Weibo PUSH strategy and optimization plan!

Strategic products are already a relatively complete discipline. Those who want to learn can search for relevant courses through search engines. This series of articles does not discuss methodology, but only shares cases that I have analyzed myself. The focus is on showing ideas, hoping to provide some reference and inspiration for everyone.

Research time: 2018.12

Research object: Weibo PUSH

Research goal: To discover problems or opportunities through research, which will serve as the basis for optimizing the plan to increase the average number of PUSH users per day.

Research background:

• User groups are divided into three categories: high frequency, medium frequency and low frequency. Roughly divided by platform active behavior (interest tags, update frequency, content interaction frequency, friend interaction frequency, etc.)
• The number of push messages and the opening rate (N messages pushed, how many users opened) are high frequency: 20 messages - 10%, medium frequency 10 messages - 5%, and low frequency ≤ 5 messages - 2%.
• Ignoring technical reasons (assuming PUSH reach rate is 100%)

Project Background

1. Product Goals

Reach users through push, maintain and increase the activity of various user groups.

2. Ideal state

By pushing content that users are interested in at the right time, you can attract clicks from various user groups, cultivate medium and low-frequency users into high-frequency users, and avoid users feeling disturbed and closing push notifications.

3. Core indicators

Average number of push opens per day: Each user opens at least one push message every day.

Number of people who opened the message = Number of people who received the message * Opening rate * Arrival rate

Number of users sent = users who have not turned off push notifications + users who have been recalled after turning them off + new users added every day (enabled by default)

Keep the push users that have not been closed and recall the closed users.

Problem Analysis

High frequency users

(1) User Profile

Personal information: 2D nickname, avatar, and mobile phone number for login.

focus on:

Behavior:

The last active time was 2017-1. The active behaviors during the research period are as follows.

environment:

Shenzhen, iPHONE6SP seasonal system version

(2) Sampling analysis

Sample range: All pushes from the afternoon of November 15th to November 23rd. Due to the length of the article, in order not to interrupt the rhythm of the research and analysis ideas, the sampling details will be omitted.

Sampling details include:

• Detailed PUSH content records (push time, release time, title, path, tag, publisher, content type)
• User Conduct
• Number of push messages
• Content ratio
• Single day analysis

Low → Medium Frequency Users

(1) User Profile

Personal information: Female, friends circle is mainly game friends, and most of them follow science bloggers.

User behavior: Push notifications were enabled on November 15, and users clicked on the push content and browsed it for a short time.

Environment: iOS, Beijing

(2) Sampling analysis

Sample range: All pushes from November 16 to November 20.

Summary of main issues

1. Inaccurate user profile leads to biased push strategy

Identify low frequency as high frequency. A user who was last active last year and has turned off push notifications received 12 push notifications in one afternoon after turning them on.

2. Number of push messages

High frequency - too many push notifications.

Defined as a high-frequency user, he pushed 12 messages within half a day, with more messages pushed on weekdays than on weekends.

Low frequency - too few push messages, medium frequency - moderate.

The special attention of low-frequency users has obviously affected the number of push notifications. Excluding [Special Attention], there were only 2 push notifications per day in the first two days.

3. Push time

(1) Push time period

(2) Push frequency

The push notifications were too frequent, with 3 pieces of content not of interest to the user pushed in 1 minute.

4. Push content type

(1) High frequency

1. Entertainment information push accounts for too high a proportion;
2. The proportion of [Guess You Like] did not increase as users’ interests became clearer, and the performance was unstable;
3. When pushing [Guess You Like] three days ago, we pushed some authors and related content that had no popularity and quality assurance.

(2) Mid and low frequencies

Proportion of pushed content:

【Guess you like】Content ratio:

Among all medium and low-frequency push notifications, content related to user [concerns] accounts for 58%. The [Guess You Like] section only pushes 3 types of content, of which [Games] account for 75%.

The push notifications to medium and low frequency users are too conservative.

5. Pushing duplicate content

(1) High frequency

The next day, the same author and similar content

The earliest push content today and the latest push content yesterday both came from the same publisher, and the content is almost the same.

Similar content from different authors on the next day or the same day

After pushing one piece of information yesterday, which the user did not click on, today two pieces of information with almost the same content posted by different users were pushed.

