Construction of Precision Sales Model for Luxury Market Based on Machine Learning

,


Introduction
In the past two hundred years, human society has experienced rapid development.After three industrial revolutions, it has brought about a huge liberation of productive forces and greatly changed production relations.In this process, it has also witnessed the development history of sales [1].Time has entered the twenty-rst century, with the rapid development of Internet technique, human beings have entered the era of mobile Internet, great changes have taken place in the way of life, and the exchange of message has become increasingly convenient and fast.After the world's message explosion, people enjoy the mobile Internet life but are also lost in the massive message [2].With the rapid development and further popularization of present-day message technique, Hadoop emerges as the times require and plays a driving role in the development of enterprises.According to the di erences between people's thoughts, behaviors and needs, the detailed analysis is carried out and then the rened analysis and research of people's overall message is carried out [3].
Big Data is a new way of combining present-day message technique and Internet technique.Enterprises use Hadoop to provide more diverse and comprehensive numbers and message support for enterprise development, so that every consumer can get more accurate numbers.At the same time, it also helps enterprises to achieve the purpose of precision sales [4].e world is constantly developing and changing.Nowadays, various transportation systems such as electric vehicles, subways, and high-speed rails have established a diversi ed transportation system, which makes people travel very fast and convenient.Intelligent collection and notebook computers quickly occupy people's life and work.It has developed into an important tool for presentday people's social interaction, knowledge, and entertainment.With the rapid development of society, for the development of enterprises, no innovation means lag, and lag will be abandoned by the society [5].With the advent of the era of experience economy,\ people's pursuit of a high-quality lifestyle is increasing.The sales of luxury brands have gradually changed from product-based to consumer demand-based.People's quality of life has been greatly improved than before, consumers have gradually mastered the right and ability to consume freedom, and present-day enterprises can also use Hadoop message to understand the actual needs of consumers or directly contact them.Based on this, enterprises should formulate scientific and perfect sales plans according to the actual needs of consumers [6].Innovation needs to go through a comprehensive social survey, grasp the development trend of the times, understand the products that lead the consumption trend in the industry, conduct product research and development, formulate reasonable and effective sales plans, and then create a wave of product consumption [7].In the era of mobile Internet, consumers gradually show a trend of personalization and selffocus [8].For the sales of enterprises, the target consumers have become diverse, the needs of consumers have become more subdivided, the means of communication to reach consumers have become increasingly abundant, and consumers' attention has become a scarce resource.How to accurately grasp the personality characteristics of consumers and meet the needs of scattered customers to achieve precision sales has become a serious challenge for today's enterprises [9].In the traditional big data analysis process, the decision tree method is the most commonly used big data analysis method, but with the emergence of massive data, the traditional decision tree method can no longer meet the needs of the information age, and the decision tree algorithm must be improved.To optimize certain parameters, in order to meet the requirements of the modern information society for massive data processing.Marketing in the digital age is a system that runs through product development to after-sales.The corporate marketing concept will impress the production method and business model.However, no matter what the method is or how loud the slogan is, in order to capture the hearts of consumers, it is necessary for the company to be in its own right.Only with good professional quality can you build a good brand.To have a place in the market, you need to have a good product.The market competition among brands has also shifted from product competition to service system competition.Therefore, the personalized design of the entire service process based on the needs of consumers has become the key point of brand marketing.This paper adopts a new optimization method, and the effect is remarkable, which is suitable for being widely put into practice.
How to improve the user's experience in the sales process through apparatus knowledge, so as to bring better business value to luxury brands, and realize value co-creation and sharing, is worthy of in-depth research [10].e precision sales model in the era of Big Data takes the actual orientation of consumers as the core principle, so as to promote the desired products more accurately for consumers, which is the best way for enterprises to obtain economic profits.e innovation of this paper is as follows: (1) We propose apparatus knowledge.To achieve efficient Hadoop apparatus knowledge, it is necessary to build an integrated Hadoop apparatus knowledge system that can support both machine learning algorithm design and large-scale numbers processing.(2) We propose a decision tree algorithm.As a predictive model in apparatus knowledge, decision tree is widely used in various domains because its output results are easy to understand and explain, and it has become a hot research topic in academia.
e dissertation is divided into five parts, organized as follows: Section 1 describes the research background and theme, Section 2 describes the related work of this paper, Section 3 discusses the relevant theories and other content of the theme of this paper, Section 4 discusses the means and experimental content and numbers of this paper.Section 5 is the summary of the full text.

