CN116186411A - Method and device for constructing user behavior prediction model based on deep recommendation model - Google Patents

Abstract

The invention discloses the construction of a user behavior prediction model, user behavior prediction method, device, equipment and medium. The construction method of the user behavior prediction model includes: generating an original sample set according to the historical behavior data of multiple users; generating each matrix corresponding to each original sample according to the behavior set in each original sample; forming a training sample set; using the training sample Set to train the deep recommendation model to obtain the user behavior prediction model. The user behavior prediction method includes: obtaining the user feature information of the user to be predicted and the item feature information of the item to be predicted; obtaining the behavior prediction probability matrix of the user to be predicted to be predicted; according to the behavior prediction probability matrix, verifying whether the item to be predicted will be recommended to the user to be predicted. The technical solution of the present invention provides a new way of constructing a user behavior prediction model to improve the accuracy of user behavior prediction.

Description

Method and device for constructing user behavior prediction model based on deep recommendation model

technical field

The present invention relates to the field of model construction, in particular to a method for constructing a user behavior prediction model, a user behavior prediction method, a device, a device, and a medium.

Background technique

With the rapid development of the Internet, the information data in the network is increasing exponentially, which leads to the redundancy of information, and a problem arises, that is, how users can obtain the information they are interested in in an effective time. At this time, the recommendation system based on user behavior prediction came into being at this opportunity. The recommendation system based on user behavior prediction greatly improves the user experience and can effectively reduce the time consumption when users search for interesting information by themselves. .

There are two main implementation methods for the above recommendation system: method 1 is to convert multiple training targets into one target according to the weight to let the model fit; method 2 is to build multiple network structures, so that different networks are responsible for different training targets, and finally after a certain A measurement method that combines multiple trainings to maximize the benefits of multiple objectives.

In the process of realizing the present invention, the inventor found that the above-mentioned prior art has the following problems: method 1 will completely split each training target, destroy the integrity of information, and then affect the accuracy of recommendation; in method 2, the components of manual participation Too many, there will be too many human factors and affect the recommendation results.

Contents of the invention

The present invention provides a construction of a user behavior prediction model, a user behavior prediction method, a device, a device, and a medium, so as to provide a new way of constructing a user behavior prediction model and improve the accuracy of user behavior prediction.

In a first aspect, an embodiment of the present invention provides a method for constructing a user behavior prediction model, the method including;

According to the historical behavior data of multiple users, the original sample set is generated. The original sample includes the user feature information of the set user, the item feature information of the set item, and the set of behaviors performed by the set user on the set item. There is a progressive relationship between behaviors;

According to the behavior set in each original sample, generate the first behavior characteristic matrix, the second behavior characteristic matrix and the third behavior characteristic matrix respectively corresponding to each original sample;

According to the user characteristic information of the set user in the original sample, set the item characteristic information, the first behavior characteristic matrix, the second behavior characteristic matrix and the third behavior characteristic matrix of the item to form a training sample set, the second behavior characteristic matrix and The third row feature matrix is used as the labeled data in the training sample;

Use the training sample set to train the deep recommendation model to obtain the user behavior prediction model.

In a second aspect, an embodiment of the present invention provides a user behavior prediction method, the method comprising:

Obtain the user feature information of the user to be predicted and the item feature information of the item to be predicted;

Input the user feature information of the user to be predicted and the item feature information of the item to be predicted into the user behavior prediction model trained by the user behavior prediction model construction method described in any embodiment of the present invention, and obtain the treatment of the user to be predicted Predict the behavior prediction probability matrix of the item;

According to the behavior prediction probability matrix, verify whether the item to be predicted is recommended to the user to be predicted.

In a third aspect, an embodiment of the present invention provides a device for constructing a user behavior prediction model, the device comprising:

The original sample set generation module is used to generate an original sample set according to the historical behavior data of multiple users. The original sample includes the user feature information of the set user, the item feature information of the set item, and the set user's execution of the set item. Behavior collection, there is a progressive relationship between the behaviors in the behavior collection;

Feature matrix generating module, for generating the first behavior feature matrix, the second behavior feature matrix and the third behavior feature matrix respectively corresponding to each original sample according to the behavior set in each original sample;

The training sample set generation module is used to set the item feature information, the first behavior feature matrix, the second behavior feature matrix and the third behavior feature matrix of the item according to the user feature information of the set user in the original sample to form a training sample set, the second row feature matrix and the third row feature matrix are used as labeled data in the training samples;

The model training module is used to use the training sample set to train the deep recommendation model to obtain the user behavior prediction model.

In a fourth aspect, an embodiment of the present invention provides a user behavior prediction device, which includes:

A feature information acquisition module, configured to acquire user feature information of the user to be predicted and item feature information of the item to be predicted;

The probability matrix acquisition module is used to input the user characteristic information of the user to be predicted and the item characteristic information of the item to be predicted into the user behavior prediction model trained by the method for constructing the user behavior prediction model as described in any embodiment of the present invention In , the behavior prediction probability matrix of the user to be predicted for the predicted item is obtained;

A recommendation verification module, configured to verify whether the item to be predicted is recommended to the user to be predicted according to the behavior prediction probability matrix.

In a fifth aspect, an embodiment of the present invention provides an electronic device, the electronic device comprising:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein,

The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor, so that the at least one processor can execute the method described in any embodiment of the present invention. A method for constructing a user behavior prediction model, or executing the user behavior prediction method described in any embodiment of the present invention.

In a sixth aspect, an embodiment of the present invention provides a computer-readable storage medium, the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement any of the embodiments of the present invention. A method for constructing a user behavior prediction model, or implementing the user behavior prediction method described in any embodiment of the present invention.

