Precision Marketing, evaluation methods and Parameter Adjustment

Precision Marketing, evaluation methods and Parameter Adjustment

Author: Hu Xiaoyan, an analyst at mingbo zhichuang (Beijing) Software Technology Company)

During this period of work, I had a little understanding of things I didn't know before. It mainly involves three aspects.

1. Precision Marketing

I used to hear or talk about precision marketing, it is also known that "Data Mining" technology is used to segment customers, conduct Association Analysis on products or services, find target customers, and predict and analyze inventory or sales volumes. But I have never understood the concept. Recently, I have to write a program that says "Precision Marketing". However, since the entire solution is to write data mining and Its Application in CRM, so how can we transition from "Precision Marketing" to "Data Mining?

Precision Marketing: based on accurate positioning, it establishes a personalized customer communication service system based on modern information technology to achieve measurable and low-cost expansion of enterprises. It is customer-oriented and analyzes the consumption behaviors and preferences of each customer, in this way, customers can be precisely located, different businesses of different customers can be precisely recommended, and target customers can be promoted and sold.

After understanding the concept, we found that "Precision Marketing" is a new commercial marketing method, and "Data Mining" is an essential tool to achieve this marketing method, it can also be said that the application of "data mining" in "commercial marketing" has promoted the "precise positioning" of marketing ", so as to effectively retain customers, attract customers and fully explore the profit potential of customers, to achieve a win-win situation for enterprises and customers.

2. Evaluation Method

In the past, there were not many evaluation models in SDABAS mining tools. Recently, I learned that the evaluation model is very useful and can well evaluate the selected attributes or algorithms. SDABAS includes performance measurement methods for various algorithms, such as performance measurement of prediction algorithms, which can verify the accuracy of prediction. Due to the many prediction methods, you can use several prediction algorithms to check that the prediction algorithm has a high precision and select the optimal model. In the case of a small amount of data, you can use a variety of verification technologies such as: discount verification, self-lifting verification, X-Verification) to verify the model's ability to predict or classify the test data, if the prediction or classification capability is poor, you can adjust the selected attributes or modify the created model accordingly, which can improve the prediction or classification accuracy of the created model for new data.

3. Model Parameter Adjustment

In the past, when using SDABAS mining tools, except for clustering algorithms, I seldom adjusted the parameter settings in other algorithms.

Recently, due to the need to do prediction and analysis work, the neural network was used, but the prediction result seemed to be not very good. Later, I tried to adjust the parameters and found that the prediction accuracy of the results was improved, it mainly refers to the value of the parameter "Training Cycle", "application ratio", and "Momentum", where the value of momentum may exceed 0.9. When classification is applied, the classification effect of ID3 is better than that of decision trees. However, ID3 is only limited to the "nominal" attribute and cannot be used for the "numerical" attribute. The classification algorithm can adjust the parameter "select standard" accordingly. Other parameters have little impact on the classification effect. Since the decision tree can be applied to the "value" attribute, its algorithm performance varies for different data sets. Several groups of data have been tried, it is found that the use of "gini_index" in "selection criteria" is better, and when the data is not ideal, the use of "gain_ratio" may not be able to establish a category.

This article is from the "explore value in data" blog. For more information, contact the author!