Related Work
Hélène suggested that experiential sales refers to a sales means in which an enterprise promotes customers' cognition, preference, and purchase by allowing target customers to perceive the quality or performance of products or services [11].It is marketing through the Internet, catering to the current trend of young people using the Internet.Miranda recommends creating environments and atmospheres that allow consumers to fully engage their senses during the buying process.
e creation of such a situation can make consumers empathize, thereby triggering their consumption ideas and driving product sales [12].Guercini suggested progress to transform the corporate sales environment perspective into a sales space, where three elements-content, context, and infrastructure-can create value for user scenarios in the e-sales space [13].Song suggested that in the Internet age, search engines such as Google/Baidu, based on the context of user behavior, push advertising message to users, which is the prototype of the concept of scene sales [14].Morillo suggested to design and manufacture products more suitable for users, improve the visibility and reputation of enterprises, and cultivate loyal customers [15].Mo suggested to make complicated message organized, programmed, and interesting to facilitate message digestion and absorption; reduce explicit and frequent product recommendations [16].Amirouche suggested to use the web platform to establish a sound and professional numbers message databank based on the collected numbers 2 Mobile Information Systems message, so as to do the postpreparation work for enterprises to carry out precision sales activities [17].Bazi suggested to use Weibo, WeChat, or some popular software that people use every day to push out sales message to quickly let the masses know and understand [18].Pencarelli suggested to expand the scope of sales groups, gain more recognition and support, and eventually become enterprises to provide precise services to consumers [19].Zhang suggested using precision sales to improve sales effectiveness and quality, obtain maximum profits, and promote the development of enterprises [20].Kauppinen suggests putting users in the three representative web scenarios of input, search, and browsing, ensuring their experience and privacy as preconditions, obtaining their behavior of inputting and recycling message on the Internet, and building a "interest guide + massive exposure" + "Entry Marketing" as the core experience sales idea [21].e aforementioned are some studies conducted by predecessors and researchers.We have come to this conclusion: Assuming that the enterprise is not professional enough, if it wants to open up other markets, it will cause differentiation of its own marketing power, and in the long run, it will be dominated by highly professional enterprises.It is necessary to use the massive data in the Internet platform to conduct accurate market segmentation, accurately locate target customers, conduct accurate customer portraits, accurately understand customer needs, accurately place advertisements, and provide customers with accurate communication and services.In the process of continuous social and economic development in the twenty-first century, people have entered the digital age.In the digital age, all marketers need to rapidly improve their digital cognitive ability, that is, they should actively improve their digital awareness and change the traditional digital thinking.

