Click rate estimation method and device, storage medium and electronic device

Abstract

Embodiments of the present application disclose a click rate estimation method and device, a storage medium and an electronic device. The method comprises: obtaining user search text input by a user, performing feature extraction processing based on the user search text to obtain semantic feature data, discrete feature data and statistical feature data; performing click rate estimation processing on promotion information of a candidate recommended word based on the semantic feature data, the discrete feature data and the statistical feature data by using a click rate estimation model to obtain the click rate of the promotion information corresponding to the candidate recommended word.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a click-through rate prediction method, apparatus, storage medium, and electronic device. Background Technology

[0002] In related technologies, search advertising is a common form of online advertising. Search engines can display search ads and organic search results together. As a bridge between advertisers and users, search engines can provide users with a personalized experience while helping advertisers expand their brand, promote products, and increase sales. Click-through rate (CTR) is a crucial metric in online advertising systems. By estimating CTR, search engines can adjust the user search experience based on the predicted CTR, while also ensuring advertisers have ample ad exposure opportunities. Summary of the Invention

[0003] This application provides a click-through rate (CTR) prediction method, apparatus, computer storage medium, and electronic device. The technical solution is as follows:

[0004] In a first aspect, embodiments of this application provide a click-through rate (CTR) prediction method, the method comprising:

[0005] Obtain the user's search text input by the user, and perform feature extraction processing based on the user's search text to obtain semantic feature data, discrete feature data, and statistical feature data;

[0006] Based on the semantic feature data, the discrete feature data, and the statistical feature data, a click-through rate (CTR) prediction model is used to predict the CTR of promotional information for candidate recommended words, thereby obtaining the CTR of promotional information corresponding to the candidate recommended words.

[0007] In some possible implementations, the step of extracting semantic feature data, discrete feature data, and statistical feature data based on the user's search text includes:

[0008] Semantic feature data is obtained by performing semantic feature extraction processing on the user's search text.

[0009] Discrete feature data is obtained by performing discrete feature query processing based on the semantic feature data;

[0010] Statistical feature data is obtained by performing feature calculations based on the semantic feature data.

[0011] In some possible implementations, the semantic feature extraction process based on the user's search text to obtain semantic feature data includes:

[0012] Search terms are obtained by extracting search terms from the user's search text, and candidate recommended terms are obtained by matching recommended terms with the search terms.

[0013] Obtain the set of historical click search terms corresponding to the candidate recommended terms;

[0014] Determine the target promotion information title corresponding to the candidate recommendation keywords;

[0015] Generate semantic feature data including the search term, the candidate recommended term, the historical click search term set, and the target promotion information title.

[0016] In some possible implementations, the discrete feature data obtained by performing discrete feature query processing based on the semantic feature data includes:

[0017] Obtain the information recommendation source mapping relationship, and query the recommendation source corresponding to the candidate recommendation word and the recommendation type corresponding to the candidate recommendation word based on the information recommendation source mapping relationship;

[0018] Obtain the search application identifier corresponding to the user's search text;

[0019] Generate discrete feature data including the recommendation source, the recommendation type, and the search application identifier.

[0020] In some possible implementations, the step of performing feature calculations based on the semantic feature data to obtain statistical feature data includes:

[0021] Similarity data is obtained by performing similarity calculations based on the search terms and the candidate recommended terms;

[0022] Determine the historical click-through rate and normalized click count corresponding to the candidate recommended word within a preset historical period, calculate the product of the normalized click count and the historical click-through rate, and obtain the normalized click-through rate data corresponding to the candidate recommended word;

[0023] Generate statistical feature data including the similarity data and the normalized click data.

[0024] In some possible implementations, the step of using a click-through rate (CTR) prediction model based on the semantic feature data, the discrete feature data, and the statistical feature data to predict the CTR of promotional information for candidate recommended terms, and obtaining the CTR of promotional information corresponding to the candidate recommended terms, includes:

[0025] The click-through rate (CTR) prediction model is used to perform CTR prediction processing on the semantic feature data, the discrete feature data, and the statistical feature data for the promotional information of the candidate recommendation words, so as to obtain the reference CTR of the promotional information corresponding to the candidate recommendation words.

[0026] The click-through rate prediction model is used to perform conversion rate prediction processing on the semantic feature data and the discrete feature data for the promotion information of the candidate recommendation words, so as to obtain the conversion rate of the promotion information corresponding to the candidate recommendation words.

[0027] The click-through rate (CTR) of the reference promotional information and the predicted promotional information conversion rate are calibrated using the CTR prediction model to obtain the CTR of the promotional information corresponding to the candidate recommendation keywords.

[0028] In some possible implementations, the step of performing click-through rate (CTR) prediction processing on the semantic feature data, the discrete feature data, and the statistical feature data for the candidate recommended terms, to obtain the reference CTR for the candidate recommended terms, includes:

[0029] The semantic feature data is subjected to semantic cross-processing to obtain a semantic cross-feature vector, and the discrete feature data is subjected to embedding processing to obtain a discrete feature vector;

[0030] Based on the semantic cross feature vector and the discrete feature vector, a vector concatenation process is performed to obtain a comprehensive feature vector;

[0031] The click-through rate (CTR) prediction model is used to predict the CTR of promotional information for candidate recommended terms by processing the comprehensive feature vector and the statistical feature data. The reference CTR of promotional information corresponding to the candidate recommended terms is then obtained.

[0032] In some possible implementations, the step of using the click-through rate prediction model to perform conversion rate prediction processing on the semantic feature data and the discrete feature data for the candidate recommended words, to obtain the conversion rate of the promotional information corresponding to the candidate recommended words, includes:

[0033] The comprehensive feature vector is processed by a click-through rate prediction model to predict the conversion rate of promotional information for candidate recommended words, thereby obtaining the conversion rate of promotional information corresponding to the candidate recommended words.

[0034] In some possible implementations, the step of performing click-through rate (CTR) calibration processing on the reference promotional information CTR and the estimated promotional information conversion rate using the CTR prediction model to obtain the CTR of the promotional information corresponding to the candidate recommendation keyword includes:

[0035] The click-through rate (CTR) of the candidate recommended keyword is obtained by multiplying the CTR of the reference promotional information and the conversion rate of the predicted promotional information using the CTR prediction model.

[0036] In some possible implementations, the method further includes:

[0037] Create an initial click-through rate (CTR) prediction model for scenarios involving promotional information click-through rate prediction;

[0038] Based on the sample user search text, feature extraction processing is performed to obtain sample semantic feature data, sample discrete feature data, and sample statistical feature data. The sample semantic feature data, the sample discrete feature data, and the sample statistical feature data are labeled with the click-through rate tags of the sample promotion information corresponding to the sample candidate recommendation words.

[0039] The initial click-through rate (CTR) prediction model is trained at least once using the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data. During the model training process, the initial CTR prediction model is used to perform CTR prediction processing on the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data for the promotional information of the sample candidate recommendation words, so as to obtain the predicted sample CTR of the promotional information corresponding to the sample candidate recommendation words. Based on the sample promotional information CTR tag and the predicted sample promotional information CTR, the model loss value is calculated. The model loss value is used to adjust the model parameters of the initial CTR prediction model to obtain the CTR prediction model after training.

[0040] In some possible implementations, the step of calculating the model loss value based on the sample promotional information click-through rate label and the predicted sample promotional information click-through rate includes:

[0041] The model loss value is obtained by calculating the click-through rate tag of the sample promotion information and the click-through rate of the predicted sample promotion information using the loss calculation formula.

[0042] The loss calculation formula satisfies the following formula:

[0043]

[0044] This represents the model loss value. Represents the crossover loss function. This indicates the click-through rate (CTR) tag for the sample promotional information. This indicates the click-through rate of the promotional information in the reference sample. This represents the click-through rate of the predicted sample promotional information. This indicates the predicted conversion rate of the sample promotional information. This represents the result of the logical AND operation between the click-through rate tag and the conversion rate tag of the sample promotional information. and This represents the network parameters of the initial click-through rate prediction model.

[0045] Secondly, embodiments of this application provide a click-through rate prediction device, the device comprising:

[0046] The feature extraction module is used to obtain the user search text input by the user, and perform feature extraction processing on the user search text to obtain semantic feature data, discrete feature data, and statistical feature data;

[0047] The model prediction module is used to perform click-through rate prediction processing on the promotional information of the candidate recommended words based on the semantic feature data, the discrete feature data, and the statistical feature data using a click-through rate prediction model, so as to obtain the click-through rate of the promotional information corresponding to the candidate recommended words.