According to the technical solution of the embodiment of the present invention, the original sample set is generated according to the historical behavior data of multiple users, and the first behavior feature matrix and the second behavior feature matrix respectively corresponding to each original sample are generated according to the behavior set in each original sample. matrix and the third behavior feature matrix, and then according to the user feature information of the set user in the original sample, set the item feature information, the first behavior feature matrix, the second behavior feature matrix and the third behavior feature matrix to form a training The sample set and labeled data, and finally use the training sample set to train the deep recommendation model to obtain a user behavior prediction model, which provides a new way to build a user behavior prediction model. Based on this new user behavior prediction model, it can effectively improve user behavior. The accuracy of behavior prediction can further improve the hit rate of recommended items to users' actual needs, and effectively reduce the time consumption when users search for interesting information by themselves.

It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will be easily understood from the following description.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained based on these drawings without creative effort.

FIG. 1 is a flowchart of a method for constructing a user behavior prediction model according to Embodiment 1 of the present invention;

FIG. 2a is a flow chart of a method for constructing a user behavior prediction model according to Embodiment 2 of the present invention;

Fig. 2b is a frame diagram of a user behavior prediction model obtained by providing a method according to Embodiment 2 of the present invention;

FIG. 3 is a flow chart of a user behavior prediction method provided according to Embodiment 3 of the present invention;

FIG. 4 is a schematic structural diagram of a device for constructing a user behavior prediction model according to Embodiment 4 of the present invention;

FIG. 5 is a schematic structural diagram of a user behavior prediction device according to Embodiment 5 of the present invention;

FIG. 6 is a schematic structural diagram of an electronic device implementing a method for constructing a user behavior prediction model and a method for user behavior prediction according to an embodiment of the present invention.

Detailed ways

In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only It is an embodiment of a part of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

It should be noted that the terms "first" and "second" in the description and claims of the present invention and the above drawings are used to distinguish similar objects, but not necessarily used to describe a specific sequence or sequence. It is to be understood that the data so used are interchangeable under appropriate circumstances such that the embodiments of the invention described herein can be practiced in sequences other than those illustrated or described herein. Furthermore, the terms "comprising" and "having", as well as any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product or device comprising a sequence of steps or elements is not necessarily limited to the expressly listed instead, may include other steps or elements not explicitly listed or inherent to the process, method, product or apparatus.

Embodiment one

Fig. 1 is a flow chart of a method for constructing a user behavior prediction model provided by Embodiment 1 of the present invention. This embodiment is applicable to the situation of constructing a user behavior prediction model, and the method can be constructed by a device for constructing a user behavior prediction model. The device for constructing the user behavior prediction model can be implemented in the form of hardware and/or software, and the device for constructing the user behavior prediction model can be configured in a terminal or server with data processing functions. As shown in Figure 1, the method includes:

S110. Generate an original sample set according to the historical behavior data of multiple users.

In this embodiment, the historical behavior data includes the behavior data of the corresponding operation behaviors of multiple users on the preset platform; for example, when a certain user purchases item a, the implementation process of the purchase operation (such as input keywords, browsing search results or clicking on item links, etc.), item feature information of item a, and user feature information of a certain user; further, the preset platform can be a shopping platform with a product recommendation function, etc.

Wherein, the original sample includes the user characteristic information of the setting user, the item characteristic information of the setting item and the set of behaviors performed by the setting user on the setting item, and there is a progressive relationship among the behaviors in the behavior set.

That is, one original sample records one or more logical user operations performed by a specific user a on a specific item b.

Further, the user feature information includes at least one of the following items: the set user's age, user gender, and the user's click data and purchase data within a certain period of time; the item feature information includes at least one of the following: Item: the item category, item brand, and number of clicks and exposures of the item within a certain period of time for the set item; the behavior set includes at least one of click, additional purchase, and purchase sorted according to the progressive relationship.

In this embodiment, optionally, according to different functions of the preset platform or types of user behavior predictions, the behavior set may also include: behaviors such as browsing, searching, or bookmarking.

In this embodiment, the progressive relationship between the various behaviors may be: behavior ordering with logical relationships; for example, if the behavior set a is {click, add purchase, purchase}, then the behavior Set a is a behavior set sorted according to the progressive relationship; correspondingly, if the behavior set b is {addition, purchase, click}, it is easy to understand that due to the conditions of the existing technology, when purchasing an item, additional purchase And the purchase behavior cannot occur before the click behavior, so the behavior set b is not a behavior set sorted according to the progressive relationship.

Optionally, generate an original sample set based on the historical behavior data of multiple users, including:

In the historical behavior data of multiple users, at least one candidate behavior set of the target user for the target item is obtained; through the logic processing module, at least one target behavior that satisfies the rationality of the data progression relationship is screened in each of the candidate behavior sets Set: Construct at least one original sample matching the target user according to the user characteristic information of the target user, the item characteristic information of the target item, and the at least one target behavior set.

Wherein, the target user is any one of the multiple users; further, in this embodiment, any one of the multiple users is first selected as the target user, and the above-mentioned original sample is performed on the current target user. After the construction behavior, select any user in the multi-user except the current target user as a new target user to continue to build a new original sample until all users in the multi-user have constructed a matching one or multiple original samples or until the original samples meeting the preset quantity requirements are constructed.

In this embodiment, for example, if the behavior set a is {click, add purchase, purchase}, then the behavior set a is a behavior set that satisfies the rationality of the data progression relationship; correspondingly, if the The above behavior set b is {add purchase, purchase, click}. It is easy to understand that, under the conditions of the existing technology, when purchasing an item, the behavior of additional purchase and purchase cannot occur before the behavior of clicking, so the behavior set b is not A collection of behaviors to satisfy the rationality of the data progression relationship.

In this embodiment, when the logic processing module is used to screen the candidate behavior sets, at least one preset target behavior set that satisfies the rationality of the data progressive relationship can be preset in the logic processing module, and then the at least A set of candidate behaviors is input into the logic processing module, and the logic processing module compares each set of candidate behaviors with the preset target behavior set in turn; if the set of candidate behaviors matches at least one preset target behavior set, Then output the candidate behavior set as the target behavior set; if the candidate behavior set does not match all preset target behavior sets, then delete the candidate behavior set.

