A recommendation ranking optimization method based on user click behavior prediction

Abstract

The present invention discloses a recommendation ranking optimization method based on user click behavior prediction, which belongs to the field of information optimization processing technology. It includes the following steps: (1) obtaining historical exposure behavior data and corresponding databases, and arranging to obtain sample data sets; (2) performing association processing on the obtained sample data sets to obtain related data; (3) performing data cleaning on the obtained related data, and filtering abnormal data; (4) performing data processing operations on the obtained cleaned data; (5) performing time data processing and text data processing on the obtained processed data; (6) training the obtained processed data with a GBDT model; (7) predicting and sorting new data on the obtained GBDT model. It is applied on the Internet platform and predicts the online public opinion on the online community, increases the exposure of valuable content, improves the content dissemination and interaction, and curbs negative content to prevent adverse effects.

Description

Technical Field

[0001] The present invention belongs to the technical field of information optimization processing, and in particular, relates to a recommendation ranking optimization method based on user click behavior prediction. Background Art

[0002] Public opinion on the Internet about emergencies refers to the sum of information containing a variety of emotions, attitudes and opinions released by the public around impending or existing emergencies using the Internet as a platform and with the help of computers and other auxiliary communication tools. The popularity of the Internet has made the influence of the Internet increasingly stronger. The frequency and harmfulness of public opinion on emergencies on the Internet have shown a clear upward trend, and have shown the characteristics of suddenness, diversity and complexity, resulting in rapid spread of public opinion and wide social impact. The important characteristics of public opinion on the Internet about emergencies at this stage are a large number of netizens, fast speed, diverse opinions and difficulty in control, which also brings great difficulties to the development of related public opinion monitoring, prediction, early warning and other work.

[0003] In the Web2.0 era, with the internal and external reasons of cognitive surplus and technological empowerment, the decentralized model allows users to quickly produce and distribute content through Weibo. Weibo has become an important window for the public to understand current affairs and express their opinions due to its strong grassroots nature, convenient publishing, diverse content, and strong interactivity. These characteristics make Weibo an excellent tool for the dissemination of online public opinion on emergencies, and on the other hand, they also provide a basis for scholars to use information technology to predict online public opinion on emergencies. Therefore, based on Weibo data, using existing information technology means to effectively predict online public opinion on emergencies is of great significance for the accurate judgment of emergencies, auxiliary emergency decision-making, and the maintenance of social stability. Summary of the invention

[0004] Problem to be solved

[0005] In view of the problems existing in the above-mentioned prior art, the present invention provides a recommendation ranking optimization method based on user click behavior prediction, which is applied on the Internet platform and predicts the online public opinion in the online community. It can increase the exposure of valuable content, improve the dissemination and interaction of the content, and curb negative content to prevent further adverse effects.

[0006] Technical Solution

[0007] To solve the above problems, the present invention adopts the following technical solutions.

[0008] A recommendation ranking optimization method based on user click behavior prediction includes the following steps:

[0009] (1) Obtain historical exposure behavior data and the corresponding database, and organize them to obtain a sample data set;

[0010] (2) performing association processing on the sample data set obtained in step (1) to obtain relevant data;

[0011] (3) Clean the relevant data obtained in step (2) and filter out abnormal data;

[0012] (4) performing data processing operations on the cleaned data obtained in step (3);

[0013] (5) performing time data processing and text data processing on the processed data obtained in step (4);

[0014] (6) Training the GBDT model on the processed data obtained in step (5);

[0015] (7) Predict and sort the new data using the GBDT model obtained in step (6).

[0016] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0017] The historical exposure behavior data described in step (1) is the exposure record of each blog and recommended blog obtained based on the user's exposure click record and the behavior mark of whether a click occurred.

[0018] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0019] The relevant data described in step (2) are multiple dimensional data of each blog and recommended blog;

[0020] The multi-dimensional data include recommendation strategy, title, description, publishing time, last modification time, all blog likes, all blog collections, all blog comments, all blog shares, blog likes in the last ten days, blog collections in the last ten days, blog comments in the last ten days, blog shares in the last ten days, blog post length, blog author registration, blog author ranking, blog author visits and blog author posting frequency.