[0048] Optionally, the feature extraction module includes:

[0049] The first extraction unit is used to perform semantic feature extraction processing based on the user's search text to obtain semantic feature data;

[0050] The second extraction unit is used to perform discrete feature query processing based on the semantic feature data to obtain discrete feature data;

[0051] The third extraction unit is used to perform feature calculation processing based on the semantic feature data to obtain statistical feature data.

[0052] Optionally, the first extraction unit is specifically used for:

[0053] Search terms are obtained by extracting search terms from the user's search text, and candidate recommended terms are obtained by matching recommended terms with the search terms.

[0054] Obtain the set of historical click search terms corresponding to the candidate recommended terms;

[0055] Determine the target promotion information title corresponding to the candidate recommendation keywords;

[0056] Generate semantic feature data including the search term, the candidate recommended term, the historical click search term set, and the target promotion information title.

[0057] Optionally, the second extraction unit is specifically used for:

[0058] Obtain the information recommendation source mapping relationship, and query the recommendation source corresponding to the candidate recommendation word and the recommendation type corresponding to the candidate recommendation word based on the information recommendation source mapping relationship;

[0059] Obtain the search application identifier corresponding to the user's search text;

[0060] Generate discrete feature data including the recommendation source, the recommendation type, and the search application identifier.

[0061] Optionally, the third extraction unit is specifically used for:

[0062] Similarity data is obtained by performing similarity calculations based on the search terms and the candidate recommended terms;

[0063] Determine the historical click-through rate and normalized click count corresponding to the candidate recommended word within a preset historical period, calculate the product of the normalized click count and the historical click-through rate, and obtain the normalized click-through rate data corresponding to the candidate recommended word;

[0064] Generate statistical feature data including the similarity data and the normalized click data.

[0065] Optionally, the model prediction module includes a first prediction unit, a second prediction unit, and a third prediction unit, specifically used for:

[0066] The first prediction unit is used to perform click-through rate prediction processing on the semantic feature data, the discrete feature data, and the statistical feature data for the promotional information of the candidate recommendation words through the click-through rate prediction model, so as to obtain the reference click-through rate of the promotional information corresponding to the candidate recommendation words;

[0067] The second prediction unit is used to perform conversion rate prediction processing on the semantic feature data and the discrete feature data for the candidate recommendation words through the click-through rate prediction model, so as to obtain the conversion rate of the promotion information corresponding to the candidate recommendation words.

[0068] The third prediction unit is used to perform click-through rate calibration processing on the click-through rate of the reference promotional information and the predicted conversion rate of the promotional information through the click-through rate prediction model, so as to obtain the click-through rate of the promotional information corresponding to the candidate recommendation keyword.

[0069] Optionally, the first prediction unit is specifically used for:

[0070] The semantic feature data is subjected to semantic cross-processing to obtain a semantic cross-feature vector, and the discrete feature data is subjected to embedding processing to obtain a discrete feature vector;

[0071] Based on the semantic cross feature vector and the discrete feature vector, a vector concatenation process is performed to obtain a comprehensive feature vector;

[0072] The click-through rate (CTR) prediction model is used to predict the CTR of promotional information for candidate recommended terms by processing the comprehensive feature vector and the statistical feature data. The reference CTR of promotional information corresponding to the candidate recommended terms is then obtained.

[0073] Optionally, the second prediction unit is specifically used for:

[0074] The comprehensive feature vector is processed by a click-through rate prediction model to predict the conversion rate of promotional information for candidate recommended words, thereby obtaining the conversion rate of promotional information corresponding to the candidate recommended words.

[0075] Optionally, the third prediction unit is specifically used for:

[0076] The click-through rate (CTR) of the candidate recommended keyword is obtained by multiplying the CTR of the reference promotional information and the conversion rate of the predicted promotional information using the CTR prediction model.

[0077] Optional click-through rate prediction devices also include:

[0078] The first training module is used to create an initial click-through rate (CTR) prediction model for the scenario of predicting CTR for promotional information.

[0079] The second training module is used to perform feature extraction processing based on sample user search text to obtain sample semantic feature data, sample discrete feature data, and sample statistical feature data, and to label the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data with the click-through rate tags of the sample promotion information corresponding to the sample candidate recommendation words.

[0080] The third training module is used to train the initial click-through rate (CTR) prediction model at least once using the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data. During the model training process, the initial CTR prediction model is used to perform CTR prediction processing on the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data for the promotional information of the sample candidate recommendation words, so as to obtain the predicted sample CTR of the promotional information corresponding to the sample candidate recommendation words. Based on the sample promotional information CTR tag and the predicted sample promotional information CTR, the model loss value is calculated, and the model parameters of the initial CTR prediction model are adjusted using the model loss value to obtain the CTR prediction model after training.

[0081] Optionally, the third training module is specifically used for:

[0082] The model loss value is obtained by calculating the click-through rate tag of the sample promotion information and the click-through rate of the predicted sample promotion information using the loss calculation formula.

[0083] The loss calculation formula satisfies the following formula:

[0084]

[0085] This represents the model loss value. Represents the crossover loss function. This indicates the click-through rate (CTR) tag for the sample promotional information. This indicates the click-through rate of the promotional information in the reference sample. This represents the click-through rate of the predicted sample promotional information. This indicates the predicted conversion rate of the sample promotional information. This represents the result of the logical AND operation between the click-through rate tag and the conversion rate tag of the sample promotional information. and This represents the network parameters of the initial click-through rate prediction model.

[0086] Thirdly, embodiments of this application provide a computer storage medium having multiple instructions adapted for loading and executing the methods described above by a processor.

[0087] Fourthly, embodiments of this application provide an electronic device, which may include: a memory and a processor; wherein the memory stores a computer program adapted to be loaded by the memory and to execute the above-described method.

[0088] The beneficial effects of the technical solutions provided in this application include at least the following:

[0089] The click-through rate (CTR) prediction method provided in this application extracts features from the user's search text to obtain semantic feature data, discrete feature data, and statistical feature data. Then, a CTR prediction model is used to predict the CTR of promotional information for candidate recommended terms, thereby obtaining the CTR of the promotional information corresponding to the candidate recommended terms. In this way, using the trained CTR prediction model can improve the prediction accuracy of the CTR of promotional information (such as search ad CTR) corresponding to candidate recommended terms. Since candidate recommended terms are obtained based on matching the user's search text, the CTR of promotional information can determine the recommendation ranking order of its corresponding candidate recommended terms among all candidate recommended terms. Therefore, this application embodiment can improve the accuracy of the recommendation ranking order of candidate recommended terms under the semantics of the current user's search text. Attached Figure Description

[0090] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0091] Figure 1 This is a schematic diagram illustrating a click-through rate prediction method provided in an embodiment of this application.

[0092] Figure 2 This is a schematic diagram of the model structure of a click-through rate prediction model provided in an embodiment of this application;

[0093] Figure 3 This is a flowchart illustrating another click-through rate prediction method provided in an embodiment of this application;

[0094] Figure 4 This is a schematic diagram of the structure of a click-through rate prediction device provided in an embodiment of this application;

[0095] Figure 5 This is a schematic diagram of the structure of a model prediction module provided in an embodiment of this application;

[0096] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0097] To make the inventive objectives, features, and advantages of the embodiments of this application more apparent and understandable, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0098] In the description of this application, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. In the description of this application, it should be noted that, unless otherwise expressly specified and limited, "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances. Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist; for example, A and / or B can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0099] In related technologies, CTR is a crucial metric in online advertising systems. Search engines can use estimated CTR to adjust other advertising parameters such as recommended keywords, ad display methods, ad placement, ad frequency, and recommended keywords. This adjustment not only improves the user's search experience but also ensures advertisers have ample ad exposure opportunities. Therefore, improving the accuracy of estimated CTR is a pressing technical problem that needs to be solved.

[0100] To address the aforementioned technical problems, this application will be described in detail below with reference to specific embodiments.

[0101] In one embodiment, such as Figure 1 As shown, a click-through rate (CTR) prediction method is proposed. This method can be implemented using a computer program and can run on a CTR prediction device based on the von Neumann architecture. This computer program can be integrated into applications or run as a standalone utility application.

[0102] Specifically, the subject of this click-through rate (CTR) prediction method is an electronic device, and the CTR prediction method includes:

[0103] S101: Obtain the user search text entered by the user, and perform feature extraction processing based on the user search text to obtain semantic feature data, discrete feature data, and statistical feature data.

[0104] User search text refers to the text entered by a user in a search context. Taking a search scenario within a search engine as an example, the text entered by the user in the search input box is the user search text.

[0105] Semantic feature data can include search terms, candidate recommended terms, historical click search term sets, and target promotional information titles. Semantic feature data can be simply understood as feature data containing semantic information.

