Introduction
Today, Wikipedia is part of everybody’s life. It is the first source that most people will check when they want to get information. This website is maintained by volunteers through open collaboration, and what makes its strength, can also bring problems. Wikipedia aims to have a neutral point of view and stay strictly factual but can we guarantee this when anyone can become a contributor? A study [1] observed that a large number of Wikipedia editors change their behavior and focus on editing controversial topics when promoted to administrators, they might be biased (consciously or not) and influence these articles. This concern about bias in Wikipedia content leads us to consider how such biases might extend to user interactions with the platform, in the Wikispeedia game.
Wikispeedia is a game where players are tasked with navigating from one Wikipedia article to another using only internal Wikipedia links. The objective of this game is entertainment, but also allowed to gather data on how humans perceive the semantic distance between different concepts. The paths that players take from the start article to the end article, along with the number of clicks and the time taken, offer valuable insights into the way people relate and categorize information[3]. We will use this insight to understand the effect of potential political bias. Our study aims to examine the political bias within Wikipedia and its potential impact on Wikispeedia players.
Our dataset is an extensive collection derived from the Wikispeedia game, featuring detailed records on how players moved from one Wikipedia article to another, a process we’ll refer to as ‘the path’ in our subsequent analysis. This dataset encompasses 4,604 Wikipedia articles as they existed in 2007, meticulously documenting user navigation and thereby shedding light on patterns in the pursuit of information. It also captures the duration users spend on these paths, providing a window into their engagement levels and the complexity of the topics encountered. Furthermore, the dataset includes categorizations for each article, facilitating focused studies across a variety of knowledge areas. This rich, layered dataset serves as the foundation for our investigation into potential political biases in Wikipedia’s content.
What about political bias and political figures in Wikispeedia ?
Firstly, let’s define political bias : this refers to a tendency to orient or modify information to make a political position more attractive. To define that, plain text of the articles was assessed in relation to two other well-known sites that reproduce Wikipedia in a politically biased way: Rationalwiki and Infogalactic. According to Media Bias/Fact check, these websites are respectively recognized as having a left-wing and right-wing bias. After training a natural language processing model (BERT embeddings with SVM for classification) on data scraped from these websites, Wikispeedia articles are classified according to the similarity with the collected text. If it’s closer to one compared the other, we assume a political bias in the article. The results are shown below:
We can see that most of articles in Wikispeedia are similar to Infogalactic and therefore classified as right wing.
Let’s characterize this repartition a bit more precisely and dive deeper in the data. A good way to understand this bias is to look the repartition through each categories of articles to see if there is a notable difference.
Even though most categories have a clear majority of articles with a right bias, in a few categories, such as chemistry or railway transport, they account for less than 50%.
It seems that article categories can have a link with the political bias in the way they are written. It can be for many reasons one of which could be that it comes from the articles’ authors. Let’s think about it. Wikipedia is collaborative website where anyone can write or complete articles. Who takes personal time to write an article on Wikipedia ? It has to be people passionated or at least interested in the subject. In that case, a chemist or someone really interested in chemistry is more likely to write an article in the correponding category.
This concept becomes particularly pertinent when we turn to politically charged topics like political parties, figures, or events. Here, the contributors could be those who hold strong opinions or affiliations, potentially coloring the articles with their biases.
To put this theory to the test, we specifically zoom in on political figures. They present a straightforward case for categorization based on political affiliations. Our method involved manually classifying political personalities based on their party or ideological leanings, using web sources for accuracy. We then correlated this classification with the perceived bias in their Wikipedia articles. The results of this analysis, revealing the intersection of political affiliation and content bias, are presented below.
There is a trend of right-leaning bias across articles about political figures from various affiliations. We performed a linear regression analysis to investigate the potential correlation between political affiliations and perceived bias in Wikipedia articles. Although the findings were not conclusively indicative due to the limited size of the dataset, we identified one significant result: the ‘affiliation_Center’ variable showed a positive correlation with a right-leaning bias in Wikipedia articles (coefficient = 0.3135, p-value = 0.002). This suggests that individuals with a centrist or no political stance may be more likely to have Wikipedia articles that exhibit a right bias.
We now have a pretty good overview in the political biases in Wikispeedia articles. We now understand that there are some biases in the way articles are written and they might permeate in the content we consume daily. This raises a crucial question: How do these biases translate into interactive online environments, specifically in games like Wikispeedia? In the next section, we delve into the Wikispeedia Analysis to investigate whether the potential biases in Wikipedia articles affect the way players navigate through this game.
How does the bias affect players during their games ?
In order to measure how bias affect player during the game we first calculated the average bias of each path and analysed the following key metrics :
- The percentage of back-clicks
- the mean rating
- the finished percentage
- the mean time people take.
We correlate these metrics with the political bias of articles on the plot below :
To facilitate comparability, we normalized these metrics to a 0-1 range by dividing each value by its respective maximum, allowing for an equal weighting across the diverse scales of data. With this plot we can visualize the biases of each path relative to one another but cannot visualize absolute performance, since the graph does not display the maximum possible value of a feature.
Well, this graph is interesting. There are significant differences in the way people navigate through Wikipedia depending on the bias of the article. The graph clearly shows that paths with an average left-leaning bias tend to be abandoned more frequently, suggesting that players often give up the game when encountering these articles. This might be attributed to difficulties in finding relevant information, leading to confusion. This theory is supported by the observation that the average time spent on left-biased paths is longer compared to those with right bias or center biases. Additionally, paths featuring a higher proportion of right-leaning articles are associated with more successful game outcomes, indicated by better ratings. However, it’s important to consider that right-bias paths also exhibit a higher frequency of back-clicks, implying that users often do not find the information they initially sought.
This analysis is valuable as it offers preliminary insights into how biases in Wikipedia articles might influence user navigation behavior, revealing preferences in information seeking.
We will now delves deeper into the specific choices players make.
Analysis of the players
In the Wikispeedia game, the players are given a starting article and destination article randomly. But they are also able to choose their own first and last article. It is the first decision players face and it can already show tendencies. Let’s first turn our attention to this aspect of the game.
Here are the articles the most choosen by the players :
Those articles are all very standard but something caught our attention in the graph.. Let’s look take a look at the first and last article chosen by players but only for the political people category. Is there a tendency to choose more right -wing people or left-wing people?