Construction of Precision Sales Model for Luxury Market
3.1.User Portrait.At the heart of machine learning is learning.Learning is a unique ability of human beings.How to make machines like humans can improve their performance through the influence of the external environment is the focus of research in the field of machine learning.e process of machine learning is a process from unknown to known.e purpose of constructing user portraits is to restore user message.
erefore, the construction of user portraits is based on consumer numbers, and standardized user label message is extracted through web Hadoop and computer technique.
is process covers basic numbers collection and behavior modeling, building accurate user portraits.ere are five stages of sales and feedback evaluation.User portraits need to collect a large amount of behavior numbers, preference numbers, demand numbers, and basic message numbers to improve the fullness of user portraits.Marketing decisions rely on credible market numbers, traditionally obtained through in-depth interviews and market research.e more complete the numbers dimension of the user portrait and the richer the character characteristics, the greater the role played in precision sales.Perform consumer segmentation on basic numbers such as consumers' online behavior numbers, in-service behavior numbers, user content preference numbers, and transaction numbers.Both indepth interviews and market research use small samples and are affected by the shortcomings of the research technique itself, so its reliability is very low.e numbers collected by user portraits are mainly divided into two types.e first type is that the collected numbers is fixed, such as gender, age, height, weight, occupation, geographic location and other message in the user's basic message numbers.In most cases, the numbers label will not change or will not change in a short time; the second type, the collected numbers changes and the change cycle is relatively short, such as user behavior numbers, preference numbers, search records in demand numbers, Commodity browsing records, hobbies type, commodity demand type and other message.By marking users' basic properties, purchasing ability, behavioral characteristics, hobbies, psychological characteristics, social webs, and so on, the user portrait prediction label is constructed, and precise sales strategies are formulated.
After a certain requirement of a user is satisfied, these users will not have the same requirement again in a short period of time.erefore, building user portraits requires continuous collection of user portrait numbers, and the best way to meet the needs of user portrait numbers collection is to use Hadoop technique.With the popularization of e-commerce and the development of web technique, people have the ability to obtain the Hadoop message of the whole sample.Process the numbers collected in the previous stage to model the behavior to abstract the factual labels of the user.
ese user numbers changes at a rapid rate, with many users changing their needs from week to week.Based on web Hadoop and computer algorithms and models, model labels are constructed in combination with corporate strategic goals, numbers conditions, and application scenarios.User portraits need to be constructed using multidimensional attribute labels, and user message is abstracted through numbers extraction.
ese numbers are mainly divided into two categories: static message numbers and dynamic message numbers.Static message numbers is generally user message that is fixed for a period of time and is also user basic attribute numbers, such as gender, age, and other properties.Dynamic message numbers is message that changes at any time, including users' access behaviors, transaction behaviors, consumption preferences, and so on, to a certain extent, reflects consumers' purchasing ability, transaction frequency, and other properties.User attribute numbers is static message numbers, mainly including the user's natural properties, social properties, lifestyle, and psychological properties.Within a certain time range, it is almost unchanged, such as gender, income, education, involving population or other properties, which can generally be obtained directly through the user's registration.
e real-time update feature of web numbers makes it more timely and accurate to reflect the real needs of consumers.Mining Hadoop and extracting "tags" of consumer characteristics is conducive to accurate sales for enterprises.

Mobile Information Systems
User portrait construction must label numbers, and the core of the user portrait model is numbers labeling.Data labeling divides the user's characteristics into different types of labels, and then labels the same type of people with the same characteristics according to the different label types.Finally, according to the free composition of label types, a complete user portrait can be formed.

Data Warehouse.
In recent years, with the continuous deepening of enterprise computer application, most enterprises have invested a lot of time and resources to establish a huge and complex message system, and accumulated a lot of valuable numbers resources.Data warehouse (dw) and online analytical processing are important components of decision support system.Different from traditional online transaction processing, it summarizes, analyzes, and processes existing numbers to provide support for decisionmaking.In recent years, dw has become a hot topic in the databank industry, and it is generally believed that it will be an important part of the future development of databank technique.Dw is the product of fierce contention in the market, and its goal is to achieve effective decision support.Faced with increasingly fierce market contention and potential financial risks, these companies are eager to have a powerful analytical tool to help them fully mine meaningful message from these massive numbers to assist senior leaders in planning and guide decision-making activities.A dw is a subject-oriented, integrated, stable, and time-varying collection of numbers.ese characteristics of the dw determine that it is fundamentally different from the traditional transaction-oriented databank.As a new research domain, dw has developed rapidly, and many universities and companies have carried out extensive and in-depth research in this domain.e overall architecture of the dw system is shown in Figure 1.
e aforementioned is the overall architecture of the data warehouse.We can understand its composition from the figure, which is very helpful for our in-depth understanding of this article.With the development of enterprises, business numbers has expanded rapidly.In addition to using these numbers for transaction processing, people hope that computers can participate more in numbers analysis and decision-making, and databank technique has been trying to make itself competent in transaction processing,various types of message processing tasks from batch processing to analytical processing.e purpose of the dw is to establish a systematic numbers storage environment, separate a large amount of numbers required for analysis and decisionmaking from the traditional operating environment, and transform the scattered and inconsistent operating numbers into integrated and unified message, thereby supporting decision-making.e dw needs to process and organize the numbers extracted from each numbers source to make it suitable for the needs of the online analysis system.A large amount of numbers stored in a dw usually originates from one or several independent numbers sources.It is worth noting that the dw is not only a large databank, although the dw generally has a large amount of numbers, it is also a collection of a series of technologies and solutions for OLAP applications.ere is a growing realization that with current computer processing power and traditional databank architectures, this functionality is simply not possible, and, on the other hand, transaction processing and analytical processing have very different properties and are directly supported by a transaction processing environment.Decisions don't work.e technical structure of the dw is shown in Figure 2.
e complete dw includes three technical contents: dw technique, online analytical processing technique and numbers mining technique.rough the message obtained by numbers mining, the potential customers in the telecommunications business can be identified, the quality of customer service can be improved, and new project areas can be discovered.In a dw, association patterns, sequence patterns and clustering patterns can be found.e problem to be researched and solved by the dw is the problem of acquiring knowledge from the databank.Unlike relational databanks, dws do not have a strict mathematical basis.A dw is a solution to a problem, not an off-the-shelf product To improve the efficiency and effectiveness of analysis and decision-making, analytical processing and its numbers must be separated from transactional processing and its numbers.3.