[0106] Discrete feature data can include recommendation source, recommendation type, and search application identifier. Discrete feature data can be understood as feature data with a finite number of values; this type of feature data can be used to represent categories or classifications. Specifically, the recommendation source can be a finite number of recommendation sources, the recommendation type can be a finite number of recommendation types, and the search application identifier can be a finite number of application identifiers.

[0107] Statistical feature data can include similarity data and normalized click-through rate (CTR) data. Statistical feature data can be understood as features that can be represented numerically. Specifically, similarity data can include similarity values, and normalized CTR data can include click-through rate values.

[0108] In some embodiments, a search request sent by a user terminal can be received, parsed, and the user's search text can be obtained. The search request may carry the user's input text. Further, semantic feature processing can be performed on the user's search text to obtain semantic feature data. Then, discrete feature query processing can be performed based on the semantic feature data to obtain discrete feature data. Finally, feature calculation processing can be performed based on the semantic feature data to obtain statistical feature data.

[0109] S102, based on semantic feature data, discrete feature data, and statistical feature data, a click-through rate prediction model is used to predict the click-through rate of promotional information for candidate recommended words, and the click-through rate of promotional information corresponding to candidate recommended words is obtained.

[0110] Click-through rate (CTR) prediction models can be a type of model that uses a multi-objective joint modeling approach to predict the CTR of promotional information corresponding to candidate keywords. Taking promotional information as search ads in a search engine as an example, in a search engine, candidate keywords are the recommended terms used to trigger the display of search ads. Candidate keywords can be keywords obtained by processing search terms extracted from user search text. Therefore, candidate keywords are associated with the search ads they trigger, and each candidate keyword can correspond to a search ad. The CTR of promotional information is the CTR of search ads.

[0111] For details, please see Figure 2 , Figure 2 This is a schematic diagram of the model structure of a click-through rate prediction model provided in an embodiment of this application. Figure 2As shown, the click-through rate (CTR) prediction model is a two-tower model, consisting of a CTR prediction tower and a conversion rate prediction tower. By inputting semantic feature data, discrete feature data, and statistical feature data into the CTR prediction model, the CTR prediction tower predicts the CTR of the reference promotional information corresponding to the candidate recommended keywords, while the conversion rate tower predicts the conversion rate of the promotional information corresponding to the candidate recommended keywords. Finally, multiplying the prediction results of the two towers yields the output of the CTR prediction model, which is the predicted CTR of the promotional information corresponding to the candidate recommended keywords. The promotional information conversion rate refers to the proportion of users who click on the promotional information and complete a conversion action, where a conversion action refers to a user completing actions such as purchasing, registering, or downloading based on the promotional information.

[0112] The click-through rate (CTR) prediction method provided in this application extracts features from the user's search text to obtain semantic feature data, discrete feature data, and statistical feature data. Then, a CTR prediction model is used to predict the CTR of promotional information for candidate recommended terms, thereby obtaining the CTR of the promotional information corresponding to the candidate recommended terms. In this way, using the trained CTR prediction model can improve the prediction accuracy of the CTR of promotional information corresponding to candidate recommended terms (such as the predicted search ad CTR). Since candidate recommended terms are obtained based on matching the user's search text, the CTR of the promotional information can determine the recommendation ranking order of its corresponding candidate recommended terms among all candidate recommended terms. Therefore, this application embodiment can improve the accuracy of the recommendation ranking order of candidate recommended terms under the current semantic meaning of the user's search text.

[0113] Please see below. Figure 3 , Figure 3 This is a flowchart illustrating another embodiment of the click-through rate prediction method proposed in this application.

[0114] Specifically, the subject of this click-through rate (CTR) prediction method is an electronic device, and the CTR prediction method includes:

[0115] S301, Obtain the user search text entered by the user, and perform semantic feature extraction processing based on the user search text to obtain semantic feature data.

[0116] Semantic feature data can include search terms, candidate recommended terms, historical click search term sets, and target promotional message titles. Search terms refer to natural search terms extracted from user search text. Candidate recommended terms refer to keywords expected to be recommended to users based on search terms, used to trigger the display of search ads. Candidate recommended terms can include two types: those ultimately recommended to users and those not recommended. The historical click search term set refers to the set of specific search terms used by users when clicking on search ads corresponding to candidate recommended terms. The target promotional message title refers to the latest promotional message title corresponding to a candidate recommended term. Taking search ads in a search engine as an example, the target promotional message title can refer to the latest ad title corresponding to a candidate recommended term. In a search engine, candidate recommended terms can be associated with corresponding search ads, which can be displayed in ad slots on the search page. The search ads displayed in these slots can change, and the ad titles can also change. The target promotional message title is the latest ad title for the search ads.

[0117] In some embodiments, the user's search text can be obtained by receiving a search request sent by a user terminal and parsing the search request.

[0118] The steps of performing semantic feature extraction processing based on user search text to obtain semantic feature data may specifically include: performing search term extraction processing based on user search text to obtain search terms; performing recommendation term matching processing based on search terms to obtain candidate recommendation terms; obtaining the historical click search term set corresponding to the candidate recommendation terms; determining the target promotion information title corresponding to the candidate recommendation terms; and generating semantic feature data including search terms, candidate recommendation terms, historical click search term set, and target promotion information title.

[0119] Specifically, search term extraction processing can include text preprocessing, word segmentation, keyword extraction, semantic analysis, keyword filtering, and ranking. These processes can be applied to the user's search text to obtain search terms. Text preprocessing can include text cleaning, case unification, and stop word filtering. Word segmentation refers to dividing the text into words or phrases. Keyword extraction refers to extracting keywords from the segmented text based on algorithms and rules. Semantic analysis refers to further verifying and refining the extracted keywords based on context and semantic understanding. Keyword filtering and ranking refers to further filtering and ranking the extracted keywords according to the needs of a specific search engine or search application.

[0120] Candidate recommended words are obtained by performing recommendation word matching based on search terms. Specifically, the search terms can be input into the recommendation word matching model to perform recommendation word matching and obtain candidate recommended words.

[0121] Obtaining the set of historical click search terms corresponding to candidate recommended terms can be achieved by querying the specific search terms used by users when clicking on search ads corresponding to candidate recommended terms from the historical interaction data between users and search engines. The set of these search terms is the set of historical click search terms.

[0122] Determine the target promotion title corresponding to the candidate recommendation term. Specifically, in the search advertising database, query the target search advertisement corresponding to the candidate recommendation term. The target search advertisement is set to be displayed when the user clicks on the candidate recommendation term. Query the latest advertisement title of the target search advertisement and use the latest advertisement title as the target promotion title.

[0123] Finally, semantic feature data is generated, including search terms, candidate recommended terms, historical click search term set, and target promotion information title.

[0124] S302, discrete feature data is obtained by performing discrete feature query processing based on semantic feature data.

[0125] Discrete feature data can include recommendation source, recommendation type, and search application identifier. Recommendation source refers to the recommended position of the promotional information corresponding to the candidate recommendation term. Recommendation type refers to the recommendation type of the promotional information corresponding to the candidate recommendation term. Taking search advertisements in a search engine as an example, multiple ad display positions can be preset on the search page. Each ad display position can be used to display search advertisements for ad recommendation. Each ad display position can be represented by a different text identifier. The recommendation source is the ad position where the search advertisement corresponding to the candidate recommendation term is located. Search advertisements can include two types of advertisements: one is search advertisements determined by an ad recommendation system, and the other is search advertisements that can be determined directly by the search engine without going through an ad recommendation system. The recommendation type is any one of these two recommendation types. Search application identifier refers to the identifier of the search application used by the user to access the search engine website. For example, for search engine website b of search engine A, users can access search engine website b through various search applications.

[0126] In some embodiments, performing S302 may specifically include: obtaining an information recommendation source mapping relationship; querying the recommendation source and recommendation type corresponding to the candidate recommendation word based on the information recommendation source mapping relationship; obtaining the search application identifier corresponding to the user's search text; and generating discrete feature data including the recommendation source, recommendation type, and search application identifier.

[0127] The information recommendation source mapping relationship can include the correspondence between at least one reference candidate recommendation word, the reference recommendation source corresponding to the reference candidate recommendation word, and the reference recommendation type corresponding to the reference candidate recommendation word. The reference recommendation source refers to the recommendation position of the promotional information corresponding to the reference candidate recommendation word, and the reference recommendation type refers to the recommendation type of the promotional information corresponding to the reference candidate recommendation word.

[0128] Information recommendation source mapping relationships can be pre-configured and stored in a specified file, making it easy to retrieve the information recommendation source mapping relationships directly from the specified file later.