[0021] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0022] The abnormal data described in step (3) is blog post data of abnormal detection or abnormal state;

[0023] The blog post data of abnormal detection or abnormal status includes blog post data generated by non-essential users, wherein non-essential users include robots, zombie fans, advertising users, and users whose fans are lower than a set value.

[0024] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0025] The data processing described in step (4) includes null value filling processing and type data one-hot encoding;

[0026] The described empty value filling process is to fill in the missing values ​​in the blog post data;

[0027] The one-hot encoding of the type data is to perform eigenvalue digital processing on the type data in the blog post data;

[0028] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0029] The time data processing in step (5) is as follows:

[0030] Convert the timestamp of the processed data.

[0031] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0032] The text data processing in step (5) is as follows:

[0033] The processed data is segmented, stop words are filtered, and doc2vec word vectors are converted.

[0034] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0035] In step (6), the processed data needs to be labeled before training the GBDT model.

[0036] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0037] The weak form of the boundary element formula for the prediction regression of the GBDT model in step (6) is as follows:

[0038]

[0039] Where k(T) represents the boundary change rate; T is the number of user clicks; G(x, y) is the weight function, which takes the basic solution of the Green's function and feeds it back to the training of the GBDT model, where x and y are the horizontal and vertical coordinates of the Cartesian space in the GBDT model respectively; q(x) is the weight value of the doc2vec word vector in the GBDT model, Ω is the calculation domain of the transient nonlinear heat conduction equation; Γ is the boundary of the calculation domain Ω; n is the number of vector conversions in step (5); ρ represents the fixed coefficient of the Green's function and is equal to 1.2; c represents the fixed coefficient of the Green's function and is equal to 0.5.

[0040] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0041] The sorting described in step (7) is to sort the predicted data in reverse order using the probability of being predicted to be 1.

[0042] Beneficial Effects

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The recommendation ranking optimization method designed by this application based on the prediction of user click behavior is applied to the Internet platform and predicts the information content on the online blog. It can increase the exposure of valuable content, improve the dissemination and interaction of the content, and curb negative content to prevent further adverse effects. Specifically, the training of the GBDT model is creatively introduced, and the weak form formula of the boundary element of the prediction regression is optimized. Due to the introduction of the weight function, the problem of improving the accuracy and stability of the prediction results is effectively solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flowchart of the recommendation ranking optimization method based on user click behavior prediction of the present invention;

[0046] Figure 2 A structural block diagram of an electronic device provided in an embodiment of the present application.

[0047] Figure 3 The structure diagram of a computer system suitable for implementing a method according to an embodiment of the present disclosure is shown in FIG. DETAILED DESCRIPTION

[0048] The present invention is further described below in conjunction with specific embodiments.

[0049] Background: In a recommendation system, related items under an item are recalled from multiple recall sources, and different strategies may recall different numbers of recommended items. These items need to be presented to users in a certain order to maximize the click rate.

[0050] Currently, recommendation sorting uses a fixed strategy sorting logic. The logic is fixed and cannot achieve the best recommendation effect. Therefore, in order to avoid using fixed sorting logic for sorting, it is necessary to design a sorting model for recommendation results to improve the click-through rate.

[0051] like Figure 1 As shown, the recommendation ranking optimization method based on user click behavior prediction of this embodiment includes the following steps:

[0052] (1) Obtain historical exposure behavior data and the corresponding database, and organize them to obtain a sample data set;

[0053] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0054] The historical exposure behavior data described in step (1) is the exposure record of each blog and recommended blog obtained based on the user's exposure click record and the behavior mark of whether a click occurred.

[0055] Specifically, building a sample data set is conducive to the subsequent training of the GBDT model and improving work efficiency; at the same time, with the development of social urbanization, my country's urbanization level is getting higher and higher, the scale of cities is getting larger and larger, and the operation of cities is getting more and more complicated. How to ensure social stability and the stability of the entire society has become a very important topic for urban managers. In the traditional processing method, urban managers discover and deal with problems through reports from grassroots grid members. In this way, problems are not discovered in time or are not reported until they have developed to a certain degree of severity (for example, mass incidents occur), resulting in high problem handling costs and causing very bad social impacts. This application uses a variety of methods to collect social public opinion (mainly Weibo data commonly used in China), and uses the GBDT model to predict the trend of problem development through the big data collection of social public opinion, so as to help urban authorities deal with problems in a targeted manner, solve problems in the early stages, and maintain social stability. In addition, it should be reminded that historical exposure behavior data can be obtained by contacting suppliers of public opinion data. For example, Weibo data can be obtained by analyzing the hot search keyword rankings.