Further, according to the user ID or other specific user identifiers, the user characteristic information matching the user identifier, the item characteristic information of the item corresponding to the user identifier and the at least one target behavior set can be spliced to form a The original sample matching the user ID; further, each user in the multi-user can construct at least one original sample matching the user ID; for example, the information contained in the original sample may be: {user a , female, item b, category c, click, add purchase, purchase}; among them, in the above information, user a is the target user, "female" is the user characteristic information of user a, and "category c" is the item of item b Feature information, "click, add purchase, purchase" is a collection of target behaviors.

S120. According to the behavior set in each original sample, generate a first behavior feature matrix, a second behavior feature matrix, and a third behavior feature matrix respectively corresponding to each original sample.

Wherein, the feature matrix in the first row may be a zero matrix, and the feature matrix in the second row and the feature matrix in the third row are both non-zero matrices.

Optionally, according to the behavior set in each original sample, generate the first behavior feature matrix, the second behavior feature matrix and the third behavior feature matrix respectively corresponding to each original sample, including: obtaining the current processing sample in each original sample The current behavior set in the set, and form a one-dimensional behavior matrix matching the current behavior set, wherein, the number of digits L of the one-dimensional behavior matrix is fixed, and each matrix bit corresponds to a set behavior;

Construct the first basic feature matrix of L*(L-1) order, sequentially obtain the first i-1 column data of the i-th column data in the one-dimensional behavior matrix, and fill it into the i-th row of the first basic feature matrix In the first i-1 column, the remaining position in the first basic feature matrix is filled with zeros to obtain the first row feature matrix, wherein the i is initialized to 1;

Construct the second basic feature matrix of L*L order, sequentially obtain the i-th column data in the one-dimensional behavior matrix, fill it into the i-th row and i-th column in the second basic feature matrix, and store the second basic feature matrix Fill the remaining positions with zeros to get the second row feature matrix;

Construct an identity matrix of order L*L as the third row characteristic matrix.

In this embodiment, first a one-dimensional behavior matrix can be generated according to the behavior sets in each original sample, and then a first behavior feature matrix, a second behavior feature matrix and a third behavior feature matrix can be generated according to the one-dimensional behavior matrix.

Specifically, when generating a one-dimensional behavior matrix based on the behavior set in each original sample, the behavior that occurred can be set to 1, and the behavior that did not occur can be set to 0, forming a one-dimensional behavior matrix composed of 0 and 1 ; Exemplarily, on the basis of S110, the target behavior set a that satisfies the rationality of the data progression relationship is {click, add purchase, purchase}, if in a certain behavior of the target user, the user clicks on item a Links to browse, and after browsing, the behavior of purchasing without adding to the shopping cart is directly carried out. In the above behaviors, the setting behavior of click and purchase behavior is 1, and the setting behavior of additional purchase is 0, which corresponds to the above The one-dimensional behavior matrix of the behavior set is (1, 0, 1). It should be noted that, in step S110, at least one target behavior set that satisfies the rationality of the data progression relationship is screened in each of the candidate behavior sets through the logical processing module, that is, the target behavior sets are all behavioral rationality Furthermore, since it is unreasonable to purchase without clicking in a certain behavior, there will be no irregularities such as (0, 0, 1) or (0, 1, 1) in the above operations. A one-dimensional behavior matrix that conforms to behavioral logic.

In a specific implementation of this embodiment, the target behavior set of the target user is set as {click, add purchase, purchase}, and the target user has click and purchase behaviors in the historical behavior data, but does not exist In the behavior of adding purchases, after the one-dimensional behavior matrix d(1, 0, 1) matching the current behavior set is formed through the above behaviors, it can be determined that the size of the corresponding first basic feature matrix is a 3*2 order matrix, First, obtain the first 0 data of the first column of the above matrix d, because the 0 data does not contain any data information, that is, the first row in the first basic feature matrix has no data information; correspondingly, the above matrix d is obtained later The data in the first 1 digit of the second column of data is "1", that is, the data contained in the second row in the first basic feature matrix is 1, and so on, the data contained in the third row in the first basic feature matrix is 1,0; then fill the above data into the corresponding positions of the first basic feature matrix, and fill the remaining positions in the first basic feature matrix with zeros, that is, the obtained first row feature matrix is:

When using the above-mentioned one-dimensional behavior matrix d(1, 0, 1) to construct the second feature matrix, it can be determined that the size of the corresponding second basic feature matrix is a 3*3 order matrix, and first the first The column data "1" is filled in the first row and the first column in the second basic feature matrix; correspondingly, the second column data "0" in the matrix d is filled in the second basic feature matrix In the second column of the row, the data "1" in the third column in the matrix d is filled into the third row and the third column in the second basic feature matrix, and the remaining positions in the second basic feature matrix are complemented Zero, the second line of the characteristic matrix is obtained as:

Further, when using the above-mentioned one-dimensional behavioral matrix d(1, 0, 1) to construct the third characteristic matrix, a 3*3 order identity matrix can be constructed as the third characteristic matrix.

S130. According to the user characteristic information of the set user in the original sample, set the item characteristic information of the item, the first behavior characteristic matrix, the second behavior characteristic matrix and the third behavior characteristic matrix to form a training sample set, and the second behavior characteristic matrix The matrix and the third row feature matrix are used as labeled data in the training samples.

Wherein, the training sample set includes training data and labeling data; specifically, the training data includes: user feature information of the set user, item feature information of the set item, the first behavior feature matrix, and the labeling data includes the second Behavior feature matrix and the third behavior feature matrix; further, the training data can be used to calculate the theoretical probability of user behavior prediction under the current model parameters; the label data can be used to calculate the actual probability of user behavior prediction, need It should be noted that the current model parameter condition can be any condition value set randomly.

