Low retention and difficult transformation are the most troublesome problems for current operators. In the process of user operation, many times we only care about the late, but found that the loss outweighs the gain. For example, after the loss of users to recall, aside from the high cost, the final recall rate is not ideal. As a matter of fact, whether it is the conversion of users to pay or the eventual outflow of users, it is directly related to their previous attributes and behaviors. Based on such data, AI algorithm can be used to lock the users who will lose or may pay in the near future in advance and take targeted operational measures to effectively extend the life cycle of users and improve the payment conversion rate.

This is the problem solved by Huawei Forecast Service, which focuses on the two core operating scenarios of user loss and payment conversion, and carries out multi-dimensional crowd oriented prediction. It supports multi-touch operation of predicted crowd through push service, in-app message and other forms of user touch.

Such a powerful and practical tool, since the launch has been the majority of developers have been a hot discussion and continued attention. In this paper, we will combine the specific functions of Huawei Forecasting Service to share tips on how to use the products in the daily operation process. We hope to help you gradually get into the best situation and have a deep understanding of Huawei Forecasting Service:

Question 1: What can the predicted population do specifically?

Predicted audience groups can be used directly for push services, in-app messaging, remote configuration and other growth services provided by AppGallery Connect. You can choose the appropriate way to reach users based on your specific business strategy.

For example, through push service, users with high probability of losing can be pushed with activity news such as “new version of gameplay and new gift package launch”, so as to promote users to stay active and prevent loss. All you need to do is select “Forecast Audience Groups” in the background of the configuration of push service. Similarly, other services, such as remote configuration and in-app messaging, can be used to filter the crowd and reach the target predictive users in this way.

Question 2: How to evaluate the accuracy of the prediction results?

The “true case rate” and “false positive case rate” shown in the forecast details page are actually an overall evaluation of the forecast results. The true sample rate represents the proportion of the number of positive samples correctly predicted by the model to the actual number of positive samples, while the false positive sample rate represents the proportion of the number of negative samples incorrectly predicted by the model to the actual number of negative samples.

For example, in the case of payment prediction, the true rate represents the percentage of the paying population that is correctly predicted to be paying by the model to the total paying population; The false positive rate represents the percentage of unpaid people who are wrongly predicted by the model to be paying. From the literal interpretation, it is not difficult to see that the higher the real example rate is, the lower the false positive example rate is, and the more accurate the prediction result is.

Question 3: Why is the prediction service opened, but the prediction task is always unable to be completed?

This comes back to the principle of predictive services. The premise of the predictive task is that your application reports user attribute and behavior data through Huawei analytics services. Therefore, before using the prediction service, it is necessary to first open the analysis service and integrate the SDK of the analysis service to ensure that the corresponding user behavior data can be reported so that the prediction task can proceed smoothly.

For example, a lot of feedback from developers about payment and repurchase predictions is that they don’t work out, depending on whether or not your app has paid events reported. Only if sufficient paid events are reported to support the training of the prediction model, such as the automatic acquisition event “In-App Purchases” (InapPurchases), can corresponding prediction results be generated.

Question 4: How exactly should custom predictions be used?

Custom prediction is an additional prediction scenario provided by the system in addition to preset churn, payment and repurchase scenarios. You can specify the user behavior you want to understand to carry out the prediction task according to the actual product operation requirements.

For example, in a game-like App where the operator may be concerned with the probability of a player passing a level, a custom prediction task can be created with a custom prediction of “passing the level” as the target prediction event. Please refer to the Custom Forecast Usage Guide for specific operation.

The answers to frequently asked questions about the use of forecasting services are shared here, and we will update more tips and tricks on the use of forecasting services from time to time. You can also click here to view other forecasting services related introduction, welcome to visit and use Huawei forecasting services.

For more details on HMS Core, see:

To participate in the developer discussion, please download the demo and sample code from the CSDN community or Reddit community. Please go to GitHub or Gitee to solve the integration problems. Please go to Stack Overflow

The original link: https://developer.huawei.com/… Author: Pepper