Research on the Construction of Precision Sales Model for Luxury Market Based on Apparatus Knowledge
As shown in the figure, when time <4, the polyline flattens out as time grows.Traditional numbers analysis and statistical techniques mainly focus on comprehensive analysis of traditional numbers structures with appropriate numbers statistical analysis means set in advance, in order to fully discover the important functions and application values of traditional numbers; compared with other traditional numbers analysis techniques, one of the important cores of Hadoop analysis technique is that it needs to fully excavate the logic behind the hidden numbers from the traditional numbers with a huge number of systems and various structures, so that the traditional numbers can maximize the application value of its functions.e relationship of the contour system over time is shown in Figure 4.
As can be seen from Figure 4, the contour system increases with time.Big numbers technique is another major technological change after the Internet of ings, cloud computing, and mobile cloud computing.e use of

Mobile Information Systems
Hadoop technique has had a huge impact on all walks of life.Big numbers technique has developed rapidly in recent years and has attracted great attention from all walks of life around the world.Under the influence of the era of Hadoop, the development of apparatus knowledge mainly includes two research directions.One is knowledge mechanism, which focuses on simulating human knowledge mechanism.e second is effective message utilization, which focuses on the in-depth mining of potential knowledge from large databanks.e efficiency comparison result curve is shown in Figure 5.
According to the test results, the arithmetic mean calculation means is used to calculate the operational efficiency of the numbers flow equilibrium height, and it is found that the means based on apparatus knowledge proposed in this paper is more efficient than the traditional means.Under the same numbers stream label, the algorithm proposed in this paper is 20% more effective than the traditional algorithm.After the theoretical discussion on the semantic web and art technique in the era of Hadoop, it is believed that for meaningful numbers support and integration in both and complex Hadoop webs, multidisciplinary technique can be crossed.Analyzing and mining these hidden message rules from a variety of message numbers is almost powerless for artificial intelligence operations.erefore, it must be closely integrated with the use of apparatus intelligence knowledge, and computer knowledge replaces manual work to analyze and mine hidden message and obtain relevant knowledge.Apparatus knowledge and numbers analysis are key technologies for transforming Hadoop into useful knowledge, and studies have shown that in many cases, the larger the scale of numbers processed, the better the performance of apparatus knowledge models.From Figures 6 and 7, it can be seen that the fault tolerance index and the accuracy rate.
It can be seen from Figures 6 and 7 that the accuracy of the algorithm in this paper is higher than that of deep knowledge.Big numbers technique has a wide range of applications.With the development of computer and web technique, numbers of hundreds of terabytes or even several petabytes are produced in all walks of life.In the development of present-day Hadoop environment, analyzing relevant numbers has become an important direction for the development of various industries.In this process, apparatus knowledge can quickly absorb knowledge and promote apparatus knowledge to a higher degree of development.Many existing apparatus knowledge means are based on  6 Mobile Information Systems memory theory.When Hadoop cannot be loaded into computer memory, it cannot be processed by many algorithms.erefore, new apparatus knowledge algorithms should be proposed to meet the needs of Hadoop processing.
e relationship between the transmission power and the number of users is shown in Figure 8.
As can be seen from Figure 8, when a web has 100 users, the algorithm proposed in this paper can increase by 20% to meet the requirements of users.Robotic knowledge has become a technique that can support message technique and drive servers.How to perform in-depth numbers' analysis on the complex and diverse Hadoop environment based on industrial robotics knowledge and efficiently integrate and utilize industrial message technique has become the current industrial is the main development direction of robotics knowledge technique research.Big numbers apparatus knowledge is not only an apparatus knowledge and algorithm design problem but also a large-scale system problem.It is neither pure apparatus knowledge nor a problem that can be solved by pure Hadoop processing technique, but a cross-cutting research topic involving two main aspects of apparatus knowledge and Hadoop processing at the same time.Big numbers apparatus knowledge systems should not only focus on apparatus knowledge means and algorithms themselves, but also how to use distributed and parallelized Hadoop processing technologies, so that Hadoop apparatus knowledge systems can effectively analyze complex numbers and process large amounts of numbers in parallel, so as to complete the operation of the relevant apparatus knowledge algorithm in a limited time frame.Apparatus knowledge algorithms in the Hadoop environment can ignore the importance of knowledge results according to certain performance standards.e implementation of the divide-andconquer strategy using distributed and parallel computing can avoid the interference caused by noise numbers and redundancy, reduce storage costs, and improve the operating efficiency of the knowledge algorithm.