S140. Use the training sample set to train the deep recommendation model to obtain a user behavior prediction model.

Wherein, the training sample set includes: user feature information of the set user, item feature information of the set item, a first behavior feature matrix, a second behavior feature matrix, and a third behavior feature matrix.

Optionally, use the training sample set to train the deep recommendation model to obtain a user behavior prediction model, including:

Obtain the target training sample in the training sample set, and input the target training sample into the deep recommendation model; through the sparse feature layer in the deep recommendation model, the user feature information of the target user in the target training sample, the item feature information of the target item and The first line of the target is to process the feature matrix to obtain the original sparse vector; the original sparse vector is processed through the dense embedding layer in the deep recommendation model to obtain a dense vector; the original sparse vector is obtained through the factorization layer in the deep recommendation model Perform logistic regression calculation with the dense vector to obtain the behavior prediction probability matrix; through the loss function layer in the depth recommendation model, according to the prediction probability matrix, the target second behavior feature matrix and the target third behavior feature matrix in the target training samples Calculate the loss function, and adjust the parameters of the deep recommendation model according to the loss function; return to the operation of obtaining the target training samples in the training sample set, until the user behavior prediction model is trained.

According to the technical solution of the embodiment of the present invention, the original sample set is generated according to the historical behavior data of multiple users, and the first behavior feature matrix and the second behavior feature matrix respectively corresponding to each original sample are generated according to the behavior set in each original sample. matrix and the third behavior feature matrix, and then according to the user feature information of the set user in the original sample, set the item feature information, the first behavior feature matrix, the second behavior feature matrix and the third behavior feature matrix to form a training The sample set and labeled data, and finally use the training sample set to train the deep recommendation model to obtain a user behavior prediction model, which provides a new way to build a user behavior prediction model. Based on this new user behavior prediction model, it can effectively improve user behavior. The accuracy of behavior prediction can further improve the hit rate of recommended items to users' actual needs, and effectively reduce the time consumption when users search for interesting information by themselves.

Embodiment two

Fig. 2a is a flow chart of a user behavior prediction method provided by Embodiment 2 of the present invention. This embodiment is based on the above-mentioned embodiment for refinement. In this embodiment, the training sample set is used to carry out the in-depth recommendation model. Training, the refinement of the user behavior prediction model is obtained as follows: obtain the target training samples in the training sample set, and input the target training samples into the deep recommendation model; The user feature information, the item feature information of the target item and the target first behavior feature matrix are processed to obtain the original sparse vector; the original sparse vector is processed through the dense embedding layer in the deep recommendation model to obtain the dense vector; through the deep recommendation model The factorization layer performs logistic regression calculation on the original sparse vector and dense vector to obtain the behavior prediction probability matrix; through the loss function layer in the depth recommendation model, according to the prediction probability matrix, the target No. The second row feature matrix and the target third row feature matrix calculate the loss function, and adjust the parameters of the deep recommendation model according to the loss function; return to the operation of obtaining the target training sample in the training sample set, until the user behavior prediction model is trained.

Correspondingly, as shown in Figure 2a, the method includes:

S210. Generate an original sample set according to the historical behavior data of multiple users.

Wherein, the original sample includes the user characteristic information of the setting user, the item characteristic information of the setting item and the set of behaviors performed by the setting user on the setting item, and there is a progressive relationship among the behaviors in the behavior set.

S220. According to the behavior set in each original sample, generate a first behavior feature matrix, a second behavior feature matrix, and a third behavior feature matrix respectively corresponding to each original sample.

S230. According to the user characteristic information of the set user in the original sample, set the item characteristic information of the item, the first behavior characteristic matrix, the second behavior characteristic matrix and the third behavior characteristic matrix to form a training sample set, and the second behavior characteristic matrix The matrix and the third row feature matrix are used as labeled data in the training samples.

S240. Obtain target training samples from the training sample set, and input the target training samples into the deep recommendation model.

As shown in Figure 2b, it should be noted that Sparse Features is the sparse feature layer in the following steps; correspondingly, Desnes Embedding is the dense embedding layer in the following steps, and the factorization layer in the following steps of the HM layer; where , Figure 2b also includes a Hide layer not included in the following steps, further, the Hide layer can be used to perform high-order crossover between feature matrices and extract deep-level information.

S250. Process the user feature information of the target user, the item feature information of the target item, and the target first behavior feature matrix in the target training sample through the sparse feature layer in the deep recommendation model to obtain an original sparse vector.

S260. Process the original sparse vector through the dense embedding layer in the deep recommendation model to obtain a dense vector.

Wherein, the dense vector can be converted into a dense vector by performing word2vec processing on the original sparse vector in the dense embedding layer.

S270. Perform logistic regression calculation on the original sparse vector and dense vector through the factorization layer in the deep recommendation model to obtain a behavior prediction probability matrix.

S280. Calculate the loss function through the loss function layer in the depth recommendation model according to the prediction probability matrix, the target second behavior feature matrix and the target third behavior feature matrix in the target training samples, and make a deep recommendation according to the loss function The parameters of the model are tuned.

Wherein, since the current model parameter condition can be any condition value set randomly, the current predicted prediction probability matrix is usually different from the actual probability of user behavior prediction calculated by labeling data, and then according to the difference between the two The difference of the parameters is adjusted using a loss function.

S290. Return to and execute the operation of obtaining the target training samples in the training sample set until the user behavior prediction model is obtained through training.