[0129] Specifically, in the information recommendation source mapping relationship, you can query the corresponding reference recommendation source and the corresponding reference recommendation type when the reference candidate recommendation word is a candidate recommendation word. The reference recommendation source is the recommendation source corresponding to the candidate recommendation word, and the reference recommendation type is the recommendation type corresponding to the candidate recommendation word.

[0130] Obtaining the search application identifier corresponding to the user's search text can specifically involve parsing the search request sent by the user's terminal to obtain the search application identifier. The search request can carry the search application identifier of the search engine website visited by the user.

[0131] Furthermore, discrete feature data including recommendation source, recommendation type, and search application identifier are generated.

[0132] S303, statistical feature data is obtained by performing feature calculation processing based on semantic feature data.

[0133] Statistical feature data can include similarity data and normalized click-through rate (CTR) data. Similarity data includes the similarity between the search term and the candidate recommended terms. Normalized CTR data refers to the normalized CTR data obtained after normalizing the historical CTRs corresponding to the candidate recommended terms.

[0134] In some embodiments, S303 may specifically include: performing similarity calculation processing based on search terms and candidate recommended terms to obtain similarity data; determining the historical click-through rate and normalized click count corresponding to the candidate recommended terms within a preset historical period, calculating the product of the normalized click count and the historical click-through rate to obtain the normalized click-through rate data corresponding to the candidate recommended terms; and generating statistical feature data including similarity data and normalized click data.

[0135] Specifically, the cosine similarity between the search term and each candidate recommended term can be calculated to obtain similarity data.

[0136] The preset historical period can be the most recent preset number of days. Historical click-through rate (CTR) refers to the CTR of candidate keywords within the preset historical period. Normalized click-through rate (NCTR) is the ratio of the number of clicks of a candidate keyword within the preset historical period to the reference number of clicks. The reference number of clicks is the difference between the maximum and minimum number of clicks among all candidate keywords within the preset historical period. Calculating the normalized CTR is necessary because keywords with a large historical CTR will have a normalized CTR that approximates the true posterior CTR after normalization, while keywords with a small historical CTR will have a normalized CTR that approximates 0. Only candidate keywords with high exposure and high click-through rates are statistically significant. Therefore, calculating the normalized CTR first, and then using the normalized CTR to calculate the normalized CTR, allows us to filter out candidate keywords with high exposure and high click-through rates.

[0137] After obtaining the similarity data and normalized click-through rate data mentioned above, statistical feature data including the similarity data and normalized click-through rate data can be generated.

[0138] S304 uses a click-through rate (CTR) prediction model to perform CTR prediction processing on semantic feature data, discrete feature data, and statistical feature data for candidate recommended words, and obtains the reference CTR of the promotional information corresponding to the candidate recommended words.

[0139] In some embodiments, performing S304 may specifically include: performing semantic cross processing on semantic feature data to obtain a semantic cross feature vector; performing embedding processing on discrete feature data to obtain a discrete feature vector; performing vector concatenation processing based on the semantic cross feature vector and the discrete feature vector to obtain a comprehensive feature vector; and performing click-through rate prediction processing on the comprehensive feature vector and statistical feature data for the candidate recommended words using a click-through rate prediction model to obtain the click-through rate of the reference promotion information corresponding to the candidate recommended words.

[0140] Semantic cross-processing is performed on semantic feature data to obtain semantic cross-feature vectors. Specifically, this can be done by mapping semantic feature data into target feature vectors through a pre-trained language model. The target feature vectors are continuous high-dimensional feature vectors. Fully connected layers are used to perform feature compression and feature cross-combination processing on these feature vectors to obtain semantic cross-feature vectors. Semantic feature data includes multiple features such as search terms, candidate recommended terms, historical click search term sets, and target promotion information titles. These can be mapped to a first feature vector using pre-trained language models, such as Bidirectional Encoder Representations from Transformers (BERT) and Generative Pre-trained Transformer (GPT) models. The first feature vector is mapped to a first feature vector, the second to a candidate recommended term, the third to a historical click search term set, and the fourth to a target promotion information title. This results in a target feature vector comprising the first, second, third, and fourth feature vectors. Further, a fully connected layer is used to perform feature compression and feature cross-combination processing on the first, second, third, and fourth feature vectors to obtain a semantic cross-feature vector.

[0141] Discrete feature vectors are obtained by embedding discrete feature data. Specifically, the word embedding of the recommendation source can be used to obtain the fifth feature vector, the word embedding of the recommendation type can be used to obtain the sixth feature vector, and the word embedding of the search application identifier can be used to obtain the seventh feature vector, resulting in a discrete feature vector including the fifth, sixth, and seventh feature vectors.

[0142] The comprehensive feature vector can include semantic cross feature vectors and discrete feature vectors. For example, the comprehensive feature vector is an m-dimensional vector. The values ​​of the first to n dimensions of the comprehensive feature vector are the values ​​of the semantic cross feature vectors, and the values ​​of the (n+1)-m dimensions of the comprehensive feature vector are the values ​​of the discrete feature vectors. m is greater than n, and both m and n are positive integers.

[0143] By using a click-through rate (CTR) prediction model to process the comprehensive feature vector and statistical feature data, the CTR of candidate recommended keywords is estimated, resulting in the reference CTR of the promotional information corresponding to the candidate recommended keywords. Specifically, this can be achieved by:

[0144] The comprehensive feature vector is compressed to obtain a compressed feature vector. Then, the click-through rate (CTR) prediction network in the CTR prediction model is used to predict the CTR of candidate recommended terms based on the compressed feature vector and statistical feature data, yielding the reference CTR for each candidate recommended term. The CTR prediction model can include a CTR prediction network and a conversion rate prediction network; the CTR prediction network predicts the CTR, and the conversion rate prediction network predicts the conversion rate.

[0145] S305 uses a click-through rate prediction model to perform conversion rate prediction processing on semantic feature data and discrete feature data for candidate recommended words, thereby obtaining the conversion rate of the promotional information corresponding to the candidate recommended words.

[0146] In some embodiments, a click-through rate (CTR) prediction model is used to perform conversion rate prediction processing on the comprehensive feature vector for candidate recommended words, thereby obtaining the conversion rate of the promotional information corresponding to the candidate recommended words. Specifically, the comprehensive feature vector is compressed to obtain a compressed feature vector, and the compressed feature vector is then used by the conversion rate prediction network in the CTR prediction model to perform conversion rate prediction processing on the candidate recommended words, thereby obtaining the conversion rate of the promotional information corresponding to the candidate recommended words.

[0147] S306 uses a click-through rate (CTR) prediction model to perform CTR calibration on the reference promotional information's CTR and conversion rate, thereby obtaining the CTR of the promotional information corresponding to the candidate recommended keywords.

[0148] In some embodiments, the click-through rate (CTR) of a reference promotional message is calculated by multiplying its CTR and conversion rate using a click-through rate (CTR) prediction model to obtain the CTR of the candidate recommendation term. Specifically, the CTR prediction model can perform a multiplication operation on the CTR of the reference promotional message and its conversion rate to obtain the CTR of the candidate recommendation term.

[0149] In some embodiments, before performing step S301, the following steps may also be performed: creating an initial click-through rate (CTR) prediction model for the scenario of predicting the CTR of promotional information; performing feature extraction processing based on sample user search text to obtain sample semantic feature data, sample discrete feature data, and sample statistical feature data; labeling the sample semantic feature data, sample discrete feature data, and sample statistical feature data with sample promotional information CTR tags corresponding to sample candidate recommendation words; using the sample semantic feature data, sample discrete feature data, and sample statistical feature data to perform at least one round of model training on the initial CTR prediction model; during the model training process, using the initial CTR prediction model to perform CTR prediction processing on the sample semantic feature data, sample discrete feature data, and sample statistical feature data for sample candidate recommendation words to obtain the predicted sample promotional information CTR corresponding to the sample candidate recommendation words; calculating the model loss value based on the sample promotional information CTR tags and the predicted sample promotional information CTR; using the model loss value to adjust the model parameters of the initial CTR prediction model to obtain the CTR prediction model after training.

[0150] The initial click-through rate (CTR) prediction model is a dual-tower model, which includes a CTR prediction tower and a conversion rate prediction tower.

