an embodiment of the present invention;

FIG. 2 is a flowchart illustrating another CTR estimation processing method according to an embodiment of the present invention;

FIG. 3A is a schematic diagram of a dual sequence modeling sub-network model in an embodiment of the present invention;

FIG. 3B is a schematic structural diagram of a CTR estimation model according to an embodiment of the present invention;

FIG. 4 is a flowchart illustrating another CTR estimation processing method according to an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of a CTR estimation processing device according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of another CTR estimation processing apparatus according to an embodiment of the present invention;

fig. 7 is a schematic diagram of a hardware structure of an electronic device according to an embodiment of the present invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Referring to fig. 1, a flowchart illustrating steps of a CTR estimation processing method according to an embodiment of the present invention is shown.

Step 110, for any target POI, acquiring a history CTR of the target POI under each query word condition according to a history query record, and acquiring N query words with the highest history CTR of the target POI, and constructing a first tag sequence of the target POI;

step 120, according to the relevant parameters of the CTR and the real-time query words of the target user, obtaining a CTR estimated value of the target user aiming at the target POI through a CTR estimated model; wherein the CTR related parameter comprises at least a first tag sequence of the target POI.

As described above, in practical applications, the information of POIs such as takeaway is too much (such as products, dishes, brands, etc.), but the emphasis of different POIs may also be different. For example, in a take-away scenario, some businesses are running brands (e.g., large KA businesses), some businesses are inclined to market items (e.g., braised chicken rice, spicy soup, etc.), some businesses are inclined to market dishes (e.g., fast food), and so forth. At present, when information of POIs is constructed, various types of information are generally mixed together, so that a merchant is difficult to accurately represent, and more noise is brought.

In practical application, the number of short Query inputs by a user in a search scene such as a take-out platform is large, and the user clicks POIs such as a merchant after inputting the Query, which is equivalent to establishing a matching relationship between the POIs and the Query, wherein the Query not only contains finer POI categories, but also has other information such as brands, tastes and the like; therefore, compared with information such as category data of POIs, the historical clicking Query can more finely characterize the POIs and reflect the diversity of the information.

Therefore, in the embodiment of the invention, the portrait tag sequence of the POI is constructed by introducing the historical click Query sequence of each POI and the like so as to accurately describe the merchant. Where a historical click Query of a POI may be understood as clicking on the POI after the user enters the Query, then the Query may be understood as a historical click Query of the POI.

Specifically, for any target POI, the history CTR of the target POI under each query word condition may be obtained according to the history query record, and N query words with the highest history CTR of the target POI are obtained, so as to construct a first tag sequence of the target POI. The history query records may include history query records of a plurality of different users, where the users may include target users or may not include target users, which is not limited in this embodiment of the present invention. Further, the history query record may be understood as record data related to CTR, such as query words input by each query, each POI exposed in a query result fed back for each query, and click behavior records, next behavior records, and the like of each POI in a specified history period of time of the user.

The obtaining mode of the history CTR of the target POI under the condition of the query word can be obtained according to the exposure times and the clicking times of the target POI under the condition of the query word. For example, a history ctr= (Click (target POI, Q) +a)/(PV (target POI, Q) +b) of the target POI under the condition of the query word may be set, where Click (target POI, Q) represents a historical Click number of the target POI under the condition of the query word Q in a preset time period in the past, PV (target POI, Q) represents a historical exposure number of the target POI under the condition of the query word Q in the preset time period in the past, and a and b are smoothing factors, and specific values thereof may be set in a customized manner according to requirements, which is not limited to the embodiment of the present invention. In addition, the number of query words included in the first tag sequence, that is, the value of N, may also be set in a customized manner according to the requirement, which is not limited in this embodiment of the present invention. For example, the value of N may be set to 50, and so on.

In addition, when the first tag sequence of the target POI is constructed, the history CTR of the target POI under the condition of each Query word can be obtained based on the history click Query of each user for the target POI, which can be obtained in the platform.

And the above-described query terms may be understood as all query terms input by the user each time a search is requested. For example, assuming that the user requests a search by triggering the search control after entering "afternoon tea", "dessert" in the search box during the search, the query term corresponding to the current search may be a combination of "afternoon tea" and "dessert", such as "afternoon tea dessert", or "afternoon tea dessert", and so on.

And further, according to the CTR related parameters and real-time query words of the target user, the CTR estimated value of the target user for the corresponding target POI can be obtained through a CTR estimated model. Wherein, the CTR related parameter at least comprises a first tag sequence of the corresponding target POI. In addition, the CTR-related parameters may include any other available parameters related to estimating the current CTR of the target user for the corresponding target POI, such as historical behavior information of the corresponding user expressing the long-term interest of the user (e.g., target POI clicked by the user in the past, target POI ordered, target POI viewed, query words entered), other description information of the target POI (e.g., target POI name, menu information, belonging category, business district, etc.), and so forth.