As shown in Figure 2b, the input data of Sparse Features contains sparse features and numerically dense features. DesnesEmbedding can map the features to k-dimensional vectors through the FM algorithm, and add the feature vectors in pairs to obtain the first-order cross weight. The multiplication operation obtains the second-order cross weight; then the FM layer combines the first-order and second-order cross weights of the input data and the corresponding feature vectors, and performs data transformation on the prediction results of the FM layer and the Hide layer in the output layer to obtain the current model parameter conditions It should be noted that since the current model parameter condition can be any conditional value set randomly, the current predicted prediction probability matrix is usually different from the actual user behavior prediction calculated by labeling data. The probabilities are different, and then use the loss function to adjust the parameters according to the difference between the two, and repeat the above operations until the final difference between the two is within the allowable range of error, that is, the user behavior prediction model is trained.

According to the technical solution of the embodiment of the present invention, the original sample set is generated according to the historical behavior data of multiple users, and the first behavior feature matrix and the second behavior feature matrix respectively corresponding to each original sample are generated according to the behavior set in each original sample. matrix and the third behavior feature matrix, and then according to the user feature information of the set user in the original sample, set the item feature information, the first behavior feature matrix, the second behavior feature matrix and the third behavior feature matrix to form a training The sample set and labeled data, and then obtain the target training samples in the training sample set, and input the target training samples into the deep recommendation model, through the sparse feature layer in the deep recommendation model, the user feature information of the target user in the target training samples, The item feature information of the target item and the target first behavior feature matrix are processed to obtain the original sparse vector, and then the original sparse vector is processed through the dense embedding layer in the deep recommendation model to obtain a dense vector, and the factor in the deep recommendation model is used to obtain the dense vector. The formula decomposition layer performs logistic regression calculation on the original sparse vector and dense vector, and after obtaining the behavior prediction probability matrix, through the loss function layer in the deep recommendation model, according to the prediction probability matrix and the target second behavior in the target training sample The feature matrix and the third behavior of the target feature matrix calculate the loss function, and adjust the parameters of the deep recommendation model according to the loss function, and finally return to the operation of obtaining the target training samples in the training sample set until the user behavior prediction model is trained, providing a A new way to build a user behavior prediction model. Based on the new user behavior prediction model, the accuracy of user behavior prediction can be effectively improved, and the hit rate of recommended items to the actual needs of users can be improved, and users are less interested in self-searching. Time consumption of information.

Embodiment three

Fig. 3 is a flow chart of a method for predicting user behavior provided by Embodiment 3 of the present invention. This embodiment is applicable to the situation of predicting user behavior, and the method can be executed by a user behavior prediction device. It can be implemented in the form of hardware and/or software, and the user behavior prediction device can be configured in a computer or server with a user behavior prediction function. As shown in Figure 3, the method includes:

S310. Obtain user feature information of the user to be predicted and item feature information of the item to be predicted.

Among them, the user characteristic information of the user to be predicted includes at least one of the following items: the age of the user to be predicted, the gender of the user, and the click data and purchase data of the user within a certain period of time; The item characteristic information of the item includes at least one of the following items: the item category of the item to be predicted, the item brand, and the number of clicks and exposures of the item within a certain period of time.

S320. Input the user feature information of the user to be predicted and the item feature information of the item to be predicted into the user behavior prediction model trained by the method of constructing the user behavior prediction model, and obtain a behavior prediction probability matrix of the user to be predicted for the item to be predicted.

Wherein, the specification of the behavior prediction probability matrix is determined by the number of behaviors to be predicted, that is, if the number of behaviors to be predicted is n, the specification of the behavior prediction probability matrix is n*n.

S330. Verify whether the item to be predicted is recommended to the user to be predicted according to the behavior prediction probability matrix.

则根据所述行为预测概率矩阵，点击行为发生的概率为第一行第一列对应的数值0.8，相应的，当发生点击行为时，加购行为发生的概率是点击行为概率*第二行第二例对应的数值，即0.8*0.2＝0.16,；当发生加购行为时，购买行为发生的概率为加购行为概率*第三行第三例对应的数值0.16*0.3＝0.048。In this embodiment, it is assumed that the behavior to be predicted is the probability that the user clicks on the item to be predicted, purchases and purchases, and the behavior prediction probability matrix obtained through the model is assumed to be

According to the behavior prediction probability matrix, the probability of click behavior is 0.8 corresponding to the value in the first row and first column. Correspondingly, when a click behavior occurs, the probability of additional purchase behavior is click behavior probability * second row No. The value corresponding to the second example is 0.8*0.2=0.16; when an additional purchase occurs, the probability of the purchase is the probability of additional purchase * the value corresponding to the third example in the third line 0.16*0.3=0.048.

In this embodiment, it is possible to verify whether the item to be predicted is recommended to the user to be predicted by setting a probability threshold, that is, when the probability of occurrence of the predicted behavior reaches the threshold, the item to be predicted is recommended to the user The user is to be predicted, otherwise no recommendation will be made.

In a specific implementation of this embodiment, on the basis of the above-mentioned behavior, when the target prediction behavior is set to click, assuming that the threshold of the click probability is 0.5, since 0.8 is greater than 0.5, the item to be predicted can be recommended to the The user to be predicted; further, if the target prediction behavior is set to click and purchase, assuming that the threshold of click and purchase probability is 0.2, the click and purchase probability obtained under the current conditions is 0.8*click weight+0.48*purchase weight, Wherein, the click weight and purchase weight can be adjusted according to actual application conditions, and when the click and purchase probabilities are greater than a preset threshold, the item to be predicted can be recommended to the user to be predicted.

In the technical solution of the embodiment of the present invention, by acquiring the user feature information of the user to be predicted and the item feature information of the item to be predicted, and then inputting the user feature information of the user to be predicted and the item feature information of the item to be predicted into the user behavior prediction model In the user behavior prediction model trained by the construction method, the behavior prediction probability matrix of the user to be predicted is obtained; finally, according to the behavior prediction probability matrix, it is verified whether the item to be predicted is recommended to the user to be predicted, Provides a new way to build a user behavior prediction model. Based on the new user behavior prediction model, the accuracy of user behavior prediction can be effectively improved, and the hit rate of recommended items to the user's actual needs can be improved, and the user's self-determination can be effectively reduced. Time spent searching for information of interest.