Decision Tree Algorithm.
With the advent of the message numbers era, the storage and calculation of massive numbers has been realized.e amount of numbers that people count and analyze is increasing, so this poses corresponding challenges to numbers storage devices and storage means.e speed of numbers processing has become the key to Hadoop technique.In the traditional Hadoop numbers analysis process, the decision tree means is the most commonly used Hadoop analysis means, but with the emergence of massive numbers, the traditional decision tree means can no longer meet the needs of the message age, and the decision tree algorithm must be improved.To optimize certain parameters, in order to meet the requirements of the present-day message society for massive numbers processing.
A decision tree is a tree-like predictive model in apparatus knowledge, usually its internal nodes represent a test on an attribute, while leaf nodes represent the final category.
e optimization of the feature algorithm refers to reclassifying the original feature set to form a new subset, which can be better processed and analyzed by computer algorithms. is means is simple and easy to extend, and can be well applied in practice.
Assuming that the set of common commodities rated by user i and user j is i ij , the Pearson's correlation coefficient is defined as follows: Among them, R i,c represents the rating of user i for product c, and R i and R j represent the average rating of user i and user j, respectively.
Let the n-dimensional rating vectors of user i and user j be i → and j → , respectively, and their similarity is defined as follows: Among them, R i,c represents the rating of the product c by the user i.
e set of common commodities rated by user i and user j is I ij , I i and I j represent the commodity sets rated by user i and user j, respectively, then the similarity between user i and user j is as follows: rough the aforementioned equation, we can get products that are similar to the user's expected value, it explains to us how this process comes, let us know it, and know why.
e generation of recommender systems greatly alleviates the problem of "message overload."Feature selection algorithms can be divided into two categories: filters and wrappers.e filter is measured by the message inside the feature set, which is a preprocessing process independent of the subsequent classification algorithm and is evaluated by indicators such as correlation coefficient and sample distance.Owing to its massive numbers characteristics, traditional storage devices can no longer meet the needs of present-day message.Owing to the huge changes in the way of message numbers processing, traditional numbers processing technologies can no longer be applied to Hadoop analysis and processing.
Let T be the numbers set, the category set is e probability of occurrence of attribute V � v i is as follows: In the example of attribute V � v i , the conditional probability of having class C j is as follows: Category message entropy calculation: Category conditional entropy: Information gain: Information entropy of attribute V: e decision tree model is used to solve many basic problems, such as multistage decision-making, table lookup, and optimization.It naturally restores the decision-making process and splits the complex decision-making process into a series of simple choices.It can intuitively explain the entire process of decision-making.e recommendation system of e-commerce builds a bridge between e-commerce and users.
It obtains the user's consumption behavior numbers and various characteristics of users in a specific way and analyzes it to predict the interests and needs of consumers and respond accordingly.e products are recommended to users to achieve a win-win situation for e-commerce and consumers.Statistics of the behaviors of four users within 31 days are shown in Table 1.
A simple analysis of Table 1 shows that users A and C only click on various products without other behaviors.It can be concluded that such users are "crawlers," that is, these users only use negative "crawlers" to obtain product message.e numbers generated by "crawlers" users often affects the results of numbers analysis.erefore, numbers with more clicks and little or no other behavior can be considered to be caused by "crawlers." Owing to the huge numbers set, it is impossible to process all the numbers at once when the memory is performing calculations, and many numbers need to be temporarily stored in the disk.e traditional decision tree algorithm is suitable for the calculation of eigenvalues with large numbers sets, so the decision tree algorithm can be used as an important weapon for Hadoop analysis.In the domain of apparatus knowledge and numbers mining, there are many scenarios such as expert decision-making systems that need to show or clarify the processing process.Most complex apparatus knowledge algorithms cannot meet such needs due to the obscure calculation process, and the decision tree model is exactly the most suitable solution.
User behavior numbers characteristics, such as what items the user has clicked, what items have been collected or purchased, what scores have been given to what items, and other characteristics related to user behavior.Construct the features that reflect the user's consumption habits, and try to dig out the user's shopping habits.e main selected features are shown in Table 2.
e establishment of a recommendation system is inseparable from the support of recommendation algorithms.
ere are many traditional recommendation algorithms, but they all have certain deficiencies.With the development of message technique, a large amount of message is digitized into numbers and processed and analyzed by computers.Since the decision tree algorithm needs to read and write numbers, the huge size of the numbers makes the read and write speed slow.Optimizing the process of the decision tree construction algorithm and reducing the read and write operations of the numbers has become an optimization direction of the decision tree algorithm.Owing to the limitations of its own architecture, the traditional decision tree algorithm still has corresponding defects in processing massive numbers features and cannot meet the needs of fully analyzing numbers features.erefore, the traditional decision tree algorithm must optimize its parameters, and then obtain the corresponding eigenvalues, and finally obtain the overall characteristics of the Hadoop.More and more industries need to analyze and process a large amount of message, find useful message in Hadoop, mine the rules contained in it and make use of it.Proper scaling of mature algorithms that have been successfully applied to relatively small-scale numbers, or the development of new algorithms suitable for Hadoop analysis, has become a current research focus.Decision tree algorithm is a discrete function approximation means, which is a typical technical means of numbers classification and processing.e decision tree algorithm first classifies the numbers, and then uses the induction principle to generate readable decision tree rules.Analysis and judgment of decision numbers: In essence, the decision tree algorithm is a technical means to solve the characteristics of numbers through classification.