Correspondingly, the CTR estimation model can be obtained in advance through training a plurality of target POI samples with known CTR true values and CTR related parameters. The process of obtaining the CTR true value and the CTR-related parameter of the target POI sample may refer to the above-mentioned process of obtaining the history CTR and the CTR-related parameter of the target POI, which is not described herein.

Referring to fig. 2, in another embodiment, step 120 may further include:

step 121, for each POI included in each historical order of the target user, acquiring a historical CTR of the POI under each query word condition according to a historical query record, acquiring M query words with highest CTR in the query words corresponding to the target user, and constructing a second tag sequence of the target user;

step 122, according to the relevant parameters of the CTR and the real-time query words of the target user, acquiring a CTR estimated value of the target user aiming at the target POI through the CTR estimated model; the CTR-related parameter includes at least the first tag sequence and the second tag sequence.

In the search scene, the phenomenon that the user repeatedly selects the same POI is obvious, and meanwhile, the preference of different users for the POI is also diversified. Taking take out as an example, the re-purchase phenomenon of the merchant is obvious, and meanwhile, the preferences of different users for the merchant are also diversified, for example, some users are brand preference type, some users are taste preference type, and some users are dish preference type. How to exploit the buyback information to mine the user's preferences is critical.

Moreover, the information important to the user in the search scene has two parts: firstly, expressing the current intention of a user by the instant Query of the user; and secondly, historical behavior information of the user expresses long-term interests of the user. How to model the matching of the instant Query of the user and the historical behavior information of the user with the POI at the same time is also an important problem which can be optimized. The single data is more fully inclusive of the user's interests in the user's history. Therefore, in the embodiment of the invention, the query word of the POI corresponding to the historical order of the target user can be introduced to construct the portrait tag sequence of the target user, namely the second tag sequence, so as to enrich the information expression of the user. For example, a tag sequence for a user represented by Query may be mined using a historical click Query sequence for POIs contained in a historical order of half a year of the user's history.

Specifically, according to historical behavior data of a target user, acquiring a historical order of the target user and POIs (point of interest) contained in each historical order of the target user, further aiming at each POI contained in each historical order of the target user, acquiring a historical CTR (control point) of the POI under each query word condition according to a historical query record, and acquiring M highest query words of CTR in query words corresponding to the target user according to the historical CTR of the POI under each query word condition, so as to construct a second tag sequence of the target user; the specific manner of obtaining the history CTR of the POI under each query term according to the history query record may be similar to the foregoing history CTR of obtaining the target POI under each query term according to the history query record, which is not described herein. The query terms corresponding to the target user may be understood as query terms related to the historical query records of each POI included in each historical order of the target user, and may include query terms input by query records of a plurality of users for the POI, instead of just query terms input by query records of the target user for the POI. Of course, in the embodiment of the present invention, it may also be provided that the query term corresponding to the target user only includes the query term input by the query record of the target user for the POI, which is not limited to the embodiment of the present invention.

The historical behavior data of the target user may be understood as behavior data of the target user in a specified period of time before the current time, for example, behavior data in the past half year, and the like.

Based on the historical behavior data of the target user, the historical orders of the target user in the corresponding past half years and POIs corresponding to each historical order can be obtained, and based on the historical query record of each user, query words input by each user for the POIs can be obtained.

For example, assuming that the historical order 1 of the target user includes POI1, that is, the target user places an order for the commodity in POI1 and generates order 1, further, the query word corresponding to POI1 may be obtained according to the historical query record, and the history CTR of POI1 under each query word condition, where the query word corresponding to POI1 may be understood as the query word corresponding to the target user. In addition, in the embodiment of the present invention, when the second tag sequence of the target user is constructed, query words with higher CTR are required to be obtained, and if all query words corresponding to each POI included in each historical order are obtained, the data size may be larger, and the query words with smaller CTR values may not affect the second tag sequence, so that for each POI included in each historical order, only the L query words with highest CTR corresponding to the POI may be obtained and used as the query words corresponding to the target user, where L is a positive integer, and the value of L may be set in a customized manner according to the requirement.

Moreover, since the target user corresponds to a plurality of historical orders, and POIs contained in different historical orders may correspond to the same query word, at this time, the historical CTRs of the same query word under different orders may be added or averaged, or the maximum value of the historical CTRs of the same query word under different orders may be directly taken as the CTR of the query word corresponding to the target user.