[0151] Based on the sample user search text, feature extraction processing is performed to obtain sample semantic feature data, sample discrete feature data, and sample statistical feature data. Specifically, this can be done as follows: Search term extraction is performed on the sample user search text to obtain sample search terms; recommendation term matching is performed on the sample search terms to obtain sample candidate recommendation terms; the historical click search term set corresponding to the sample candidate recommendation terms is obtained; the sample promotion information title corresponding to the sample candidate recommendation terms is determined, and sample semantic feature data including sample search terms, sample candidate recommendation terms, the historical click search term set, and the sample promotion information title is generated; the sample recommendation source and sample recommendation type corresponding to the sample candidate recommendation terms are obtained, and the sample search application identifier corresponding to the sample user search text is obtained, generating sample discrete feature data including sample recommendation source, sample recommendation type, and sample search application identifier; similarity calculation is performed on the sample search terms and sample candidate recommendation terms to obtain sample similarity data; the historical click-through rate and sample normalized click count corresponding to the sample candidate recommendation terms within a preset historical period are determined; the product of the sample normalized click count and the historical click-through rate is calculated to obtain the sample normalized click-through rate data corresponding to the sample candidate recommendation terms; and sample statistical feature data including sample similarity data and sample normalized click data is generated.

[0152] An initial click-through rate (CTR) prediction model is used to predict the CTR of sample candidate recommendation words based on sample semantic feature data, sample discrete feature data, and sample statistical feature data. Specifically, this involves: using the CTR prediction tower in the initial CTR prediction model to predict the CTR of sample candidate recommendation words based on sample semantic feature data, sample discrete feature data, and sample statistical feature data, obtaining the reference sample CTR of sample candidate recommendation words; using the conversion rate prediction tower in the initial CTR prediction model to predict the conversion rate of sample candidate recommendation words based on sample semantic feature data and sample discrete feature data, obtaining the predicted sample conversion rate of sample candidate recommendation words; and finally, using the initial CTR prediction model to perform CTR calibration on the reference sample CTR and the predicted sample conversion rate of sample candidate recommendation words, obtaining the predicted CTR of sample candidate recommendation words.

[0153] The model loss value is calculated based on the click-through rate (CTR) tags of the sample promotional information and the predicted CTR of the sample promotional information. Specifically, the model loss value is obtained by calculating the CTR tags of the sample promotional information and the predicted CTR of the sample promotional information using a loss calculation formula. The loss calculation formula satisfies the following formula:

[0154]

[0155] This represents the model loss value. Represents the crossover loss function. This label represents the click-through rate of the sample promotional information. This indicates the click-through rate of the promotional information in the reference sample. This indicates the predicted click-through rate of the sample promotional information. This indicates the predicted conversion rate of the sample promotional information. This represents the logical AND operation result between the click-through rate tag and the conversion rate tag of the sample promotional information. and This represents the network parameters of the initial click-through rate prediction model.

[0156] It is understandable that the sample promotional information click-through rate tag includes the actual click-through rate of the promotional information corresponding to the sample candidate recommended keywords; the sample promotional information conversion rate tag includes the actual conversion rate of the promotional information corresponding to the same sample candidate recommended keywords.

[0157] The initial click-through rate (CTR) prediction model is adjusted using the model loss value to obtain a trained CTR prediction model. The trained CTR prediction model must meet certain training termination conditions, which may include factors such as the loss function value being less than or equal to a preset loss function threshold, or the number of iterations reaching a preset threshold. Specific training termination conditions can be determined based on actual circumstances and are not specifically limited here.

[0158] The click-through rate (CTR) prediction method provided in this application involves: extracting semantic features from user search text to obtain semantic feature data; performing discrete feature query processing on the semantic feature data to obtain discrete feature data; performing feature calculation processing on the semantic feature data to obtain statistical feature data; then using a CTR prediction model to perform CTR prediction processing on the semantic feature data, discrete feature data, and statistical feature data for candidate recommended words, thereby obtaining the CTR of the reference promotional information corresponding to the candidate recommended words; then using the CTR prediction model to perform conversion rate prediction processing on the semantic feature data and discrete feature data for candidate recommended words, thereby obtaining the conversion rate of the promotional information corresponding to the candidate recommended words; and finally using the CTR prediction model to perform CTR calibration processing on the reference promotional information CTR and promotional information conversion rate, thereby obtaining the CTR of the promotional information corresponding to the candidate recommended words. Thus, this embodiment of the application utilizes the natural law of promotional information first being exposed, then clicked, and finally converted. It combines the conversion rate of promotional information (such as search ad conversion rate) to estimate and calibrate the click-through rate (CTR) of promotional information (such as search ad CTR). Compared to traditional CTR estimation tasks that overestimate CTR due to noise data (such as accidental clicks), this embodiment uses the conversion rate of promotional information to calibrate the estimated CTR, solving the problem of overestimating CTR in related technologies, thereby obtaining a more accurate CTR. Furthermore, the CTR of promotional information can determine the recommendation ranking of its corresponding candidate recommendation terms among all candidate recommendation terms. Therefore, this embodiment of the application can improve the accuracy of the recommendation ranking of candidate recommendation terms under the current user's search text semantics.

[0159] The following will combine Figure 4 This application provides a detailed description of the click-through rate prediction device provided in its embodiments. It should be noted that... Figure 4 The click-through rate prediction device shown is used to perform the functions described in this application. Figures 1-3 The methods shown in the embodiments are for illustrative purposes only, illustrating the parts relevant to the embodiments of this application. For specific technical details not disclosed, please refer to this application. Figures 1-3 The example shown.

[0160] Please see Figure 4This diagram illustrates the structure of a click-through rate (CTR) prediction device according to an embodiment of this application. The CTR prediction device 1 can be implemented as all or part of a device through software, hardware, or a combination of both. According to some embodiments, the CTR prediction device 1 includes a feature extraction module 11 and a model prediction module 12, specifically used for:

[0161] Feature extraction module 11 is used to obtain the user search text input by the user, and perform feature extraction processing based on the user search text to obtain semantic feature data, discrete feature data, and statistical feature data;

[0162] The model prediction module 12 is used to perform click-through rate prediction processing on the promotional information of the candidate recommended words based on the semantic feature data, the discrete feature data, and the statistical feature data using a click-through rate prediction model, so as to obtain the click-through rate of the promotional information corresponding to the candidate recommended words.

[0163] Optionally, the feature extraction module 11 includes:

[0164] The first extraction unit is used to perform semantic feature extraction processing based on the user's search text to obtain semantic feature data;

[0165] The second extraction unit is used to perform discrete feature query processing based on the semantic feature data to obtain discrete feature data;

[0166] The third extraction unit is used to perform feature calculation processing based on the semantic feature data to obtain statistical feature data.

[0167] Optionally, the first extraction unit is specifically used for:

[0168] Search terms are obtained by extracting search terms from the user's search text, and candidate recommended terms are obtained by matching recommended terms with the search terms.

[0169] Obtain the set of historical click search terms corresponding to the candidate recommended terms;

[0170] Determine the target promotion information title corresponding to the candidate recommendation keywords;

[0171] Generate semantic feature data including the search term, the candidate recommended term, the historical click search term set, and the target promotion information title.

[0172] Optionally, the second extraction unit is specifically used for:

[0173] Obtain the information recommendation source mapping relationship, and query the recommendation source corresponding to the candidate recommendation word and the recommendation type corresponding to the candidate recommendation word based on the information recommendation source mapping relationship;

[0174] Obtain the search application identifier corresponding to the user's search text;

[0175] Generate discrete feature data including the recommendation source, the recommendation type, and the search application identifier.

[0176] Optionally, the third extraction unit is specifically used for:

[0177] Similarity data is obtained by performing similarity calculations based on the search terms and the candidate recommended terms;

[0178] Determine the historical click-through rate and normalized click count corresponding to the candidate recommended word within a preset historical period, calculate the product of the normalized click count and the historical click-through rate, and obtain the normalized click-through rate data corresponding to the candidate recommended word;

[0179] Generate statistical feature data including the similarity data and the normalized click data.

[0180] Optionally, please see Figure 5 This is a schematic diagram of the structure of a model prediction module provided in an embodiment of this application. Figure 5 As shown, the model prediction module 12 includes a first prediction unit 121, a second prediction unit 122, and a third prediction unit 123, specifically used for:

[0181] The first prediction unit 121 is used to perform click-through rate prediction processing on the semantic feature data, the discrete feature data, and the statistical feature data for the promotional information of the candidate recommendation words through the click-through rate prediction model, so as to obtain the reference click-through rate of the promotional information corresponding to the candidate recommendation words.

[0182] The second prediction unit 122 is used to perform conversion rate prediction processing on the semantic feature data and the discrete feature data for the promotion information of the candidate recommendation words through the click-through rate prediction model, so as to obtain the conversion rate of the promotion information corresponding to the candidate recommendation words.

[0183] The third prediction unit 123 is used to perform click-through rate calibration processing on the click-through rate of the reference promotional information and the predicted conversion rate of the promotional information through the click-through rate prediction model, so as to obtain the click-through rate of the promotional information corresponding to the candidate recommendation words.

