Articles

Foreword:

The data comes from the flow, and the direction of the flow is the current direction of the company's marketing advertising. In the era of big data, the real value of data technology should be to enhance the marketing value of traffic. I am not a marketing background, but I have also completed the digital marketing specialization of the University of Illinois at Urbana-Champaign , so this article will explore the application of data to the value of traffic marketing from the perspective of data technology.

When it comes to marketing, Marketing fact, I can not do without is often mentioned in the previous article in the 4P principle, since the main marketing comes from this 4P links with consumers, we have to do is to strengthen those ties. Today's article mainly discusses the application of strengthening the connection between product placement and users, which is also one of the most important parts of digital marketing.

There are two problems that need to be solved to explore the relationship between users. The first question is how can I find the right users for my distribution? The second question is how to measure the delivery effect?

Data-driven advertising and marketing process

The user contacts generate records, and the records become data. When the records become a streamline, then the data becomes a user behavior stream. After all the user behavior streams of the user are together, we can define you through data analysis The large user groups and user patterns. Here is an example. A user visits your website on the APP . From the beginning of the visit, the user touch point appears, his mobile phone information, his network information, the content he viewed on the website, and the website he saw Which advertisements, which advertisements have been clicked on, and what kind of interaction behaviors have occurred with the ad slots? A series of behaviors of these users are the behavior streams constituted by the touch points of the users in the process of using the APP . These behavior streams are our data source.

After we collect all the user touch data, how can we get a richer and more comprehensive understanding of users, so that we can more accurately place advertisements, or make the interaction between advertisements and users more accurate? So how to segment users and match user segmentation tags with marketing scenarios has become a core issue. Because we need to expose the right ads to the right users in the right scenes to achieve the best interests of advertisers.

Which label users your advertisement is suitable for is a challenge in the whole process. This requires the accumulation of data, and then through machine learning algorithms to accurately match the product advertisement with the target crowd label. Marketing data sources are generally divided into demographic attributes such as user gender and age, behavior data, and secondary consumption data.

Balance of precision and coverage

Target positioning for precision marketing through user tag matching is actually not enough to maximize the marketing effect. This requires us to trade off between "precision placement" and "overlay placement" . Excessive pursuit of precise placement will result in smaller marketing coverage and insufficient coverage of potential users. Advertisers will seek to reach as many users as possible to ensure the improvement of brand recognition, but this will lead to later conversions. The efficiency is reduced, causing unnecessary costs and waste of resources. How to find the balance between coverage and precision is the biggest challenge facing brand marketing.

Advertising effect measurement

The four steps of selection - placement - measurement - optimization are the four steps in digital advertising marketing. The first selection and placement is the establishment of the user tags we mentioned before, and the placement of advertisements for the right users under the appropriate marketing scenarios. But after the launch, the entire marketing campaign did not end. We need to further measure the effect of advertising, and use the final statistical results to further targeted optimization of the placement and selection strategies to achieve better marketing effects.

Here we need to talk about two types of advertising, the first is performance advertising, the other is brand advertising.

Here is an example to let everyone distinguish between the two types of advertising.

The above is brand advertising, its main purpose is to highlight brand value. The official definition is: the direct purpose of establishing a product brand image, increasing the brand's market share, and highlighting the position of the brand in the minds of consumers. This is a long-term "brainwashing" way to influence consumers' minds to achieve a long-term effect. The purpose is to remember the brand.

The above picture is a typical performance advertisement, its purpose is very clear, it is to urge you to buy, buy and buy. The official definition is that in a performance-based advertising system, advertisers only need to pay for measurable results. That is to say, this kind of advertisement naturally comes with measurable attributes, and some indicators can be used to measure the effect of the advertisement. For example: purchase volume, CTR ( click through rate ), paid conversion rate, etc. Through the measurement of these established indicators, advertisers can charge for Party B’s delivery platform, and Party B’s platform can also use indicators to adjust and optimize their own delivery strategies.

Through the above example, we can clearly realize that the measurement of the effectiveness of performance advertising is very clear, and the quantitative operation after the user is exposed to the advertisement is measured by indicators. However, the measurement of the effect of brand advertising is very metaphysical, because the purpose of its advertising is not to produce any quick effects (such as purchase) among users in the short term. The purpose of brand advertising is to make the brand rooted in consumers in the long term. In the minds of consumers, consumers can "remember" the brand. So the question is, how to measure the brand's status in consumers' hearts? The following six methods can indirectly understand the effect of brand advertising.

• Know the popularity of the company's products through online surveys

• Monitor the traffic of new and old blog posts

• Social listen and observe online conversation content

• Measure the interactive performance of social media

• Research search data

• Monitor website traffic data and performance

Among the above-mentioned methods, the first method online survey is also the method currently used by most advertising platforms, such as the Sohu Big Data Center solution:

The user's preference for advertisements is determined through the user's self-selection of advertisements, so that double precision placement between advertisements and crowds can be achieved. The alternative advertisements are accurately placed on the target group, and the target group chooses the advertisement that they want to know the most, and the accuracy of the advertisement placement is enhanced through the user's choice.

After the advertisement is broadcast, Sohu will track changes in users' subjective attitudes through online questionnaires . In order to measure the effect, based on big data, these questionnaires are accurately presented to users who have “watched ads” and “have not seen ads”. By comparing the answers of the people exposed to the ads and those who are not exposed to the ads, it helps brand owners understand that the user group is watching. Changes in subjective attitudes after advertising.

Digital marketing technology

Classification algorithm

Use known classification criteria to infer unknown categories of data. Putting it in the marketing application is to use the classification model trained by known user tags to establish user tags for the remaining large number of users, so as to describe the portrait of the entire user group.

Clustering Algorithm

Classify the population without knowing the classification criteria. For example, I have a 1 00 million users in the absence of classification criteria, I subdivide them how to manage it? This requires a clustering algorithm to give all users the characteristics for automatic classification. Each user classified into a category will use similar user behaviors, so that 10,000 square meters can be based on the clustering results and different clustering categories. User behavior to perform accurate user management for users.

Algorithm combination

We need to know that classification and clustering algorithms are the ultimate goal of the algorithm. There are many algorithms that really achieve the purpose of classification and clustering. For example, classification algorithms include xgboost , decision tree , random forest , SVM , LDA, etc. . In different segments of the marketing scene, the effectiveness of each algorithm is not the same, so the vote Algorithms Voting way to judge the results, the model is a way of strengthening. For example . 3 algorithms is determined that the user A class 2 algorithms is judged B class, then the majority, the final prediction result is A . This is the application of Voting Model .

AB Test

The application of AB test is very extensive. His principle is based on statistical hypothesis testing, which is intended to test whether there is a significant gap between the experimental group and the control group. In the example that Sohu uses questionnaire surveys to measure the effectiveness of brand advertising just described, it is the application of AB test . The same questionnaire is distributed to users who have seen the advertisement and the users who have not seen the advertisement , and measure whether the final brand advertising effect is significant through data collection and statistics.