[0056] (2) The sample data set obtained in step (1) is subjected to association processing. Here, the association processing method is to establish a mapping relationship between the user and the feature data. The conventional ORM relationship mapping (ORM stands for Object Relational Mapping) obtains the relevant data, that is, to associate a specific user feature with the data generated by it, which is equivalent to creating a file;

[0057] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0058] The relevant data described in step (2) are multiple dimensional data of each blog and recommended blog;

[0059] The multi-dimensional data include recommendation strategy, title, description, publishing time, last modification time, all blog likes, all blog collections, all blog comments, all blog shares, blog likes in the last ten days, blog collections in the last ten days, blog comments in the last ten days, blog shares in the last ten days, blog post length, blog author registration, blog author ranking, blog author visits and blog author posting frequency.

[0060] For the above steps, you can also improve the effect by the following expansion methods:

[0061] A sample data set is obtained, wherein the sample data set includes positive sample data with positive labels and unlabeled unlabeled sample data; specifically, the content that conforms to multiple dimensional data is positively labeled and recorded as positive sample data, otherwise, it is classified as unlabeled unlabeled sample data to be further enriched.

[0062] (3) Clean the relevant data obtained in step (2) and filter out abnormal data. The purpose of this operation is to reduce the number of samples and noise, and to identify and process missing data, incomplete data, and unreasonable data;

[0063] At the same time, the specific operation methods are as follows:

[0064] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0065] The abnormal data described in step (3) is blog post data of abnormal detection or abnormal state;

[0066] The blog post data of abnormal detection or abnormal status includes blog post data generated by non-essential users, wherein non-essential users include robots, zombie fans, advertising users, and users whose fans are lower than a set value.

[0067] Furthermore, the method of data cleaning is relatively conventional, for example, it can be achieved by using the method mentioned in the following patent, patent number: CN201810143012.8, publication number: CN108363782A, which discloses a data cleaning method and a data cleaning system, which mentions the following operation method:

[0068] “Step S10: selecting a data source to be cleaned from heterogeneous data sources through a graphical interface; wherein the heterogeneous data sources include text files and database data;

[0069] Step S11: Editing data cleaning rules through the graphical interface;

[0070] Step S12: performing data cleaning through the graphical interface.

[0071] (4) performing data processing operations on the cleaned data obtained in step (3);

[0072] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0073] The data processing described in step (4) includes null value filling processing and type data one-hot encoding;

[0074] The described empty value filling process is to fill in the missing values ​​in the blog post data;

[0075] The one-hot encoding of the type data is to perform eigenvalue digital processing on the type data in the blog post data.

[0076] The empty value filling process can be improved in the database, or the following method can be used:

[0077] Fill the missing values ​​according to the attribute type of the missing values ​​in the blog data and the average value of the values ​​in the blog data

[0078] The type data is one-hot encoded, and the details are as follows:

[0079] Onehot encoding is also called unique heat encoding. It is a single-bit valid encoding. It mainly uses an N-bit state register to encode N states. Each state has its own independent register bit, and only one bit is valid at any time. Onehot encoding is the representation of categorical variables as binary vectors. This first requires mapping categorical values ​​to integer values. Then, each integer value is represented as a binary vector, which is a zero value except for the index of the integer, which is marked as 1.

[0080] (5) performing time data processing and text data processing on the processed data obtained in step (4);

[0081] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0082] The time data processing in step (5) is as follows:

[0083] Convert the timestamp of the processed data.

[0084] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0085] The text data processing in step (5) is as follows:

[0086] The processed data is segmented, stop words are filtered, and doc2vec word vectors are converted.

[0087] It should be noted that

[0088] The timestamp conversion can be done in the following ways:

[0089] Get the offset of the specified time data, including the steps of finding the timestamp of Kafka and getting the offset of the specified time data through the timestamp of Kafka; set the specified offset for Kafka; call the Kafka Consumer in the KafkaApi and use the seek method to jump to the specified Offset of the specified Partition for consumption; start the Kafka consumer program to make the set time consistent with the timestamp corresponding to the time data.