And then, according to the CTR value of each query corresponding to the target user, obtaining M highest query words in the query words corresponding to the target user, and constructing a second tag sequence of the target user. Wherein the M query terms may be sequentially arranged in the second tag sequence in order of CTR values from high to low.

After a first tag sequence representing a target POI portrait and a second tag sequence representing a target user portrait are obtained, a CTR estimated value of the target user for the target POI can be obtained through the CTR estimated model according to CTR related parameters and real-time query words of the target user; the CTR-related parameters at this time may include at least the first tag sequence and the second tag sequence.

Optionally, in an embodiment of the present invention, the step 121 may further include:

Step 1211, obtaining a target POI and a history query word corresponding to each history order of the target user;

step 1212, for any one of the history orders, based on the history CTR of each history query word corresponding to the history order, obtaining the weight of each history query word corresponding to the history order, and ensuring that the weight and value of the history query word under the same history order are specified values;

step 1213, adding weights of the same historical query term under each historical order to obtain a CTR of the historical query term corresponding to the target user;

step 1214, obtaining M query words with highest CTR from the historical query words of the target user, and constructing a second tag sequence of the target user.

The specific value of the specified numerical value can be set in a self-defined manner according to the requirement, and the embodiment of the invention is not limited. For example, a specified value of 1 may be set, and then Query within each historical order may be normalized to its history CTR as a weight. The Query within the historical order may be understood as the Query term corresponding to each POI contained in the historical order.

For example, assume that the historical behavior data of the target user includes a historical order 1, a historical order 2 and a historical order 3, where the historical order 1 includes POI1, the historical order 2 includes POI2, the historical order 3 includes POI3, query words corresponding to POI1, POI2 and POI3 are obtained according to the historical query records, and the history CTR under each query word condition is as follows:

Assuming the specified value at this time is a, the weight of each query term placed in each of the historical orders may further be determined based on the specified value.

Taking historical order 1 as an example, the weight w1 of Query1 placed in the order may be CTR 1/(ctr1+ctr2+ctr3) a, the weight w2 of Query2 may be CTR 2/(ctr1+ctr2+ctr3) a, and the weight w3 of Query3 may be CTR 3/(ctr1+ctr2+ctr3) a;

accordingly, it may be obtained that under the historical order 2, the Query2 weight w4 may be CTR 2/(ctr2+ctr3+ctr5) a, and under the historical order 3, the Query2 weight w7 may be CTR 2/(ctr2+ctr6+ctr7) a.

Further, the CTR of the Query word Query2 corresponding to the target user is w2+w4+w7. Therefore, based on the CTR of the query words corresponding to each target user, M query words with the highest CTR can be obtained, and a second tag sequence of the target user can be constructed.

Optionally, in the embodiment of the present invention, the CTR estimation model is a combination of a dual-sequence modeling sub-network model and a CTR estimation sub-network model;

accordingly, the step 122 may further include:

step 1221, obtaining a feature vector of the first tag sequence and the second tag sequence after crossing through the dual-sequence modeling sub-network model;

Step 1222, according to the feature vector, the real-time query word of the target user and other CTR related parameters, acquiring a CTR predicted value of the target user for the target POI under the condition of the real-time query word through a CTR predicted sub-network model; wherein the other CTR-related parameters are other parameters than the first tag sequence and the second tag sequence in the CTR-related parameters.

The construction scheme of the first tag sequence and the second tag sequence is given above. In order to achieve the purpose of simultaneously modeling the matching between the instant Query of the user and the historical behavior information of the user and the candidate target POIs, the embodiment of the invention also provides a double-sequence matching modeling scheme. That is, the CTR estimation model is set as a combination of a dual-sequence modeling sub-network model and a CTR estimation sub-network model. The dual-sequence modeling sub-network model can be used for fusing the second tag sequence of the user and the first tag sequence of the target POI so as to extract the cross information among the Query in the sequence.

Then, the feature vector of the first tag sequence and the second tag sequence after crossing can be obtained by using the dual-sequence modeling sub-network model at this time, that is, the first tag sequence and the second tag sequence are input into the dual-sequence modeling sub-network model at the same time, so as to learn the crossing information between the first tag sequence at the target POI side and the second tag sequence at the user side, and output the feature vector representing the crossing information. Further, according to the feature vector, the real-time query word of the target user and other CTR related parameters, a CTR predicted value of the target user for the target POI under the condition of the real-time query word can be obtained through a CTR predicted sub-network model.

The dual-sequence modeling sub-network model and the CTR estimation sub-network model may be any available machine learning model, which is not limited in this embodiment of the present invention.