[0184] Optionally, the first prediction unit 121 is specifically used for:

[0185] The semantic feature data is subjected to semantic cross-processing to obtain a semantic cross-feature vector, and the discrete feature data is subjected to embedding processing to obtain a discrete feature vector;

[0186] Based on the semantic cross feature vector and the discrete feature vector, a vector concatenation process is performed to obtain a comprehensive feature vector;

[0187] The click-through rate (CTR) prediction model is used to predict the CTR of promotional information for candidate recommended terms by processing the comprehensive feature vector and the statistical feature data. The reference CTR of promotional information corresponding to the candidate recommended terms is then obtained.

[0188] Optionally, the second prediction unit 122 is specifically used for:

[0189] The comprehensive feature vector is processed by a click-through rate prediction model to predict the conversion rate of promotional information for candidate recommended words, thereby obtaining the conversion rate of promotional information corresponding to the candidate recommended words.

[0190] Optionally, the third prediction unit 123 is specifically used for:

[0191] The click-through rate (CTR) of the candidate recommended keyword is obtained by multiplying the CTR of the reference promotional information and the conversion rate of the predicted promotional information using the CTR prediction model.

[0192] Optionally, the click-through rate prediction device 1 also includes:

[0193] The first training module is used to create an initial click-through rate (CTR) prediction model for the scenario of predicting CTR for promotional information.

[0194] The second training module is used to perform feature extraction processing based on sample user search text to obtain sample semantic feature data, sample discrete feature data, and sample statistical feature data, and to label the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data with the click-through rate tags of the sample promotion information corresponding to the sample candidate recommendation words.

[0195] The third training module is used to train the initial click-through rate (CTR) prediction model at least once using the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data. During the model training process, the initial CTR prediction model is used to perform CTR prediction processing on the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data for the promotional information of the sample candidate recommendation words, so as to obtain the predicted sample CTR of the promotional information corresponding to the sample candidate recommendation words. Based on the sample promotional information CTR tag and the predicted sample promotional information CTR, the model loss value is calculated, and the model parameters of the initial CTR prediction model are adjusted using the model loss value to obtain the CTR prediction model after training.

[0196] Optionally, the third training module is specifically used for:

[0197] The model loss value is obtained by calculating the click-through rate tag of the sample promotion information and the click-through rate of the predicted sample promotion information using the loss calculation formula.

[0198] The loss calculation formula satisfies the following formula:

[0199]

[0200] This represents the model loss value. Represents the crossover loss function. This indicates the click-through rate (CTR) tag for the sample promotional information. This indicates the click-through rate of the promotional information in the reference sample. This represents the click-through rate of the predicted sample promotional information. This indicates the predicted conversion rate of the sample promotional information. This represents the result of the logical AND operation between the click-through rate tag and the conversion rate tag of the sample promotional information. and This represents the network parameters of the initial click-through rate prediction model.

[0201] Please refer to Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. For example, the electronic device in this embodiment may specifically be a server, and the electronic device may include one or more of the following components: a processor 110, a memory 120, an input device 130, an output device 140, and a bus 150. The processor 110, memory 120, input device 130, and output device 140 can be connected via the bus 150.

[0202] Processor 110 may include one or more processing cores. Processor 110 connects to various parts of the electronic device using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 120, and by calling data stored in memory 120. Optionally, processor 110 may be implemented using at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). Processor 110 may integrate one or more of a central processing unit (CPU), graphics processing unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the displayed content; and the modem handles wireless communication. It is understood that the modem may also not be integrated into processor 110 and may be implemented separately using a communication chip.

[0203] The memory 120 may include random access memory (RAM) or read-only memory (ROM). Optionally, the memory 120 may include a non-transitory computer-readable storage medium. The memory 120 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (e.g., touch function, sound playback function, image playback function, etc.), instructions for implementing the various method embodiments described below, etc. The operating system may be the Android system, including systems deeply developed based on the Android system, the iOS system developed by Apple Inc., including systems deeply developed based on the iOS system, or other systems.

[0204] In order for the operating system to distinguish the specific application scenarios of third-party applications, it is necessary to establish data communication between the third-party applications and the operating system. This would allow the operating system to obtain the current scenario information of the third-party applications at any time, and then perform targeted system resource adaptation based on the current scenario.

[0205] The input device 130 is used to receive input instructions or data, and includes, but is not limited to, a keyboard, mouse, camera, microphone, or touch device. The output device 140 is used to output instructions or data, and includes, but is not limited to, a display device and a speaker. In one example, the input device 130 and the output device 140 can be combined, and the input device 130 and the output device 140 can be a touch display screen.

[0206] The touch display screen can be designed as a full-screen, curved screen, or irregularly shaped screen. It can also be designed as a combination of a full-screen and a curved screen, or a combination of an irregularly shaped screen and a curved screen; however, this application does not limit the specific design of the touch display screen.

[0207] In addition, those skilled in the art will understand that the structure of the electronic device shown in the above figures does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements. For example, the electronic device may also include radio frequency circuits, input units, sensors, audio circuits, Wireless Fidelity (WiFi) modules, power supplies, Bluetooth modules, etc., which will not be described in detail here.

[0208] exist Figure 6 In the illustrated electronic device, the processor 110 can be used to call the program for the click-through rate prediction method stored in the memory 120, and specifically perform the following operations:

[0209] Obtain the user's search text input by the user, and perform feature extraction processing based on the user's search text to obtain semantic feature data, discrete feature data, and statistical feature data;

[0210] Based on the semantic feature data, the discrete feature data, and the statistical feature data, a click-through rate (CTR) prediction model is used to predict the CTR of promotional information for candidate recommended words, thereby obtaining the CTR of promotional information corresponding to the candidate recommended words.

[0211] In one embodiment, when the processor 110 performs the step of extracting semantic feature data, discrete feature data, and statistical feature data based on the user search text, it specifically performs the following operations:

[0212] Semantic feature data is obtained by performing semantic feature extraction processing on the user's search text.

[0213] Discrete feature data is obtained by performing discrete feature query processing based on the semantic feature data;

[0214] Statistical feature data is obtained by performing feature calculations based on the semantic feature data.

[0215] In one embodiment, when the processor 110 performs the step of extracting semantic features based on the user's search text to obtain semantic feature data, it specifically performs the following operations:

[0216] Search terms are obtained by extracting search terms from the user's search text, and candidate recommended terms are obtained by matching recommended terms with the search terms.

[0217] Obtain the set of historical click search terms corresponding to the candidate recommended terms;

[0218] Determine the target promotion information title corresponding to the candidate recommendation keywords;

[0219] Generate semantic feature data including the search term, the candidate recommended term, the historical click search term set, and the target promotion information title.

[0220] In one embodiment, when the processor 110 performs the step of obtaining discrete feature data by performing discrete feature query processing based on the semantic feature data, it specifically performs the following operations:

[0221] Obtain the information recommendation source mapping relationship, and query the recommendation source corresponding to the candidate recommendation word and the recommendation type corresponding to the candidate recommendation word based on the information recommendation source mapping relationship;

[0222] Obtain the search application identifier corresponding to the user's search text;

[0223] Generate discrete feature data including the recommendation source, the recommendation type, and the search application identifier.

[0224] In one embodiment, when the processor 110 performs the step of performing feature calculation processing based on the semantic feature data to obtain statistical feature data, it specifically performs the following operations:

[0225] Similarity data is obtained by performing similarity calculations based on the search terms and the candidate recommended terms;

[0226] Determine the historical click-through rate and normalized click count corresponding to the candidate recommended word within a preset historical period, calculate the product of the normalized click count and the historical click-through rate, and obtain the normalized click-through rate data corresponding to the candidate recommended word;

[0227] Generate statistical feature data including the similarity data and the normalized click data.

[0228] In one embodiment, when the processor 110 performs the step of estimating the click-through rate (CTR) of promotional information for candidate recommended terms using a click-through rate prediction model based on the semantic feature data, the discrete feature data, and the statistical feature data, and obtains the CTR of the promotional information corresponding to the candidate recommended terms, the processor 110 specifically performs the following operations:

[0229] The click-through rate (CTR) prediction model is used to perform CTR prediction processing on the semantic feature data, the discrete feature data, and the statistical feature data for the promotional information of the candidate recommendation words, so as to obtain the reference CTR of the promotional information corresponding to the candidate recommendation words.

[0230] The click-through rate prediction model is used to perform conversion rate prediction processing on the semantic feature data and the discrete feature data for the promotion information of the candidate recommendation words, so as to obtain the conversion rate of the promotion information corresponding to the candidate recommendation words.

[0231] The click-through rate (CTR) of the reference promotional information and the predicted promotional information conversion rate are calibrated using the CTR prediction model to obtain the CTR of the promotional information corresponding to the candidate recommendation keywords.