[0090] The text data processing method, specifically, the word segmentation method, is processed according to user characteristics and blog post characteristics, wherein user characteristics include activity characteristics within the recent time T, self characteristics, fan characteristics, social characteristics, technical characteristics, and time characteristics, wherein blog post characteristics include basic characteristics, time characteristics, text characteristics, and topic influence characteristics, and stop words also include punctuation marks, numbers, email addresses, phone numbers, dates, URLs, and other elements;

[0091] Among them, doc2vec word vector conversion technology is relatively mature, for example: academic paper 1 (Hu Yameng. Research on text sentiment analysis based on dictionary and Doc2 vec fusion [D]. Wuhan Post and Telecommunications Research Institute.), academic paper 2 (Yue Wenying. Chat content filtering based on Doc2 Vec and SVM [J]. Computer System Applications, 2018, 027(007): 127-132.), academic paper 3 (Jia Xiaoting, Wang Mingyang, Cao Yu. Research on Chinese single document automatic summarization method combining Doc2Vec and improved clustering algorithm [J]. Data Analysis and Knowledge Discovery, 2018, 2(2): 10.), etc. At the same time, the following methods can be used:

[0092] "According to the doc2vec model, each target text in the target text set is subjected to n vector conversions to obtain each target vector set; wherein the target vector set includes n target vectors; each target vector set is similarly matched with a sentence vector generated according to the input text, and the target text corresponding to the target vector set with the highest matching degree with the sentence vector is extracted as the matching text."

[0093] (6) Training the GBDT model on the processed data obtained in step (5);

[0094] In step (6), the processed data needs to be labeled before training the GBDT model. Specifically, whether the user clicks the data is used as a 0 or 1 label based on the processed data.

[0095] In the above-mentioned recommendation ranking optimization method based on user click behavior prediction,

[0096] The weak form of the boundary element formula for the prediction regression of the GBDT model in step (6) is as follows:

[0097]

[0098] Where k(T) represents the boundary change rate; T is the number of user clicks; G(x, y) is the weight function, which takes the basic solution of the Green's function and feeds it back to the training of the GBDT model, where x and y are the horizontal and vertical coordinates of the Cartesian space in the GBDT model respectively; q(x) is the weight value of the doc2vec word vector in the GBDT model, Ω is the calculation domain of the transient nonlinear heat conduction equation; Γ is the boundary of the calculation domain Ω, and n is the number of vector conversions in step (5); ρ represents the fixed coefficient of the Green's function and is equal to 1.2, and c represents the fixed coefficient of the Green's function and is equal to 0.5.

[0099] It should be noted that model methods such as XGBOOST, RF, LR, and SVM can also be trained.

[0100] Further analysis shows that when training each regression tree of the GBDT model, a positive sample training subset is constructed based on the positive sample data in the sample data set, and a negative sample training subset is constructed by sampling the unlabeled sample data in the sample data set. The positive sample training subset is combined with the multiple negative sample training subsets to obtain the training set of the current regression tree, and the current regression tree is trained based on the training set of the current regression tree, and then a gradient boosting decision tree GBDT model is constructed based on each regression tree. For example: the set is: [Algorithm 1, Algorithm 2, Algorithm 3], the set of hyperparameter combinations is: [hyperparameter combination 1, hyperparameter combination 2, ..., hyperparameter combination 10], and the training set includes: training set 1, training set 2, ..., training set 8. Then, selecting "algorithm 1 + hyperparameter combination 1 + training set 1" can determine a candidate decision tree model, selecting "algorithm 2 + hyperparameter combination 1 + training set 1" can determine another candidate decision tree model, selecting "algorithm 1 + hyperparameter combination 2 + training set 1" can determine another candidate decision tree model, selecting "algorithm 1 + hyperparameter combination 1 + training set 2" can determine another candidate decision tree model, and so on. At the same time, the weight function is optimized by combining the algorithm described in (II), and the formula is as follows:

[0101]

[0102] In the formula, F(z, y) represents the optimized value of the weight function and is fed back to the weight value of the doc2vec word vector; (z, y) represents the vertical coordinate value and the vertical coordinate value of the Cartesian space in the GBDT model respectively; r is the distance between the source point and the field point in the Cartesian space in the GBDT model; G(x, y) is the same as the weight function in Formula I; b(x) represents the lateral search direction function of the Cartesian space in the GBDT model; r α It represents the exponential ratio between the r value and the unknown coefficient of the weight function; α is the unknown coefficient; dr represents the differential of the r value.