For example, the Self-Attention structure is utilized to extract the cross information between Query in the sequence, so that the dual-sequence modeling sub-network model can be set as the Self-Attention structure, and at least one Self-Attention structure corresponding to the user side and the target POI side can be set in the dual-sequence modeling sub-network model, which is not limited to the embodiment of the invention. Accordingly, the CTR estimate subnetwork model may be set to DNN (Deep NeuralNetworks, deep neural network) network architecture, and so on.

Optionally, in an embodiment of the present invention, the dual-sequence modeling sub-network model includes an embedded layer, an attention mechanism layer, and a Pooling (Pooling) layer that are cascaded in sequence;

accordingly, the step 1221 may further include:

s1, acquiring a first matrix representation of the real-time query word through a first embedding layer, and simultaneously acquiring a second matrix representation of the first tag sequence and a third matrix representation of the second tag sequence through a second embedding layer respectively;

S2, representing the input of a K matrix in an attention mechanism layer of a replacement user side through the first matrix to obtain an input matrix of the attention mechanism layer;

s3, obtaining an output vector of the attention mechanism layer, inputting the output vector into the pooling layer, and obtaining the feature vector after crossing through the pooling layer.

In constructing the dual sequence modeling sub-network model, in order to fully learn the intersection information between the target POI and the user, the following two aspects can be considered:

1. the first tag sequences of the target POI side are all centered on the target POI, so that the first tag sequences can be crossed inside to extract multi-dimensional information of the multi-dimensional target POI; meanwhile, the method can cross with the instant Query (namely real-time Query word) of the user so as to extract target POI information matched with the current intention of the user;

2. the second tag sequence at the user side has wider interests, and the first tag sequence at the merchant side and the user instant Query can be used as a context for crossing so as to extract the user interests which are most matched with the current intention of the user and the candidate target POI. In addition, unlike the first tag sequence on the merchant side, the second tag sequence on the user side may have query words that are far away because of the large variation range of interests or demands of the user, so that the internal intersection of the sequences needs to be avoided to reduce noise.

Therefore, in the embodiment of the invention, by improving the Self-Attention structure, K in the Attention mechanism layer (for example, the Self-Attention structure) corresponding to the second tag sequence at the user side is replaced by the real-time query word of the target user by the second tag sequence at the user side, and Q and V are still unchanged, so that the following purposes are achieved:

1. introducing real-time query words of users, and ensuring that sequences at two sides are crossed with the real-time query words;

2. the second tag sequence at the user side is prevented from being crossed internally, and the tag sequences at the user side and the target POI side are ensured to be crossed;

3. the first tag sequence on the merchant side is internally crossed.

Specifically, in order to complete the above-mentioned intersecting process, matrix representations of the real-time query word, the first tag sequence and the second tag sequence may also be obtained through an Embedding (Embedding) layer, respectively, and since the length of the real-time query word of the target user is significantly smaller than that of the first tag sequence and the second tag sequence, different Embedding layers may be adopted. The first matrix representation of the real-time query word can be acquired through the first embedding layer, and the second matrix representation of the first tag sequence and the third matrix representation of the second tag sequence can be acquired through the second embedding layer. In addition, in the embodiment of the present invention, the enhancement layers corresponding to the target POI side and the user side may also be set respectively, so that the second matrix representation of the first tag sequence and the third matrix representation of the second tag sequence are obtained through the enhancement layers corresponding to the user side and the target POI side respectively, which is not limited to the embodiment of the present invention.

Accordingly, the dual sequence modeling sub-network model may be configured to include attention mechanism layers corresponding to the user side and the target POI side, respectively, and then after the third matrix representation of the second tag sequence on the user side is obtained through the embedding layer, the third matrix representation of the first tag sequence on the user side may be used as input to the attention mechanism layer on the user side, and after the second matrix representation of the first tag sequence on the target POI side is obtained through the embedding layer, the third matrix representation of the second tag sequence on the user side may be used as input to the attention mechanism layer on the target POI side.

Before the third matrix is input into the attention mechanism layer of the user side, the input of the K matrix in the attention mechanism layer of the user side can be replaced through the first matrix representation, and Q and V remain unchanged, so that the input matrix of the attention mechanism layer of the user side is obtained, the internal crossing of the second label sequence of the user side is avoided, and meanwhile, the crossing of the sequences at both sides and the real-time query word is ensured.

And further, the output vectors of the attention mechanism layers at the user side and the target POI side can be obtained, the pooled layers are input, and the feature vectors after crossing are obtained through the pooled layers. Then, the CTR estimated sub-network model may obtain the CTR estimated value of the target user for the corresponding target POI under the current real-time query term condition according to the feature vector and other CTR related parameters.

Optionally, in an embodiment of the present invention, the embedded layer includes any one of a Word embedding layer, a Position Embeddding layer, and a Segment Embedding layer, or a combination of any plurality of these layers; the attention mechanism layer comprises a multi-head attention mechanism; the pooling layer includes an average pooling layer.