Embodiment four

FIG. 4 is a schematic structural diagram of an apparatus for constructing a user behavior prediction model provided by Embodiment 4 of the present invention. As shown in Figure 4, the device includes:

The original sample set generation module 410 is used to generate an original sample set according to the historical behavior data of multiple users. The original sample includes the user feature information of the set user, the item feature information of the set item, and the set user's execution of the set item. Behavior set, there is a progressive relationship between each behavior in the behavior set;

The feature matrix generation module 420 is used to generate the first behavior feature matrix, the second behavior feature matrix and the third behavior feature matrix respectively corresponding to each original sample according to the behavior set in each original sample;

The training sample set generation module 430 is used to set the item feature information, the first behavior feature matrix, the second behavior feature matrix and the third behavior feature matrix of the item according to the user feature information of the set user in the original sample to form a training sample set. The sample set, the feature matrix in the second row and the feature matrix in the third row are used as labeled data in the training sample;

The model training module 440 is configured to use the training sample set to train the deep recommendation model to obtain a user behavior prediction model.

According to the technical solution of the embodiment of the present invention, the original sample set is generated according to the historical behavior data of multiple users, and the first behavior feature matrix and the second behavior feature matrix respectively corresponding to each original sample are generated according to the behavior set in each original sample. matrix and the third behavior feature matrix, and then according to the user feature information of the set user in the original sample, set the item feature information, the first behavior feature matrix, the second behavior feature matrix and the third behavior feature matrix to form a training The sample set and labeled data, and finally use the training sample set to train the deep recommendation model to obtain a user behavior prediction model, which provides a new way to build a user behavior prediction model. Based on this new user behavior prediction model, it can effectively improve user behavior. The accuracy of behavior prediction can further improve the hit rate of recommended items to users' actual needs, and effectively reduce the time consumption when users search for interesting information by themselves.

On the basis of the foregoing embodiments, the original sample set generation module 410 may include:

The alternative behavior set acquisition unit is used to acquire at least one alternative behavior set of the target user for the target item in the historical behavior data of multiple users;

A behavior set screening unit, configured to filter at least one target behavior set that satisfies the rationality of the data progression relationship in each of the candidate behavior sets through a logic processing module;

An original sample construction unit, configured to construct at least one original sample matching the target user according to the user characteristic information of the target user, the item characteristic information of the target item, and the at least one target behavior set.

On the basis of the foregoing embodiments, the feature matrix generating module 420 may include:

Behavior matrix generation unit, used to obtain the current behavior set in the current processing sample in each original sample, and form a one-dimensional behavior matrix matching the current behavior set, wherein, the number of digits L of the one-dimensional behavior matrix is fixed , each matrix bit corresponds to setting behavior;

The first row feature matrix construction unit is used to construct the first basic feature matrix of L*(L-1) order, and sequentially obtain the first i-1 column data of the i-th column data in the one-dimensional behavior matrix, and fill it to In the first i-1 column of the i-th row in the first basic feature matrix, and zero-fill the remaining positions in the first basic feature matrix to obtain the first row feature matrix, wherein the i is initialized to 1;

The second behavioral feature matrix construction unit is used to construct the second basic feature matrix of L*L order, sequentially obtain the i-th column data in the one-dimensional behavioral matrix, and fill it into the i-th row of the second basic feature matrix column, and zero-fill the remaining position in the second basic feature matrix to obtain the second row feature matrix;

The third row is a characteristic matrix construction unit for constructing an identity matrix of order L*L, as the third row is a characteristic matrix.

On the basis of the foregoing embodiments, the model training module 440 may include:

A target training sample acquisition unit, configured to acquire target training samples in the training sample set, and input the target training samples into the deep recommendation model;

The original sparse vector generation unit is used to process the user feature information of the target user in the target training sample, the item feature information of the target item, and the target first behavior feature matrix through the sparse feature layer in the deep recommendation model to obtain the original sparse vector ;

The dense vector generation unit is used to process the original sparse vector through the dense embedding layer in the depth recommendation model to obtain a dense vector;

The behavior prediction probability matrix acquisition unit is used to perform logistic regression calculation on the original sparse vector and dense vector through the factorization layer in the depth recommendation model to obtain the behavior prediction probability matrix;

The parameter adjustment unit is used to calculate the loss function according to the prediction probability matrix, the target second behavior feature matrix and the target third behavior feature matrix in the target training sample through the loss function layer in the depth recommendation model, and according to the loss The function adjusts the parameters of the deep recommendation model;

Returning to the execution unit, used to return and execute the operation of obtaining the target training samples in the training sample set until the user behavior prediction model is trained.

The device for constructing a user behavior prediction model provided by an embodiment of the present invention can execute the method for constructing a user behavior prediction model provided by any embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the method.

Embodiment five

FIG. 5 is a schematic structural diagram of a user behavior prediction device provided by Embodiment 5 of the present invention. As shown in Figure 5, the device includes:

Feature information acquisition module 510, configured to acquire user feature information of the user to be predicted and item feature information of the item to be predicted;

The probability matrix acquisition module 520 is used to input the user feature information of the user to be predicted and the item feature information of the item to be predicted into the user behavior prediction model trained by the construction method of the user behavior prediction model, and obtain the predicted item of the user to be predicted The behavior prediction probability matrix of ;

The recommendation verification module 530 is configured to verify whether the item to be predicted is recommended to the user to be predicted according to the behavior prediction probability matrix.