[0103] In the above steps (5) and (6), onehot encoding, timestamp, and doc2vec are combined into the GBDT model. At the same time, the training data contains text information such as the title and description of the blog post. The machine cannot learn from text information alone. We use doc2vec to convert the text into a vectorized representation.

[0104] (7) The GBDT model obtained in step (6) predicts and sorts new data. Specifically, the sorting described in step (7) is to sort the predicted data in reverse order using the probability of prediction being 1. The sorting is performed by training the model, which eliminates the fixed rows that previously used a single sorting logic. The recommended items with a high click probability can be moved forward to increase the possibility of clicks, improve the click-through rate, and optimize the user experience.

[0105] In summary, the recommendation ranking optimization method designed by this application based on user click behavior prediction is applied to the Internet platform and predicts the information content on online blogs. It can increase the exposure of valuable content, improve the dissemination and interaction of content, and curb negative content to prevent further adverse effects. Specifically, the training of the GBDT model is creatively introduced, and the weak form formula of the boundary element of the prediction regression is optimized. Due to the introduction of the weight function, the problem of improving the accuracy and stability of the prediction results is effectively solved.

[0106] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0107] It is understandable that the related features in the above methods and devices can be referenced to each other. In addition, the "first", "second" and the like in the above embodiments are used to distinguish the embodiments, but do not represent the advantages and disadvantages of the embodiments.

[0108] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0109] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the description of the above specific languages ​​is for disclosing the best mode of the present invention.

[0110] Figure 2 A structural block diagram of an electronic device according to an embodiment of the present application is shown.

[0111] The above-mentioned implementation method describes a recommendation ranking optimization method based on user click behavior prediction. In a possible design, the above-mentioned recommendation ranking optimization method based on user click behavior prediction can be integrated into an electronic device. Figure 2 As shown in , the electronic device 500 may include a processor 501 and a memory 502 .

[0112] The memory 502 is used to store a program that supports the processor to execute the data processing method or resource allocation method in any of the above embodiments, and the processor 501 is configured to execute the program stored in the memory 502 .

[0113] The memory 502 is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor 501 to implement the steps in the above embodiment:

[0114] (1) Obtain historical exposure behavior data and the corresponding database, and organize them to obtain a sample data set;

[0115] (2) performing association processing on the sample data set obtained in step (1) to obtain relevant data;

[0116] (3) Clean the relevant data obtained in step (2) and filter out abnormal data;

[0117] (4) performing data processing operations on the cleaned data obtained in step (3);

[0118] (5) performing time data processing and text data processing on the processed data obtained in step (4);

[0119] (6) Training the GBDT model on the processed data obtained in step (5);

[0120] (7) Predict and sort the new data using the GBDT model obtained in step (6).

[0121] Figure 3It is a structural diagram of a computer system suitable for implementing a recommendation ranking optimization method based on user click behavior prediction according to an embodiment of the present application.

[0122] like Figure 3 As shown, the computer system 600 includes a processor (CPU, GPU, FPGA, etc.) 601, which can perform part or all of the processing in the embodiments shown in the above figures according to the program stored in the read-only memory (ROM) 602 or the program loaded from the storage part 608 to the random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the system 600 are also stored. The processor 601, ROM 602 and RAM 603 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0123] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed, so that a computer program read therefrom is installed into the storage section 608 as needed.

[0124] In particular, according to an embodiment of the present application, the method described above with reference to the accompanying drawings can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program tangibly contained on a readable medium thereof, and the computer program includes a program code for executing the method in the accompanying drawings. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 609, and / or installed from a removable medium 611.

[0125] The flowchart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operations of the system, method and computer program product according to various embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0126] The units or modules involved in the embodiments described in this application may be implemented by software or hardware. The units or modules described may also be set in a processor, and the names of these units or modules do not constitute limitations on the units or modules themselves in some cases.

[0127] As another aspect, the present application further provides a computer-readable storage medium, which may be a computer-readable storage medium included in the node described in the above embodiment; or a computer-readable storage medium that exists independently and is not assembled into a device. The computer-readable storage medium stores one or more programs, and the programs are used by one or more processors to execute the method described in the present application.