FIG. 3A is a schematic diagram of a dual-sequence modeling sub-network model. It can be divided into three parts, namely an Embedding Layer, a Multi-head self-attention fusion Layer (Multi-head self-attention mechanism joint Layer) and a Pooling Layer. The concrete explanation is as follows:

1. the embedded Layer serves as an input Layer of the dual-sequence modeling sub-network model, and includes Position Embedding and Field embedded (i.e., segment Embedding) to represent inputs in addition to embedded (i.e., word embedded) for each Item (Query) in the sequence. That is, the result E of the Layer of the coding of the ith query term in the tag sequence _i ＝PE(pos，i)+FE(field _i ) +IE (i), wherein PE (pos, i) represents the Position Embedding result of the ith query term in the tag sequence, pos represents the CTR weight of the ith query term in the tag sequence, FE (field _i ) Field results, field, representing the i-th query term in the tag sequence _i Representing the sequence domain to which the current query word belongs (first tag training/second tag sequence), each field having its id number, e.g. field_id e [0,1]IE (i) represents Word encoding results of the i-th query term in the tag sequence.

1.1 Position Embedding is used to identify position information in the input sequence, representing weight information for each Query in the sequence. The operation process of the positioning module can be as follows:

wherein w_ctr, i.e. pos as described above, is the input sequenceThe weight of the ith Query (i.e., CTR referenced by the Query term is determined when constructing the tag sequence), i being the dimensions i, d of the currently input Query in the input tag sequence _model For Position Embedding layer output dimension, d _model The value of (2) can be set in a self-defined way according to the requirement, and the embodiment of the invention is not limited.

1.2 Field editing). Different sequence domains (target POI side tag sequences/user side tag sequences) are given to different embedded vectors, specifically, the embedded vectors corresponding to the different sequence domains can be set in a self-defined manner according to requirements, and the embodiment of the invention is not limited.

PositionEmbedding

2、Multi-head self-attention fusion layer

2.1 Processing the first tag sequence, the second tag sequence and the real-time query word by an Embedding Layer to obtain E _u 、E _p 、E _o The method comprises the steps of respectively representing an Embedding result of a user side tag sequence, an Embedding result of a merchant side tag sequence and an Embedding result of a real-time query word;

2.2)Multi-head Self-Attention：

respectively generating a parameter matrix W of Q, K and V in each Self-attribute structure _Q 、W _k 、W _v The values of the parameter matrix can be obtained according to requirements or through model training, and the embodiment of the invention is not limited.

Projecting the input to the Q, K and V matrixes, and replacing the input of the K matrix in the attention mechanism layer at the user side with an Embedding result E of the real-time query word _o Finally, inputting Q, K and V obtained by splicing;

Q _u ＝E _u W _Q ；Q _p ＝E _p W _Q

K _u ＝E _o W _k ；K _p ＝E _p W _k

V _u ＝E _u W _v ；V _p ＝E _p W _v

Q＝Concat(Q _u ，Q _p )；K＝Concat(K _u ，K _p )；V＝Concat(V _u ，V _p )

wherein Q is _u 、K _u 、V _u Representing in turn the results of the projection of the input of the user-side attention mechanism layer onto its Q, K, V matrix, Q _p 、K _p 、V _p In turn representing the result of the input of the target POI side attention mechanism layer projected onto its Q, K, V matrix, respectively.

The course of the attention mechanism can be as follows:

if the attention mechanism layer includes a multi-Head attention mechanism, then it is assumed that for any Head _i ＝Attention(Q _i 、K _i 、V _i ) The result of the multi-head attention mechanism is multi head=concat (head) ₁ ，head ₂ ，…，head ₂ )W _o Wherein W is _o 、W _Q 、W _k 、W _v Are parameters that can be learned by training.

Finally, vectors can be obtained through a MeanPooling mode and input into a CTR estimated subnetwork model.

Fig. 3B is a schematic structural diagram of a CTR estimation model. The CTR estimated sub-network model structure can be DNN, CNN and the like, and the double-sequence modeling sub-network model can be used as a part of the CTR estimated model to participate in end-to-end training when the model training is carried out.

In addition, in the embodiment of the invention, when the user searches each time, the CTR value of each candidate target POI can be estimated in real time through the mode, and then each candidate target POI can be sequenced according to the real-time CRT predicted value from big to small, so that the adaptation degree of the sequencing result and the user searching requirement is improved.

According to the method for predicting the CTR by constructing the tag sequence of the target POI, the tag sequence of the user is constructed, and the structure of the CTR prediction model is improved on the basis of the method for predicting the CTR by constructing the tag sequence of the target POI, so that the purposes of simultaneously modeling the matching between the real-time Query of the user and the historical behavior information of the user and the candidate target POI in a search scene are achieved, and the accuracy of the CTR prediction result is improved.

