Entity Fusion on Cross-lingual Game Knowledge Graphs

Explore a continual entity alignment method for growing KGs (2022)

Cooperated with Interactive Entertainment Group (IEG), Tencent in 2022.

My jobs in this project include two parts:

  1. Constructing growing KGs based on Tecent’s game datasets.
  2. Designing a continial entity alignment method for growing KGs.

Constructing Growing Game KGs

Tecent provided me with a data dump they crawled from online gaming platforms (Google Play and App Store). Each item in the data dump is a game entity and its attributive information. What I need to process is cleaning these information and dividing game entities by their release time.

Pre-processing of Tecent's data.

To construct two KGs, I need to define the relations between game entities. The problem about relations is that the relation types between game entities are so few. I can only excavate two types of relation: ‘similar_game’ and ‘develop_by’ using Tecent’s data. Anyway, the statistics of constructed KGs are shown below.

Left: Statistics of App Store game KG. Right: Statistics of Google Play game KG.

Then, I divided game entities into sequential sets by their release time. Following pictures present the amount of games release over time (scale in weeks).

Game release trendency in App Store.
Game release trendency in Google Play.

Continual Entity Alignment

I have to design a continual entity alignment method to deal with alignment on two growing KGs. This task is very novel as a research setting. So I work this task as an ISWC2022 submission.