[0232] In one embodiment, when the processor 110 performs the step of estimating the click-through rate (CTR) of promotional information for candidate recommended terms by using a click-through rate (CTR) prediction model on the semantic feature data, the discrete feature data, and the statistical feature data, and obtaining the reference CTR of promotional information corresponding to the candidate recommended terms, the processor 110 specifically performs the following operations:

[0233] The semantic feature data is subjected to semantic cross-processing to obtain a semantic cross-feature vector, and the discrete feature data is subjected to embedding processing to obtain a discrete feature vector;

[0234] Based on the semantic cross feature vector and the discrete feature vector, a vector concatenation process is performed to obtain a comprehensive feature vector;

[0235] The click-through rate (CTR) prediction model is used to predict the CTR of promotional information for candidate recommended terms by processing the comprehensive feature vector and the statistical feature data. The reference CTR of promotional information corresponding to the candidate recommended terms is then obtained.

[0236] In one embodiment, when the processor 110 performs the step of estimating the conversion rate of promotional information for candidate recommended words by using the click-through rate prediction model on the semantic feature data and the discrete feature data, and obtaining the conversion rate of promotional information corresponding to the candidate recommended words, the processor 110 specifically performs the following operations:

[0237] The comprehensive feature vector is processed by a click-through rate prediction model to predict the conversion rate of promotional information for candidate recommended words, thereby obtaining the conversion rate of promotional information corresponding to the candidate recommended words.

[0238] In one embodiment, when the processor 110 performs the step of calibrating the click-through rate (CTR) of the reference promotional information and the estimated conversion rate of the promotional information using the CTR prediction model to obtain the CTR of the promotional information corresponding to the candidate recommended keyword, it specifically performs the following operations:

[0239] The click-through rate (CTR) of the candidate recommended keyword is obtained by multiplying the CTR of the reference promotional information and the conversion rate of the predicted promotional information using the CTR prediction model.

[0240] In one embodiment, the processor 110 also performs the following operations:

[0241] Create an initial click-through rate (CTR) prediction model for scenarios involving promotional information click-through rate prediction;

[0242] Based on the sample user search text, feature extraction processing is performed to obtain sample semantic feature data, sample discrete feature data, and sample statistical feature data. The sample semantic feature data, the sample discrete feature data, and the sample statistical feature data are labeled with the click-through rate tags of the sample promotion information corresponding to the sample candidate recommendation words.

[0243] The initial click-through rate (CTR) prediction model is trained at least once using the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data. During the model training process, the initial CTR prediction model is used to perform CTR prediction processing on the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data for the promotional information of the sample candidate recommendation words, so as to obtain the predicted sample CTR of the promotional information corresponding to the sample candidate recommendation words. Based on the sample promotional information CTR tag and the predicted sample promotional information CTR, the model loss value is calculated. The model loss value is used to adjust the model parameters of the initial CTR prediction model to obtain the CTR prediction model after training.

[0244] In one embodiment, when the processor 110 performs the step of calculating the model loss value based on the sample promotion information click-through rate label and the predicted sample promotion information click-through rate, it specifically performs the following operations:

[0245] The model loss value is obtained by calculating the click-through rate tag of the sample promotion information and the click-through rate of the predicted sample promotion information using the loss calculation formula.

[0246] The loss calculation formula satisfies the following formula:

[0247]

[0248] This represents the model loss value. Represents the crossover loss function. This indicates the click-through rate (CTR) tag for the sample promotional information. This indicates the click-through rate of the promotional information in the reference sample. This represents the click-through rate of the predicted sample promotional information. This indicates the predicted conversion rate of the sample promotional information. This represents the result of the logical AND operation between the click-through rate tag and the conversion rate tag of the sample promotional information. and This represents the network parameters of the initial click-through rate prediction model.

[0249] This application also provides a computer-readable storage medium storing at least one instruction that is executed by a processor to implement the click-through rate prediction method as described in the above embodiments.

[0250] This application also provides a computer program product that stores at least one instruction, which is loaded and executed by the processor to implement the click-through rate prediction method as described in the above embodiments.

[0251] Those skilled in the art will recognize that the functions described in the embodiments of this application in one or more of the above examples can be implemented using hardware, software, firmware, or any combination thereof. When implemented using software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transfer of a computer program from one place to another. Storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0252] The above description is merely an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A click rate prediction method, characterized in that, The method includes: The system acquires the user's search text and performs feature extraction processing on the search text to obtain semantic feature data, discrete feature data, and statistical feature data. The semantic feature data includes search terms, candidate recommended terms, a set of historical clicked search terms, and the title of the target promotional information. The discrete feature data includes the recommendation source, recommendation type, and search application identifier. The statistical feature data includes similarity data and normalized click-through rate (CTR) data. The similarity data includes the similarity between the search terms and the candidate recommended terms, and the normalized CTR data is obtained by normalizing the historical CTRs corresponding to the candidate recommended terms. Based on the semantic feature data, the discrete feature data, and the statistical feature data, a click-through rate (CTR) prediction model is used to predict the CTR of promotional information for candidate recommended words, thereby obtaining the CTR of promotional information corresponding to the candidate recommended words. The step of using a click-through rate (CTR) prediction model based on the semantic feature data, the discrete feature data, and the statistical feature data to predict the CTR of promotional information for candidate recommended terms, and obtaining the CTR of promotional information corresponding to the candidate recommended terms, includes: using the CTR prediction model to predict the CTR of promotional information for candidate recommended terms based on the semantic feature data, the discrete feature data, and the statistical feature data, to obtain a reference CTR of promotional information corresponding to the candidate recommended terms; using the CTR prediction model to predict the conversion rate of promotional information for candidate recommended terms based on the semantic feature data and the discrete feature data, to obtain the conversion rate of promotional information corresponding to the candidate recommended terms; and using the CTR prediction model to perform CTR calibration processing on the reference CTR of promotional information and the predicted conversion rate of promotional information, to obtain the CTR of promotional information corresponding to the candidate recommended terms.

2. The method of claim 1, wherein, The semantic feature data, discrete feature data, and statistical feature data obtained by feature extraction processing based on the user's search text include: Semantic feature data is obtained by performing semantic feature extraction processing on the user's search text. Discrete feature data is obtained by performing discrete feature query processing based on the semantic feature data; Statistical feature data is obtained by performing feature calculations based on the semantic feature data.

3. The method of claim 2, wherein, The semantic feature data obtained by extracting semantic features based on the user's search text includes: Search terms are obtained by extracting search terms from the user's search text, and candidate recommended terms are obtained by matching recommended terms with the search terms. Obtain the set of historical click search terms corresponding to the candidate recommended terms; Determine the target promotion information title corresponding to the candidate recommendation keywords; Generate semantic feature data including the search term, the candidate recommended term, the historical click search term set, and the target promotion information title.

4. The method of claim 3, wherein, The discrete feature data obtained by performing discrete feature query processing based on the semantic feature data includes: Obtain the information recommendation source mapping relationship, and query the recommendation source corresponding to the candidate recommendation word and the recommendation type corresponding to the candidate recommendation word based on the information recommendation source mapping relationship; Obtain the search application identifier corresponding to the user's search text; Generate discrete feature data including the recommendation source, the recommendation type, and the search application identifier.

5. The method of claim 3, wherein, The statistical feature data obtained by performing feature calculation based on the semantic feature data includes: Similarity data is obtained by performing similarity calculations based on the search terms and the candidate recommended terms; Determine the historical click-through rate and normalized click count corresponding to the candidate recommended word within a preset historical period, calculate the product of the normalized click count and the historical click-through rate, and obtain the normalized click-through rate data corresponding to the candidate recommended word; Generate statistical feature data including the similarity data and the normalized click data.

6. The method of claim 1, wherein, The step of using a click-through rate (CTR) prediction model to predict the CTR of promotional information for candidate recommended terms, based on the semantic feature data, the discrete feature data, and the statistical feature data, to obtain the reference CTR of promotional information corresponding to the candidate recommended terms, includes: The semantic feature data is subjected to semantic cross-processing to obtain a semantic cross-feature vector, and the discrete feature data is subjected to embedding processing to obtain a discrete feature vector; Based on the semantic cross feature vector and the discrete feature vector, a vector concatenation process is performed to obtain a comprehensive feature vector; The click-through rate (CTR) prediction model is used to predict the CTR of promotional information for candidate recommended terms by processing the comprehensive feature vector and the statistical feature data. The reference CTR of promotional information corresponding to the candidate recommended terms is then obtained.