Referring to fig. 4, a flowchart illustrating steps of another CTR estimation processing method according to an embodiment of the present invention is shown.

Step 210, for each POI included in each historical order of the target user, acquiring a historical CTR of the POI under each query word condition according to a historical query record, acquiring M query words with highest CTR in the query words corresponding to the target user, and constructing a second tag sequence of the target user;

step 220, for any target POI, according to CTR related parameters and real-time query words of the target user, acquiring a CTR predicted value of the target user for the target POI through a CTR predicted model; wherein the CTR-related parameter comprises at least the second tag sequence.

Optionally, in an embodiment of the present invention, the step 220 may further include:

step 221, for any target POI, acquiring a history CTR of the target POI under each query word condition according to a history query record, and acquiring N query words with the highest history CTR of the target POI, and constructing a first tag sequence of the target POI;

step 222, according to the relevant parameters of the CTR and the real-time query words of the target user, acquiring a CTR estimated value of the target user aiming at the target POI through the CTR estimated model; wherein the CTR-related parameter includes at least the first tag sequence and the second tag sequence.

Optionally, in an embodiment of the present invention, the CTR estimation model is a combination of a dual-sequence modeling sub-network model and a CTR estimation sub-network model, and the step 222 may further include:

step 2221, obtaining, by the dual-sequence modeling sub-network model, a feature vector of the first tag sequence and the second tag sequence after crossing;

step 2222, according to the feature vector, the real-time query word of the target user and other CTR related parameters, obtaining a CTR predicted value of the target user for the target POI under the condition of the real-time query word through a CTR predicted sub-network model;

In the embodiment of the invention, the second tag sequence of the target user can be considered when CTR estimation is performed so as to enrich the information expression of the target user, thereby improving the accuracy of CTR estimation results.

The method for obtaining the second tag sequence and subsequently performing CTR estimation by combining the second tag sequence of the target user and the first tag sequence of the target POI at the same time may refer to the above embodiments, and details of the setting of the CTR estimation model are not described herein.

Referring to fig. 5, a schematic structural diagram of a CTR estimation processing device according to an embodiment of the present invention is shown.

The CTR estimation processing device of the embodiment of the invention comprises: a first tag sequence construction module 310 and a first CTR estimation module 320.

The functions of the modules and the interaction relationship between the modules are described in detail below.

The first tag sequence construction module 310 is configured to, for any target POI, obtain, according to a history query record, a history CTR of the target POI under each query word condition, and obtain N query words with highest history CTR of the target POI, and construct a first tag sequence of the target POI;

the first CTR estimation module 320 is configured to obtain, according to the CTR related parameter and the real-time query word of the target user, a CTR estimated value of the target user for the target POI through a CTR estimation model;

Optionally, the first CTR estimation module includes:

Optionally, the second tag sequence constructs a sub-module, specifically for:

the information crossing unit is specifically configured to:

Referring to fig. 6, a schematic structural diagram of another CTR estimation processing apparatus according to an embodiment of the present invention is shown.

The CTR estimation processing device of the embodiment of the invention comprises: a second tag sequence construction module 410 and a second CTR estimation module 420.

A second tag sequence construction module 410, configured to obtain, for each POI included in each historical order of the target user, a historical CTR of the POI under each query term according to a historical query record, obtain M query terms with highest CTR among query terms corresponding to the target user, and construct a second tag sequence of the target user;

the second CTR estimation module 420 is configured to obtain, for any target POI, a CTR estimated value of the target user for the target POI through a CTR estimation model according to a CTR related parameter and a real-time query word of the target user;

Wherein the CTR-related parameter comprises at least the second tag sequence.

Optionally, the second CTR estimation module includes:

The CTR estimation processing device provided in the embodiment of the present invention can implement each process implemented in the method embodiments of fig. 1 to 4, and in order to avoid repetition, a description is omitted here.

Preferably, the embodiment of the present invention further provides an electronic device, including: the processor, the memory, store on the memory and can be on the computer program of the operation of processor, this computer program realizes each process of the above-mentioned CTR estimated processing method embodiment when being carried out by the processor, and can reach the same technical result, in order to avoid repetition, will not be repeated here.

The embodiment of the invention also provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, realizes the processes of the CTR estimation processing method embodiment, and can achieve the same technical effects, and in order to avoid repetition, the description is omitted here. Wherein the computer readable storage medium is selected from Read-Only Memory (ROM), random access Memory (RandomAccess Memory, RAM), magnetic disk or optical disk.