In the technical solution of the embodiment of the present invention, by acquiring the user feature information of the user to be predicted and the item feature information of the item to be predicted, and then inputting the user feature information of the user to be predicted and the item feature information of the item to be predicted into the user behavior prediction model In the user behavior prediction model trained by the construction method, the behavior prediction probability matrix of the user to be predicted is obtained; finally, according to the behavior prediction probability matrix, it is verified whether the item to be predicted is recommended to the user to be predicted, Provides a new way to build a user behavior prediction model. Based on the new user behavior prediction model, the accuracy of user behavior prediction can be effectively improved, and the hit rate of recommended items to the user's actual needs can be improved, and the user's self-determination can be effectively reduced. Time spent searching for information of interest.

Embodiment six

FIG. 6 shows a schematic structural diagram of an electronic device 10 that can be used to implement an embodiment of the present invention. Electronic device is intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other suitable computers. Electronic devices may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (eg, helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are by way of example only, and are not intended to limit implementations of the inventions described and/or claimed herein.

As shown in FIG. 6, the electronic device 10 includes at least one processor 11, and a memory connected in communication with the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores There is a computer program executable by at least one processor, and the processor 11 can operate according to a computer program stored in a read-only memory (ROM) 12 or loaded from a storage unit 18 into a random access memory (RAM) 13. Various appropriate actions and processes are performed. In the RAM 13, various programs and data necessary for the operation of the electronic device 10 are also stored. The processor 11 , ROM 12 , and RAM 13 are connected to each other through a bus 14 . An input/output (I/O) interface 15 is also connected to the bus 14 .

Multiple components in the electronic device 10 are connected to the I/O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit 19 allows the electronic device 10 to exchange information/data with other devices through a computer network such as the Internet and/or various telecommunication networks.

Processor 11 may be various general and/or special purpose processing components having processing and computing capabilities. Some examples of processor 11 include, but are not limited to, central processing units (CPUs), graphics processing units (GPUs), various dedicated artificial intelligence (AI) computing chips, various processors that run machine learning model algorithms, digital signal processing processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes various methods and processes described above, for example, a method for constructing a user behavior prediction model, or a method for user behavior prediction.

Correspondingly, the method for constructing the user behavior prediction model includes: according to the historical behavior data of multiple users, generating an original sample set, the original sample includes the user feature information of the set user, the item feature information of the set item, and the set user feature information. For the behavior set executed by the set item, there is a progressive relationship between the behaviors in the behavior set; according to the behavior set in each original sample, generate the first behavior feature matrix and the second behavior feature matrix corresponding to each original sample and the third behavior feature matrix; according to the user feature information of the set user in the original sample, set the item feature information of the item, the first behavior feature matrix, the second behavior feature matrix and the third behavior feature matrix to form a training sample set , the second behavior feature matrix and the third behavior feature matrix are used as labeled data in the training samples; the deep recommendation model is trained using the training sample set to obtain the user behavior prediction model.

The user behavior prediction method includes: obtaining user feature information of the user to be predicted and item feature information of the item to be predicted; inputting the user feature information of the user to be predicted and the item feature information of the item to be predicted into the user behavior prediction model through construction In the user behavior prediction model obtained by the method training, the behavior prediction probability matrix of the user to be predicted is obtained; according to the behavior prediction probability matrix, it is verified whether the item to be predicted is recommended to the user to be predicted.

In some embodiments, the method for constructing the user behavior prediction model and the user behavior prediction method can be implemented as computer programs, which are tangibly contained in a computer-readable storage medium, such as the storage unit 18 . In some embodiments, part or all of the computer program may be loaded and/or installed on the electronic device 10 via the ROM 12 and/or the communication unit 19 . When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the data reduction method in the distributed training described above can be performed. Alternatively, in other embodiments, the processor 11 may be configured in any other appropriate way (for example, by means of firmware) to execute the method for constructing the user behavior prediction model and the user behavior prediction method.

Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips Implemented in a system of systems (SOC), load programmable logic device (CPLD), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include being implemented in one or more computer programs executable and/or interpreted on a programmable system including at least one programmable processor, the programmable processor Can be special-purpose or general-purpose programmable processor, can receive data and instruction from storage system, at least one input device, and at least one output device, and transmit data and instruction to this storage system, this at least one input device, and this at least one output device an output device.

Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, so that the computer program causes the functions/operations specified in the flowcharts and/or block diagrams to be implemented when executed by the processor. A computer program may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of the present invention, a computer readable storage medium may be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus or device. A computer readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer readable storage medium may be a machine readable signal medium. More specific examples of machine-readable storage media would include one or more wire-based electrical connections, portable computer discs, hard drives, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), optical fiber, compact disk read only memory (CD-ROM), optical storage, magnetic storage, or any suitable combination of the foregoing.

In order to provide interaction with the user, the systems and techniques described herein can be implemented on an electronic device having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display)) for displaying information to the user. monitor); and a keyboard and pointing device (eg, a mouse or a trackball) through which the user can provide input to the electronic device. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and can be in any form (including Acoustic input, speech input or, tactile input) to receive input from the user.

The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., as a a user computer having a graphical user interface or web browser through which a user can interact with embodiments of the systems and techniques described herein), or including such backend components, middleware components, Or any combination of front-end components in a computing system. The components of the system can be interconnected by any form or medium of digital data communication, eg, a communication network. Examples of communication networks include: local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

A computing system can include clients and servers. Clients and servers are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also known as a cloud computing server or a cloud host. It is a host product in the cloud computing service system to solve the problems of difficult management and weak business expansion in traditional physical hosts and VPS services. defect.

It should be understood that steps may be reordered, added or deleted using the various forms of flow shown above. For example, each step described in the present invention may be executed in parallel, sequentially, or in a different order, as long as the desired result of the technical solution of the present invention can be achieved, there is no limitation herein.