7. The method of claim 6, wherein, The step of using the click-through rate prediction model to predict the conversion rate of promotional information for candidate recommended terms by analyzing the semantic feature data and the discrete feature data, and obtaining the conversion rate of promotional information corresponding to the candidate recommended terms, includes: The comprehensive feature vector is processed by a click-through rate prediction model to predict the conversion rate of promotional information for candidate recommended words, thereby obtaining the conversion rate of promotional information corresponding to the candidate recommended words.

8. The method of claim 1, wherein, The step of performing click-through rate (CTR) calibration processing on the reference promotional information CTR and the estimated promotional information conversion rate using the CTR prediction model to obtain the CTR of the promotional information corresponding to the candidate recommended keywords includes: The click-through rate (CTR) of the candidate recommended keyword is obtained by multiplying the CTR of the reference promotional information and the conversion rate of the predicted promotional information using the CTR prediction model.

9. The method of claim 1, wherein, The method further includes: Create an initial click-through rate (CTR) prediction model for scenarios involving promotional information click-through rate prediction; Based on the sample user search text, feature extraction processing is performed to obtain sample semantic feature data, sample discrete feature data, and sample statistical feature data. The sample semantic feature data, the sample discrete feature data, and the sample statistical feature data are labeled with the click-through rate tags of the sample promotion information corresponding to the sample candidate recommendation words. The initial click-through rate (CTR) prediction model is trained at least once using the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data. During the model training process, the initial CTR prediction model is used to perform CTR prediction processing on the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data for the promotional information of the sample candidate recommendation words, so as to obtain the predicted sample CTR of the promotional information corresponding to the sample candidate recommendation words. Based on the sample promotional information CTR tag and the predicted sample promotional information CTR, the model loss value is calculated. The model loss value is used to adjust the model parameters of the initial CTR prediction model to obtain the CTR prediction model after training.

10. The method of claim 9, wherein, The calculation of the model loss value based on the click-through rate label of the sample promotional information and the click-through rate of the predicted sample promotional information includes: The model loss value is obtained by calculating the click-through rate tag of the sample promotion information and the click-through rate of the predicted sample promotion information using the loss calculation formula. The loss calculation formula satisfies the following formula: This represents the model loss value. Represents the crossover loss function. This indicates the click-through rate (CTR) tag for the sample promotional information. This indicates the click-through rate of the promotional information in the reference sample. This represents the click-through rate of the predicted sample promotional information. This indicates the predicted conversion rate of the sample promotional information. This represents the result of the logical AND operation between the click-through rate tag and the conversion rate tag of the sample promotional information. and This represents the network parameters of the initial click-through rate prediction model.

11. A click-through rate prediction device, characterized in that, The device includes: The feature extraction module is used to acquire the user's search text and perform feature extraction processing on the user's search text to obtain semantic feature data, discrete feature data, and statistical feature data. The semantic feature data includes search terms, candidate recommended terms, a set of historical clicked search terms, and the title of the target promotional information. The discrete feature data includes the recommendation source, recommendation type, and search application identifier. The statistical feature data includes similarity data and normalized click-through rate (CTR) data. The similarity data includes the similarity between the search terms and the candidate recommended terms, and the normalized CTR data is obtained by normalizing the historical CTR corresponding to the candidate recommended terms. The model prediction module is used to perform click-through rate prediction processing on the promotional information of the candidate recommended words based on the semantic feature data, the discrete feature data, and the statistical feature data using a click-through rate prediction model, so as to obtain the click-through rate of the promotional information corresponding to the candidate recommended words; The model prediction module includes a first prediction unit, a second prediction unit, and a third prediction unit, specifically used for: the first prediction unit performing click-through rate (CTR) prediction processing on the semantic feature data, the discrete feature data, and the statistical feature data for candidate recommended terms using a click-through rate (CTR) prediction model, to obtain a reference CTR for the candidate recommended terms; the second prediction unit performing CTR prediction processing on the semantic feature data and the discrete feature data for candidate recommended terms using the CTR prediction model, to obtain a conversion rate for the candidate recommended terms; and the third prediction unit performing CTR calibration processing on the reference CTR and the predicted conversion rate for the candidate recommended terms using the CTR prediction model, to obtain the CTR for the candidate recommended terms.

12. The apparatus of claim 11, wherein, The feature extraction module includes: The first extraction unit is used to perform semantic feature extraction processing based on the user's search text to obtain semantic feature data; The second extraction unit is used to perform discrete feature query processing based on the semantic feature data to obtain discrete feature data; The third extraction unit is used to perform feature calculation processing based on the semantic feature data to obtain statistical feature data.

13. The apparatus of claim 12, wherein, The first extraction unit is specifically used for: Search terms are obtained by extracting search terms from the user's search text, and candidate recommended terms are obtained by matching recommended terms with the search terms. Obtain the set of historical click search terms corresponding to the candidate recommended terms; Determine the target promotion information title corresponding to the candidate recommendation keywords; Generate semantic feature data including the search term, the candidate recommended term, the historical click search term set, and the target promotion information title.

14. The apparatus of claim 13, wherein, The second extraction unit is specifically used for: Obtain the information recommendation source mapping relationship, and query the recommendation source corresponding to the candidate recommendation word and the recommendation type corresponding to the candidate recommendation word based on the information recommendation source mapping relationship; Obtain the search application identifier corresponding to the user's search text; Generate discrete feature data including the recommendation source, the recommendation type, and the search application identifier.

15. The apparatus of claim 13, wherein, The third extraction unit is specifically used for: Similarity data is obtained by performing similarity calculations based on the search terms and the candidate recommended terms; Determine the historical click-through rate and normalized click count corresponding to the candidate recommended word within a preset historical period, calculate the product of the normalized click count and the historical click-through rate, and obtain the normalized click-through rate data corresponding to the candidate recommended word; Generate statistical feature data including the similarity data and the normalized click data.

16. The apparatus of claim 11, wherein, The first prediction unit is specifically used for: The semantic feature data is subjected to semantic cross-processing to obtain a semantic cross-feature vector, and the discrete feature data is subjected to embedding processing to obtain a discrete feature vector; Based on the semantic cross feature vector and the discrete feature vector, a vector concatenation process is performed to obtain a comprehensive feature vector; The click-through rate (CTR) prediction model is used to predict the CTR of promotional information for candidate recommended terms by processing the comprehensive feature vector and the statistical feature data. The reference CTR of promotional information corresponding to the candidate recommended terms is then obtained.

17. The apparatus of claim 16, wherein, The second prediction unit is specifically used for: The comprehensive feature vector is processed by a click-through rate prediction model to predict the conversion rate of promotional information for candidate recommended words, thereby obtaining the conversion rate of promotional information corresponding to the candidate recommended words.

18. The apparatus of claim 11, wherein, The third prediction unit is specifically used for: The click-through rate (CTR) of the candidate recommended keyword is obtained by multiplying the CTR of the reference promotional information and the conversion rate of the predicted promotional information using the CTR prediction model.

19. The apparatus of claim 11, wherein, The click-through rate prediction device further includes: The first training module is used to create an initial click-through rate (CTR) prediction model for the scenario of predicting CTR for promotional information. The second training module is used to perform feature extraction processing based on sample user search text to obtain sample semantic feature data, sample discrete feature data, and sample statistical feature data, and to label the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data with the click-through rate tags of the sample promotion information corresponding to the sample candidate recommendation words. The third training module is used to train the initial click-through rate (CTR) prediction model at least once using the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data. During the model training process, the initial CTR prediction model is used to perform CTR prediction processing on the sample semantic feature data, the sample discrete feature data, and the sample statistical feature data for the promotional information of the sample candidate recommendation words, so as to obtain the predicted sample CTR of the promotional information corresponding to the sample candidate recommendation words. Based on the sample promotional information CTR tag and the predicted sample promotional information CTR, the model loss value is calculated, and the model parameters of the initial CTR prediction model are adjusted using the model loss value to obtain the CTR prediction model after training.

20. The apparatus according to claim 19, characterized in that, The third training module is specifically used for: The model loss value is obtained by calculating the click-through rate tag of the sample promotion information and the click-through rate of the predicted sample promotion information using the loss calculation formula. The loss calculation formula satisfies the following formula: This represents the model loss value. Represents the crossover loss function. This indicates the click-through rate (CTR) tag for the sample promotional information. This indicates the click-through rate of the promotional information in the reference sample. This represents the click-through rate of the predicted sample promotional information. This indicates the predicted conversion rate of the sample promotional information. This represents the result of the logical AND operation between the click-through rate tag and the conversion rate tag of the sample promotional information. and This represents the network parameters of the initial click-through rate prediction model.

21. A computer storage medium, comprising, The computer storage medium stores a plurality of instructions adapted for loading by a processor and executing the method as described in any one of claims 1 to 10.

22. An electronic device, comprising: include: A processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and executed as described in any one of claims 1 to 10.

Images