Conclusion
In today's increasingly diversified and personalized consumer demand, corporate sales work wants to occupy a place in the fiercely competitive market environment, with the help of Hadoop and present-day Internet technique, to achieve fine market segmentation and accurate market positioning, targeted sales advertising, providing products and services that meet consumer needs, thereby reducing the company's sales and publicity investment, motivating consumers to buy, improving customer satisfaction, and finally achieving a win-win effect of customer satisfaction and ideal economic benefits for the company.User portraits are very important for precision sales.Precision sales must accurately grasp the characteristics of users and then recommend products or push advertisements for them according to the characteristics of users.Finally, the product characteristics and user needs can be reached in order to complete precision sales based on user portraits.According to the test results, the arithmetic mean calculation means is used to calculate the operational efficiency of the numbers flow balance height.
e algorithm proposed in this paper is 20% more effective than the traditional algorithm.To achieve a better result, it means that the company's products need to bring good economic benefits.e realization of this goal is not only as simple as sales but also runs through the entire process of production, sales, and after-sales.At the same time, we also have some problems that need to be solved: In practice, smalland medium-sized enterprises also need to build a precision marketing model based on their own actual conditions to ensure its feasibility.I also hope that this research can provide certain reference for enterprises to achieve precision marketing.

Figure 8 :Figure 5 :
Figure 8: Relationship between transmission power and number of users.
It is a databank technique as a basic means of storing numbers and managing resources, using statistical analysis technique as an effective means for analyzing numbers and extracting message, and using artificial intelligence technique as a scientific way to mine knowledge and discover laws.erefore, it is a technique in which many disciplines are combined and applied comprehensively.Dw technique involves numbers extraction, numbers storage, numbers processing, numbers processing performance, and numbers performance.e problem to be solved is the problem of obtaining message from the databank.
C k  , and select an attribute V to divide T into multiple subsets.V is n nonoverlapping values v 1 , v 2 , . . ., v n  , then T is divided into n subsets T 1 , T 2 ,. .., T n , of which all instances in T i have the value v i .Let |T| be the number of instances of numbers set T, |T i | be the number of instances of V � v i , |C j | � freq(C j , T) be the number of instances of C j , and |C jv | be the number of Mobile Information Systems full instances with class C j out of V � v i instances.en there are: e probability of occurrence of category C j is as follows:

Table 1 :
User behavior statistics.