Same day, similar content from different authors

If the user did not click, the same content would be pushed again one hour later.

The same attention screen

Accounting for 75% of the [following users] on that day.

Accounting for 100% of the [following users] on that day.

Hard advertising (not processed yet)

(2) Mid and low frequencies

Push the same [Followed User] content every other day or on the same day.

It accounts for 33.3% of the total [following users].

It accounts for 22.2% of the total [following users].

6. Project Planning & Prioritization

[Improve user portrait] The accuracy of the user portrait directly affects the judgment of the push system and is the basis of push, so it has a high priority.

[Optimize the number of push notifications] [Duplicate push content] Too many push notifications and duplicate content are the two major factors that directly lead to users losing interest or even closing push notifications or uninstalling them. They have a high priority.

[Guess You Like] is the cold start stage, which is an important stage for guiding low- and medium-frequency users to become high-frequency users and improving user portraits to enter a virtuous cycle. It has a high priority.

[High push frequency] is currently just a judgment based on common sense. In fact, each user's usage time and amount of information received are different. It can be gradually optimized after the user settles in. Priority Medium

[Entertainment] The proportion is too high. According to the 2017 Weibo User Development Report, the high proportion of entertainment is within a relatively reasonable range. We can conduct a sample analysis after more user data is accumulated. Medium priority.

Requirements Document

1. Project Background

(1) Product status

Currently, there are three types of users on the platform: high frequency, medium frequency and low frequency. Low-frequency users have few interest tags and update them slowly, their interaction frequency with friends is low and most of their friends are low-frequency users, resulting in very limited interest and relationship materials that can be pushed to them. High-frequency users are just the opposite, and mid-frequency users are somewhere in between.

The number of people who have turned off push notifications has been increasing recently, 50% of whom are high-frequency users. After users turn off system notifications, they will not be able to receive push notifications.

The number of people who open push notifications for low-frequency users has always been unsatisfactory due to the small number of messages sent and poor content quality.

Currently, high-frequency users receive 20 push notifications per day, medium-frequency users receive 10 push notifications per day, and low-frequency users receive less than 5 push notifications per day. The opening rate of high-frequency users is 10%, medium-frequency users is 5%, and low-frequency users is 2%.

(2) Project scope

This version does not consider the arrival rate issue for the time being, and the push logic of [Special Attention] is immune to the push strategy.

(3) Pushing materials

(4) Push sources and push restrictions

(5) User portrait

(Image source: Tomato and the Sheep, "Push System from 0 to 1 (VII): Establishing Push User Portraits")

2. Project Objectives

By solving the following problems, the average number of people who open Push per day can be increased.

3. Requirements Overview

• Push content that users are interested in to increase push click rate.
• Push at the right time and control the number and frequency. Avoid disturbing users and prevent them from closing push notifications or even uninstalling the app. Slow down the trend of users closing push notifications.
• In addition to the operational plan, recall users who have turned off push notifications from a product perspective.

4. Detailed requirements

After sampling analysis, 11 types of problems were found. This version focuses on 8 types of problems with high priority.

Sample cases and detailed problem analysis:

Sampling details include:

• Detailed PUSH content records (push time, release time, title, path, tag, publisher, content type)
• User Conduct
• Number of push messages
• Content ratio
• Single day analysis

Problem 1: Inaccurate user profile leads to biased push strategy

Problem description : This user was last active last year and had turned off push notifications. After turning them on, 12 messages were pushed out in one afternoon.

From the current solution, we can see that high-frequency users receive 20 messages per day, medium-frequency users receive 10 messages per day, and low-frequency users receive less than 5 messages per day.

In the absence of any anomalies in the push plan, it is clear that Weibo's definition of high-frequency users is too loose and its assumption of user patience is too optimistic. Blindly calling user history without considering the time interval, untimely updating of user portraits, and incorrect classification of user types will lead to excessive push messages, disturbing users.

Solution: Improve user portraits based on user behavior indicators to make user type positioning more accurate. You can consider raising the threshold for becoming a high-frequency user, extending the active time of high-frequency users, and reducing the chance of disturbing users.

• Adjust the current definitions of high, medium, and low frequency users, and raise the threshold for becoming a medium or high frequency user, such as appropriately extending the unit period defined as active;
• For high-frequency users, add [Recent Activity] statistics and take the average value per unit period.

