Marketing Strategy and Customer Engagement Analysis of Catering Brands – From Short Video Perspective

. In recent years, with the popularity of short video platforms, more and more peo-ple tend to use short videos to share their lives. In the catering industry, many phenomenal pop videos created by many catering brands are continuously shared on the short video platform, which has brought about a doubling of brand reve-nue. In order to speciﬁcally explore the relationship between short videos and ca-tering marketing, in this paper, we ﬁrst design a questionnaire and collected hun-dreds of samples. Then, based on the samples, we analyze the users’ preferences for catering brands and related short video attributes. We also analyze the correla-tion between user attributes, sharing behavior, and customer engagement. Finally, we establish a discriminant model, a machine learning method, to predict users’ sharing and visiting behaviors according to their attributes. The experimental re-sults show the effectiveness of our method.


Introduction
In recent years, there have been some phenomenal catering marketing cases in the Chinese Mainland.In July 2020, the Chinese catering giant Haidilao launched a birthday blessing service, and the "social death" video shot by customers was widely spread on the short video platform.Soon after the service was launched, the number of customers of Haidilao increased significantly.In March 2021, the lesser-known bar brand Helens launched the popular product "Coke Bucket", and many custom-ers shared their happy time after consuming "Coke Bucket" on the short video plat-form.For a moment, there was a long line at the entrance of the Hellens tavern, and the Hellens tavern expanded rapidly.On September 10, 2021, Helens International Holding Co., Ltd. was officially listed on the main board of the Hong Kong Stock Exchange.It only took half a year for Helen to go public from the sale of the block-buster.It happens that there is a similar case, in July 2021, with more than 10000 franchise stores in the country, Mixue Ice cream & Tea became popular on the short video platform with a theme song of Snow City, which successfully brainwashed many young people as a magical magic tune.Since then, the popularity and sales volume of the Snow City have reached a higher level.Another example is that KFC China launched "Crazy Thursday" promotion activity in 2018 and quickly spread it through the short video platform in 2019.As a result, Yum!!China (the parent com-pany of KFC China) achieved rapid growth in revenue in 2019, and the growth itself was largely brought by the business of KFC (as illustrated in Figure 1(c,d)).Although Yum China's revenue declined due to the impact of the COVID-19 pandemic in 2020, the proportion of KFC business in Yum China's revenue has increased year by year (as illustrated in Figure 2).
In these examples, it is not difficult to see that the short video platform plays an important role in revenue growth.On the one hand, as shown in Figure 3, according to CNNIC statistics, the number of users and the user utilization rate of short video platforms in Chinese Mainland are increasing year by year.More and more people like to share their lives through short video platforms.On the other hand, these mar-keting or shared videos attract potential customers from multiple senses such as vi-sion and hearing.
Gao et al. [2] proposed a possible explanation from the perspective of psycholo-gy called Short video customer inspi-ration (SVCI).Short video customer inspiration (SVCI) is the embodiment of the concept of customer inspiration in the short video context.It is the activation state that customers get from the idea of receiving marketing guidance to the internal pursuit of consumptionrelated goals in marketing short videos, thus having a positive impact on customer integration.
However, the problem is that there is a lack of quantitative analysis of the impact of short videos on commercial brands and customer engagement.As a result, we cannot formulate effective marketing strategies.In order to solve the above problems, in this paper we first design a questionnaire and have collected hundreds of valid results.Secondly, we analyze the correlation between user (customer) attributes and customer engagement behavior.Finally, we design a linear discriminant analy-sis(LDA) model, and analyze the short video and mar-keting types that users poten-tially like to make more accurate marketing strategies.Customer engagement (CE) has received more and more attention from both busi-nesses and academia recently because it can bring faster revenue growth to enterpris-es [4].The main-stream definition is mainly from the perspective of psychology, be-havior, and the integration of psychology and behavior.In terms of psychology, CE can be seen as a psychological state generated by the interaction between customers and brands [1].In terms of behavior, CE can be seen as a kind of incentive that goes beyond buying [5].According to these concepts, we designed the questions about customer engagement in the questionnaire.

Linear discriminant analysis
Linear Discriminant Analysis (LDA) [3] is a very common technique for feature ex-traction and dimension reduction.It has been used widely in many applications such as pattern classification applications [6,7].The main idea of LDA is to project the data in a high-dimensional space into a lower-dimensional space, and after the projection, ensure that the intra-class variance of each category is small and the mean difference between categories is large, which means that the highdimensional data of the same category are projected into a lower dimensional space, and the same categories are clustered together, while different categories are far apart.In this paper, we use an LDA model to predict customer engagement behavior.This design is based on the user attributes of the short video platform and the cus-tomer engagement and preferences we need to investigate.In this paper, we mainly focus on the relationship between user attributes and customer engagement, as well as the prompts brought by user preferences.
We collected a total of 300 valid samples.Some basic information is given in the Table I.And We will analyze our data in detail in the next two sections.