Fig. 7 is a schematic diagram of a hardware structure of an electronic device implementing various embodiments of the present invention.

The electronic device 500 includes, but is not limited to: radio frequency unit 501, network module 502, audio output unit 503, input unit 504, sensor 505, display unit 506, user input unit 507, interface unit 508, memory 509, processor 510, and power source 511. It will be appreciated by those skilled in the art that the electronic device structure shown in fig. 7 is not limiting of the electronic device and that the electronic device may include more or fewer components than shown, or may combine certain components, or a different arrangement of components. In the embodiment of the invention, the electronic equipment comprises, but is not limited to, a mobile phone, a tablet computer, a notebook computer, a palm computer, a vehicle-mounted terminal, a wearable device, a pedometer and the like.

It should be understood that, in the embodiment of the present invention, the radio frequency unit 501 may be used to receive and send information or signals during a call, specifically, receive downlink data from a base station, and then process the downlink data with the processor 510; and, the uplink data is transmitted to the base station. Typically, the radio frequency unit 501 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier, a duplexer, and the like. In addition, the radio frequency unit 501 may also communicate with networks and other devices through a wireless communication system.

The electronic device provides wireless broadband internet access to the user through the network module 502, such as helping the user to send and receive e-mail, browse web pages, access streaming media, and the like.

The audio output unit 503 may convert audio data received by the radio frequency unit 501 or the network module 502 or stored in the memory 509 into an audio signal and output as sound. Also, the audio output unit 503 may also provide audio output (e.g., a call signal reception sound, a message reception sound, etc.) related to a specific function performed by the electronic device 500. The audio output unit 503 includes a speaker, a buzzer, a receiver, and the like.

The input unit 504 is used for receiving an audio or video signal. The input unit 504 may include a graphics processor (Graphics Processing Unit, GPU) 5041 and a microphone 5042, the graphics processor 5041 processing image data of still pictures or video obtained by an image capturing device (e.g., a camera) in a video capturing mode or an image capturing mode. The processed image frames may be displayed on the display unit 506. The image frames processed by the graphics processor 5041 may be stored in the memory 509 (or other storage medium) or transmitted via the radio frequency unit 501 or the network module 502. Microphone 5042 may receive sound and may be capable of processing such sound into audio data. The processed audio data may be converted into a format output that can be transmitted to the mobile communication base station via the radio frequency unit 501 in case of a phone call mode.

The electronic device 500 also includes at least one sensor 505, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor includes an ambient light sensor that can adjust the brightness of the display panel 5061 according to the brightness of ambient light, and a proximity sensor that can turn off the display panel 5061 and/or the backlight when the electronic device 500 is moved to the ear. As one of the motion sensors, the accelerometer sensor can detect the acceleration in all directions (generally three axes), and can detect the gravity and direction when stationary, and can be used for recognizing the gesture of the electronic equipment (such as horizontal and vertical screen switching, related games, magnetometer gesture calibration), vibration recognition related functions (such as pedometer and knocking), and the like; the sensor 505 may further include a fingerprint sensor, a pressure sensor, an iris sensor, a molecular sensor, a gyroscope, a barometer, a hygrometer, a thermometer, an infrared sensor, etc., which are not described herein.

The display unit 506 is used to display information input by a user or information provided to the user. The display unit 506 may include a display panel 5061, and the display panel 5061 may be configured in the form of a liquid crystal display (Liquid Crystal Display, LCD), an Organic Light-Emitting Diode (OLED), or the like.

The user input unit 507 is operable to receive input numeric or character information and to generate key signal inputs related to user settings and function control of the electronic device. Specifically, the user input unit 507 includes a touch panel 5071 and other input devices 5072. Touch panel 5071, also referred to as a touch screen, may collect touch operations thereon or thereabout by a user (e.g., operations of the user on touch panel 5071 or thereabout using any suitable object or accessory such as a finger, stylus, etc.). Touch panel 5071 may include two parts, a touch detection device and a touch controller. The touch detection device detects the touch azimuth of a user, detects a signal brought by touch operation and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device, converts the touch information into touch point coordinates, sends the touch point coordinates to the processor 510, and receives and executes commands sent by the processor 510. In addition, the touch panel 5071 may be implemented in various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch panel 5071, the user input unit 507 may include other input devices 5072. In particular, other input devices 5072 may include, but are not limited to, physical keyboards, function keys (e.g., volume control keys, switch keys, etc.), trackballs, mice, joysticks, and so forth, which are not described in detail herein.