Claims (10)

1. A method for building a user behavior prediction model, comprising: According to the historical behavior data of multiple users, the original sample set is generated. The original sample includes the user feature information of the set user, the item feature information of the set item, and the set of behaviors performed by the set user on the set item. There is a progressive relationship between behaviors; According to the behavior set in each original sample, generate the first behavior characteristic matrix, the second behavior characteristic matrix and the third behavior characteristic matrix respectively corresponding to each original sample; According to the user characteristic information of the set user in the original sample, set the item characteristic information, the first behavior characteristic matrix, the second behavior characteristic matrix and the third behavior characteristic matrix of the item to form a training sample set, the second behavior characteristic matrix and The third row feature matrix is used as the labeled data in the training sample; Use the training sample set to train the deep recommendation model to obtain the user behavior prediction model. 2. The method according to claim 1, wherein the user characteristic information includes at least one of the following: The user's user age, user gender, and user's click data and purchase data within a certain period of time are set; The item feature information includes at least one of the following: The item category, item brand, and number of hits and exposures of the item within a certain period of time for the set item; The behavior set includes at least one of click, additional purchase and purchase sorted according to the progressive relationship. 3. The method according to claim 1, characterized in that, according to the historical behavior data of multiple users, the original sample set is generated, comprising: In the historical behavior data of multiple users, at least one candidate behavior set of the target user for the target item is obtained; Screening at least one target behavior set that satisfies the rationality of the data progression relationship in each of the candidate behavior sets through a logic processing module; Construct at least one original sample matching the target user according to the user feature information of the target user, the item feature information of the target item, and the at least one target behavior set. 4. according to the method described in any one of claim 1-3, it is characterized in that, according to the behavior set in each original sample, generate the first behavior characteristic matrix, the second behavior characteristic matrix and the first behavior characteristic matrix respectively corresponding to each original sample The three-line feature matrix includes: Obtain the current behavior set in the current processing sample in each original sample, and form a one-dimensional behavior matrix matching the current behavior set, wherein, the number of bits L of the one-dimensional behavior matrix is fixed, and each matrix bit corresponds to a set fixed behavior; Construct the first basic feature matrix of L*(L-1) order, sequentially obtain the first i-1 column data of the i-th column data in the one-dimensional behavior matrix, and fill it into the i-th row of the first basic feature matrix In the first i-1 column, the remaining position in the first basic feature matrix is filled with zeros to obtain the first row feature matrix, wherein the i is initialized to 1; Construct the second basic feature matrix of L*L order, sequentially obtain the i-th column data in the one-dimensional behavior matrix, fill it into the i-th row and i-th column in the second basic feature matrix, and store the second basic feature matrix Fill the remaining positions with zeros to get the second row feature matrix; Construct an identity matrix of order L*L as the third row characteristic matrix. 5. The method according to claim 1, wherein the deep recommendation model is trained using a training sample set to obtain a user behavior prediction model, comprising: Obtain target training samples from the training sample set, and input the target training samples into the deep recommendation model; Through the sparse feature layer in the deep recommendation model, the user feature information of the target user in the target training sample, the item feature information of the target item, and the target first behavior feature matrix are processed to obtain the original sparse vector; Process the original sparse vector through the dense embedding layer in the deep recommendation model to obtain a dense vector; Perform logistic regression calculation on the original sparse vector and dense vector through the factorization layer in the deep recommendation model to obtain a behavior prediction probability matrix; Through the loss function layer in the depth recommendation model, the loss function is calculated according to the prediction probability matrix, the target second behavior feature matrix and the target third behavior feature matrix in the target training sample, and the depth recommendation model is performed according to the loss function Parameter adjustment; Return to the operation of obtaining the target training samples in the training sample set until the user behavior prediction model is trained. 6. A user behavior prediction method, characterized in that, comprising: Obtain the user feature information of the user to be predicted and the item feature information of the item to be predicted; Input the user feature information of the user to be predicted and the item feature information of the item to be predicted into the user behavior prediction model trained by the method according to any one of claims 1-5, and obtain the behavior of the user to be predicted on the item to be predicted Prediction probability matrix; According to the behavior prediction probability matrix, verify whether the item to be predicted is recommended to the user to be predicted. 7. A device for constructing a user behavior prediction model, comprising: The original sample set generation module is used to generate an original sample set according to the historical behavior data of multiple users. The original sample includes the user feature information of the set user, the item feature information of the set item, and the set user's execution of the set item. Behavior collection, there is a progressive relationship between the behaviors in the behavior collection; Feature matrix generating module, for generating the first behavior feature matrix, the second behavior feature matrix and the third behavior feature matrix respectively corresponding to each original sample according to the behavior set in each original sample; The training sample set generation module is used to set the item feature information, the first behavior feature matrix, the second behavior feature matrix and the third behavior feature matrix of the item according to the user feature information of the set user in the original sample to form a training sample set, the second row feature matrix and the third row feature matrix are used as labeled data in the training samples; The model training module is used to use the training sample set to train the deep recommendation model to obtain the user behavior prediction model. 8. A user behavior prediction device, characterized in that it comprises: A feature information acquisition module, configured to acquire user feature information of the user to be predicted and item feature information of the item to be predicted; The probability matrix acquisition module is used to input the user characteristic information of the user to be predicted and the item characteristic information of the item to be predicted into the user behavior prediction model obtained by training the method according to any one of claims 1-5, and obtain the user behavior prediction model to be predicted. Predict the user's behavior prediction probability matrix for the predicted item; A recommendation verification module, configured to verify whether the item to be predicted is recommended to the user to be predicted according to the behavior prediction probability matrix. 9. An electronic device, characterized in that the electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, the computer program is executed by the at least one processor, so that the at least one processor can perform any one of claims 1-5 The method for constructing a user behavior prediction model, or, implementing the user behavior prediction method described in claim 6 . 10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method described in any one of claims 1-5 when executed. A method for constructing a user behavior prediction model, or realizing the user behavior prediction method described in claim 6 .

Images