Reliability and validity analysis
In order to evaluate the effectiveness of the questionnaire, we conduct the reliability and validity analysis.The reliability of internal consistency was evaluated with Cronbach's alpha coefficient.As reported in Table II, the value of the reliability coefficient is 0.833, greater than 0.8, which indicates that the reliability quality of the research data is high.
Then we analysis the validity of the questionnaire.As shown in Table III, KMO and Bartlett tests were used for validity verification.It can be seen from the above table that the KMO value is 0.785, between 0.7 and 0.8.The research data is suitable for extracting information.

Analysis method 4.1 Correlation analysis
We select twelve typical user attributes (marked with z 1 , z 2 , ..., z 12 , as shown in the Table VI) and five customer engagement behaviors (marked with y 1 , y 2 , ..., y 5 , as shown in the Table VI) from the questionnaire to try to explore their relationships.Then, We make a correlation analysis between these user attributes z i and customer engagement y j (Table IV).In order to be more intuitive, we also conduct a visual analysis, as shown in Figure 4.
We can draw some interesting conclusions from the anal-ysis.1) Chain brand catering has advantages in customer engagement.There is a significant positive correlation between the preferences of restaurant chain brands and customer en-gagement behavior.2) Older people may have more difficulty in customer engagement.For example, even if they see the shared catering short videos, they may not visit this restaurant.3) People in cities with higher levels are more likely to have customer engagement behavior.It may be because there are more different restaurants and convenient transportation.4) Those who interact actively on short video platforms (such as those who like to share their favorite videos) are more likely to have customer engagement behaviors.5) There is no correlation between the behavior of ordering takeout and user engagement.6) Those who like delicacy short videos are more likely to have customers' behavior.This shows the importance of precision marketing (recommending relevant short videos).

LDA model
In order to further explore the relationship between these user attributes and customer engagement, we use LDA model to predict customer engagement y1 , y2 .Given data set D � ��� � , y i �� i�� m , y i � ��,�� , let X i ,  i , Σ i represent the i � ��,�� set , mean vector and covariance matrix respectively.If the data is projected onto the line ω, the projections of the centers of the two types of samples on the line are ω T μ � and ω T μ � and the covariance are ω T Σ 0 and ω T Σ 1 .
The optimization goal of the model is, on the one hand, to make the projection points of similar samples as close as possible, so the covariance of the projection points of similar samples should be as small as possible; on the other hand, to make the projection points of different samples as far as possible, so that the distance between the class centers is as large as possible.Considering both, the goal is to maximize:  In this linear discriminant analysis, we use 300 samples as our dataset.We ran-domly selected 80% data as the training set to train the discriminant analysis model.The remaining 20% data is used as a test set to verify the effectiveness of the mod-el.We can evaluate the effectiveness of the model by three indicators: accuracy, recall and F1 score.First, the accuracy rate refers to the proportion of samples that belong to a certain category in the actual situation when the prediction is made.Second, recall rate refers to the proportion of samples correctly predicted for a cate-gory when it is actually a category.Third, the F1 score value refers to the weighted comprehensive index of accuracy and recall.
The experimental results are shown in Table V, which reflect that we can predict user engagement behavior (y1 , y 2 ) through user attributes, and the results are satisfied.According to our survey, more than 95% of the people have watched short videos related to catering.and about 81.2% of these people will go to relevant restaurants because of these short videos.This actually reflects that short videos can bring more customers.Now that we know the importance of short videos, how to push relevant videos to specific people in the marketing strategy is even more important.In the previous section, we specifically analyzed this problem.We can use our LDA model to predict what kind of people are more likely to share and be affected (i.e., custom engage-ment) by short videos.When the target group is determined, we can spend less money to obtain better marketing results.

Discussion 2: What short videos do customers prefer?
When we determine the target audience, what style of short videos to make is also an significant issue.In order to better understand user preferences, we counted the results of User Preferences category in the questionnaire.As shown in Figure 5, the majority of people think that the reason for the success-ful marketing of these catering brands is wildly spread and impressive(brainwashing).It is worth noting that they do not think the user's secondary creation is critical.This actually reflects from the side that we need to focus on the widely spread point.As we mentioned before, it is important to learn how to predict target users.Whether from videos that attract users or those that arouse users' desire to share, their com-mon features are close to life and have unique content, which actually requires us to grasp the current popular elements and be close to life when marketing.

Conclusion
In this paper we first designed a questionnaire and collected hundreds of samples.Then, based on the samples, we an-alyzed the correlation between user attributes and customer engagement behavior.The LAD model is built to provide a model that can predict customer engagement behavior.The experimental results show that the method is effective.Finally, we also analyze users' preferences for short videos and give specific suggestions.

Table 1 .
Sample basic information statistics

Table 2 .
Cronbach reliability analysis 3.1 Questionnaire design and staticsIn order to collect relevant data, we designed a questionnaire.The question-naire includes four categories (User Attributes, Customer Engagement, Other Information and User Preferences) and 55 questions in total.The details of the ques-tionnaire are in the appendix (Table VI, Section VII).

Table 5 .
LDA results 5.1 Discussion 1: Are short videos important to customer engagement?

Table 6 .
Details of the questionnaire