[0128] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the present application.

[0129] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0130] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0132] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0133] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0134] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0135] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0136] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0137] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A recommendation ranking optimization method based on user click behavior prediction, characterized by: The following steps are involved: (1) Obtain historical exposure behavior data and the corresponding database, and organize them to obtain a sample data set; (2) Performing association processing on the sample data set obtained in step (1) to obtain relevant data; (3) Clean the relevant data obtained in step (2) and filter out abnormal data; (4) performing data processing operations on the cleaned data obtained in step (3); (5) performing time data processing and text data processing on the processed data obtained in step (4); (6) Training the GBDT model on the processed data obtained in step (5); (7) Predict and sort the new data using the GBDT model obtained in step (6); Among them, in step (6), the processed data needs to be labeled before training the GBDT model; The weak form of the boundary element formula for the prediction regression of the GBDT model in step (6) is as follows: Among them, k(T) represents the boundary change rate; T is the number of user clicks; G(x, y) is the weight function, which takes the basic solution of the Green function and feeds it back to the training of the GBDT model, where x and y are the horizontal and vertical coordinate values ​​of the Cartesian space in the GBDT model respectively; q(x) is the weight value of the doc2vec word vector in the GBDT model, and Ω is the calculation domain of the transient nonlinear heat conduction equation; Γ The computational domain Ω The boundary of , n is the number of vector transformations in step (5); ρ represents the fixed coefficient of the Green's function and is equal to 1.2, c represents the fixed coefficient of the Green's function and is equal to 0.5; At the same time, the weight function is optimized by combining the algorithm described in the following formula (II): In the formula, F(z,y) It is expressed as the optimized value of the weight function and fed back to the weight value of the doc2vec word vector; （ z, y) represent the vertical coordinate value and ordinate value of the Cartesian space in the GBDT model respectively; r is the distance between the source point and the field point in the Cartesian space in the GBDT model; G (x, y) is the same as the weight function in Formula I; b (x) represents the lateral search direction function of the Cartesian space in the GBDT model; r α It represents the exponential ratio between the r value and the unknown coefficient of the weight function; α is the unknown coefficient; dr represents the differential of the r value.

2. The recommendation ranking optimization method based on user click behavior prediction according to claim 1 is characterized in that: The historical exposure behavior data described in step (1) is the exposure record of each blog and recommended blog obtained based on the user's exposure click record and the behavior mark of whether a click occurred.

3. The recommendation ranking optimization method based on user click behavior prediction according to claim 1 is characterized in that: The relevant data described in step (2) are multiple dimensional data of each blog and recommended blog; The multi-dimensional data include recommendation strategy, title, description, publishing time, last modification time, all blog likes, all blog collections, all blog comments, all blog shares, blog likes in the last ten days, blog collections in the last ten days, blog comments in the last ten days, blog shares in the last ten days, blog post length, blog author registration, blog author ranking, blog author visits and blog author posting frequency.

4. The recommendation ranking optimization method based on user click behavior prediction according to claim 1 is characterized in that: The abnormal data described in step (3) is blog post data of abnormal detection or abnormal status; The blog post data of abnormal detection or abnormal status includes blog post data generated by non-essential users, wherein non-essential users include robots, zombie fans, advertising users, and users whose fans are lower than a set value.

5. The recommendation ranking optimization method based on user click behavior prediction according to claim 1 is characterized in that: The data processing described in step (4) includes null value filling processing and type data one-hot encoding; The described empty value filling process is to fill in the missing values ​​in the blog post data; The one-hot encoding of the type data is to perform eigenvalue digital processing on the type data in the blog post data.

6. The recommendation ranking optimization method based on user click behavior prediction according to claim 1 is characterized in that: The time data processing in step (5) is as follows: Convert the timestamp of the processed data.

7. The recommendation ranking optimization method based on user click behavior prediction according to claim 1 is characterized in that: The text data processing in step (5) is as follows: The processed data is segmented, stop words are filtered, and doc2vec word vectors are converted.

8. The recommendation ranking optimization method based on user click behavior prediction according to claim 1 is characterized in that: The sorting described in step (7) is to sort the predicted data in reverse order using the probability of being predicted as 1.

Images