The [Number of Push Messages] in the table refers to the number of push messages corresponding to the [High Frequency User] push plan, that is, it is a reduction based on the original plan and is not a cumulative reduction.

Problem 2: User profiles are inaccurate and users are not interested in push content

Problem description : The mismatch between pushed content and user interests is one of the reasons for the low open rate.

Solution : Establish the user’s interest tags based on user behavior, and match and sort them with content tags.

Statistics are collected on various material combinations, and the opening rates of various user groups when each content type is pushed in different time periods.

Problem 3: High frequency - too many push notifications & low frequency - too few push notifications

Problem description: For high-frequency users, in the 8-day full-day sampling, the proportion of more than 20 posts is 37.5%, which is equivalent to 20 posts accounting for 25%.

The special attention of low-frequency users has obviously affected the number of push notifications. Excluding [Special Attention], there were only 2 push notifications per day in the first two days.

Solution:

(1) Reduce the number of push notifications from high-frequency users

Reduce rules:

Statistics are kept on the opening rates of various push content, and adjustments are made at any time based on the opening rates of each unit cycle.

The number of push messages per day is controlled at around 15. When the opening rate drops below 15, they are supplemented based on collaborative filtering from [Guess You Like], [Social Hotspots], and [Local Information].

If there is similar content or the same author in the push, see the solution to problem 5 below.

(2) Increase the number of push messages for low-frequency users

Add rules:

• The total number of push messages is ≤10.
• According to the solution to problem 4, the diversity of content in [Guess You Like] is increased so that the system can collect user behavior indicators and interest tags as quickly as possible, and turn low-frequency users into medium- and high-frequency users as quickly as possible.

Question 4: High frequencies - [Guess you like] are unstable, and mid- and low-frequency - [Guess you like] are too conservative.

Problem description: The proportion of [Guess You Like] does not increase as user interests become clearer, and the pushed content is very hesitant.

Solution:

1) Increase the coverage of [Guess you like] content types

Low-frequency users: +40%, medium-frequency users +30%, high-frequency users +20%

Weibo interest type:

(2) Segment user groups and assign different weights to each user group

New users: Cold start phase

Returning users

IF User

High frequency users

Problem 5: High, Medium, and Low Frequency - Duplicate Push Content

Problem description:

high frequency

• The next day, the same author and similar content
• Similar content from different authors on the next day or the same day
• On the same day, the same follow-up was all over the screen. The two ceae account for 75% and 100% of the [following users] on that day respectively

Low Frequency

Most of the [Following Users] content pushed is concentrated in two authors, accounting for 33% and 22% respectively

Solution:

• Optimize content identification strategies, merge and eliminate duplicate pushed content, and extract information such as authors, content, and video keywords and categories for filtering.
• The proportion of [followed users] pushed on the same day is ≤25%. The total within 3 days is ≤30%. Total within 5 days ≤ 35%
• The number of [Following Users] pushed within the same day is ≤ 2 for each person. The total number within 3 days is ≤ 3. The total number within 5 days is ≤ 5

Problem 6: The push time is not appropriate

Problem description:

Push will disturb users when they are studying, working or sleeping.

Solution:

(Data source: Public data compilation)

The usage time of Weibo social APP users is relatively evenly distributed after 10 o'clock, with a small usage peak at 22 o'clock.

The push time should be when people are relatively free, that is, it coincides with the time when people use their mobile phones most frequently. The best times to send messages are on the way to work and during breakfast time (around 7-9 o'clock), lunch break (around 12-14 o'clock), on the way home from get off work (around 18-19 o'clock), and before going to bed (around 21-22 o'clock).

Question 7: Recall users who have turned off push notifications

Problem description : The number of people who close push accounts has been increasing recently, 50% of whom are high-frequency users

Solution:

According to the solution to problem 2, when the user's interest weight in the content is ≥5, a pop-up window will pop up to remind the user to turn on push. After clicking OK, a guide to turn on push will be displayed.

When the user performs operations such as following, liking, browsing, etc. that do not require jumping to the page, it pops up after the operation feedback. When a user comments, forwards, or searches, the result page pops up when the operation is completed.

Author: Murasakihara Shinnosuke

Source: Murasakihara Shinnosuke