Fake View Analytics in Online Video Services

Liang Chen,Yipeng Zhou,Dah Ming Chiu

DOI: https://doi.org/10.48550/arXiv.1312.5050

2013-12-18

Abstract:Online video-on-demand(VoD) services invariably maintain a view count for each video they serve, and it has become an important currency for various stakeholders, from viewers, to content owners, advertizers, and the online service providers themselves. There is often significant financial incentive to use a robot (or a botnet) to artificially create fake views. How can we detect the fake views? Can we detect them (and stop them) using online algorithms as they occur? What is the extent of fake views with current VoD service providers? These are the questions we study in the paper. We develop some algorithms and show that they are quite effective for this problem.

Multimedia,Cryptography and Security

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to detect false view counts in online video services. Specifically, the paper focuses on the following issues:

1. **How to detect false view counts?**
The paper proposes a series of algorithms to identify false view counts generated by bots or botnets.

2. **Can false views be detected and blocked in real - time when they occur?**
The paper explores the effectiveness of online algorithms to achieve real - time detection and blocking of false views.

3. **What is the extent of false views in current online video service providers?**
Through cooperation with Tencent Video, the paper analyzes actual data and evaluates the prevalence of false views in existing systems.

### Background and Motivation

Video - on - Demand (VoD) services usually maintain a view count for each video, and this count is very important for different stakeholders:

- **Viewers**: The view count can be used as a recommendation indicator.
- **Content providers**: The view count reflects the popularity of the video and helps to determine the value of the video.
- **Advertisers**: The view count may affect the evaluation of advertising effectiveness.
- **Online service providers**: The view count is regarded as a kind of "currency" and is crucial for business operations.

However, the existence of false view counts poses a threat to all these stakeholders. For example, in 2012, Google announced the removal of 2 billion false view records from some major record company websites. Such behavior not only affects the real value of the content but also increases the operating costs of service providers and undermines user trust.

### Detection Methods

The paper proposes multiple methods to detect false views, mainly including:

1. **Feature extraction based on user behavior**:
- **A single user watches the same video multiple times in a short period**: If a user watches the same video multiple times in a short period, it is likely to be a false view.
- **A large number of requests from the same IP address in a short period**: If the same IP address sends a large number of requests in a short period, it may also be a false view.

2. **Entropy function**:
- **User entropy**: By calculating the entropy value of a user watching different videos, normal users have a higher entropy value, while false - viewing users have a lower entropy value.
- **IP entropy**: Similarly, by calculating the entropy value of an IP address accessing different videos, normal IPs and IPs generating false views can be distinguished.

3. **Video entropy**:
- By calculating the entropy value of a video requested by different IP addresses, videos whose view counts are mainly contributed by a small number of IP addresses can be identified, and these videos are likely to have false views.

### Experimental Results

Through cooperation with Tencent Video, the paper obtains a large amount of actual data and conducts a detailed analysis. The experimental results show that the proposed algorithms are very effective in detecting false views. In particular, by combining user entropy, IP entropy, and video entropy, false - viewing behaviors can be identified more accurately.

### Conclusion

The methods and algorithms proposed in the paper provide an effective tool for online video service providers to detect and prevent false views, thereby protecting the real value of the content and user trust.