Shulou(Shulou.com)06/03 Report--

With the popularity of smart phones and the increasing abundance of APP forms, the number of applications installed on mobile devices has increased sharply. Users will also generate a lot of online and offline behavior data in the process of using these APP every day. These data reflect the interests and needs of users, and if they can be deeply mined and used reasonably, they can guide the operation of users. If we can predict the next behavior of users in advance, and even know the possibility of unloading and loss of users in advance, we can better guide the optimization of products and the fine operation of users.

Big data service provider promotes the "number" of application statistics products, which can make a comprehensive statistical analysis of APP from multi-indicators and multi-dimensions, such as user attributes, usage behavior, industry comparison, and so on. In addition to basic statistics, channel statistics, burying point statistics and other functions, one of the features of the number is that it can predict user behavior based on big data, helping operators predict the possibility of user loss, unloading, and payment, thus contributing to the refined operation and lifecycle management of APP.

In the process of practice, there are two difficulties in predicting user behavior based on big data: first, developers need to use a variety of means to decompose the target problem; second, the data will have different performance on specific problems.

"number" uses data analysis to model, and the general process of predicting user behavior includes the following:

1. Target problem decomposition

(1) identify the problems that need to be predicted

(2) define the span of a period of time in the future.

2. Analyze the sample data

(1) extract the historical payment records of all users, which may only account for a few thousandths of all records, and the amount of data will be very small.

(2) analyze payment records to understand the composition of paying users, such as age, sex, purchasing power and product categories consumed.

(3) extract the historical data of non-paying users. Here, according to the needs of the product, you can add conditions or extract unconditionally, such as extracting active and non-paying users, or extracting directly without conditions.

(4) analyze the composition of non-paying users.

3. The characteristics of building the model.

(1) the original data may be used directly as features.

(2) some data can be used better after transformation, such as age, which can be transformed into juvenile, middle-aged, old-aged and other characteristics.

(3) the generation of cross features, such as "middle-aged" and "female", can be combined into one feature.

4. Calculate the correlation of features.

(1) calculate the feature saturation and filter the saturation

(2) calculate the feature IV, Chi-square and other indexes to filter the feature correlation.

5. Select relevant models for modeling.

(1) choose the appropriate parameters for modeling.

(2) after the model is trained, the accuracy, recall rate and AUC of the model are calculated to evaluate the model.

(3) if you think the performance of the model is acceptable, you can do verification on the verification set, and after the verification is passed, the model is saved and predicted.

6. Forecast

Load the above saved model, and load the prediction data for prediction.

7. Monitoring

Finally, operators also need to monitor the key indicators of each forecast, identify and solve problems in a timely manner, and prevent unexpected situations, resulting in invalid forecasts or deviations in prediction results.

The above is the overall process of "number" to predict user behavior. Generally speaking, the key to analysis and modeling lies in big data's collection and handling of big data's details. In the whole process of user behavior prediction, there are many methods and models for technicians to choose, but for practical users, there is no best choice, only a more suitable choice.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.