Further, the touch panel 5071 may be overlaid on the display panel 5061, and when the touch panel 5071 detects a touch operation thereon or thereabout, the touch operation is transmitted to the processor 510 to determine a type of touch event, and then the processor 510 provides a corresponding visual output on the display panel 5061 according to the type of touch event. Although in fig. 7, the touch panel 5071 and the display panel 5061 are two independent components for implementing the input and output functions of the electronic device, in some embodiments, the touch panel 5071 and the display panel 5061 may be integrated to implement the input and output functions of the electronic device, which is not limited herein.

The interface unit 508 is an interface for connecting an external device to the electronic apparatus 500. For example, the external devices may include a wired or wireless headset port, an external power (or battery charger) port, a wired or wireless data port, a memory card port, a port for connecting a device having an identification module, an audio input/output (I/O) port, a video I/O port, an earphone port, and the like. The interface unit 508 may be used to receive input (e.g., data information, power, etc.) from an external device and transmit the received input to one or more elements within the electronic apparatus 500 or may be used to transmit data between the electronic apparatus 500 and an external device.

The memory 509 may be used to store software programs as well as various data. The memory 509 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data (such as audio data, phonebook, etc.) created according to the use of the handset, etc. In addition, the memory 509 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device.

The processor 510 is a control center of the electronic device, connects various parts of the entire electronic device using various interfaces and lines, and performs various functions of the electronic device and processes data by running or executing software programs and/or modules stored in the memory 509, and calling data stored in the memory 509, thereby performing overall monitoring of the electronic device. Processor 510 may include one or more processing units; preferably, the processor 510 may integrate an application processor that primarily handles operating systems, user interfaces, applications, etc., with a modem processor that primarily handles wireless communications. It will be appreciated that the modem processor described above may not be integrated into the processor 510.

The electronic device 500 may also include a power supply 511 (e.g., a battery) for powering the various components, and preferably the power supply 511 may be logically connected to the processor 510 via a power management system that performs functions such as managing charging, discharging, and power consumption.

In addition, the electronic device 500 includes some functional modules, which are not shown, and will not be described herein.

It should be noted that, in this document, 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 one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) comprising instructions for causing a terminal (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method according to the embodiments of the present invention.

The embodiments of the present invention have been described above with reference to the accompanying drawings, but the present invention is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and many forms may be made by those having ordinary skill in the art without departing from the spirit of the present invention and the scope of the claims, which are to be protected by the present invention.

Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, and are not repeated herein.

In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.

The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk, etc.

The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.

Claims

1. The CTR estimation processing method is characterized by comprising the following steps of:

aiming at each POI contained in each historical order of a target user, acquiring a historical CTR of the POI under the condition of each query word according to a historical query record, acquiring M highest query words in the query words corresponding to the target user, and constructing a second tag sequence of the target user;

according to the CTR related parameters and real-time query words of a target user, a CTR estimated value of the target user for the target POI is obtained through a CTR estimated model; wherein the CTR related parameters at least comprise the first tag sequence and the second tag sequence, and the CTR estimation model is a combination of a dual-sequence modeling sub-network model and a CTR estimation sub-network model;

the acquiring the CTR predicted value of the target user aiming at the target POI through a CTR pre-estimation model according to the CTR related parameters and the real-time query words of the target user comprises the following steps:

2. The method according to claim 1, wherein the step of obtaining the history CTR of the POI under each query term according to the history query record, obtaining M query terms with highest CTR among the query terms corresponding to the target user, and constructing the second tag sequence of the target user includes:

3. The method of claim 2, wherein the dual sequence modeling sub-network model comprises an embedded layer, an attention mechanism layer, a pooling layer, which are cascaded in sequence;

4. The method of claim 3, wherein the embedded layer comprises any one of a Word embedding layer, a Position Embeddding layer, a Segment Embedding layer, or a combination of any plurality thereof; the attention mechanism layer comprises a multi-head attention mechanism; the pooling layer includes an average pooling layer.

5. A CTR estimate processing apparatus, comprising:

the first CTR estimation module includes:

the second tag sequence construction sub-module is used for acquiring a historical CTR of each POI under the condition of each query word according to a historical query record aiming at each POI contained in each historical order of a target user, acquiring M query words with highest CTR in the query words corresponding to the target user and constructing a second tag sequence of the target user;

the first CTR estimation sub-module is used for acquiring a CTR estimated value of the target user aiming at the target POI through a CTR estimation model according to CTR related parameters and real-time query words of the target user;

Wherein the CTR-related parameter comprises at least the first tag sequence and the second tag sequence; the CTR estimation model is a combination of a dual-sequence modeling sub-network model and a CTR estimation sub-network model,

6. An electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, which when executed by the processor, implements the steps of the CTR estimation processing method according to any one of claims 1 to 4.

7. A computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, which when executed by a processor, implements the steps of the CTR estimation processing method according to any one of claims 1 to 4.