Sentiment and alignment: a character analysis in Harry Potter fan fiction

People use fan fiction to create and share their interpretations of a story and its characters. In this study, the interpretation of characters in fan fiction is analyzed by studying the sentiment associated with characters in Harry Potter fan fiction. In particular, the sentiment of text mentioning the characters’ first and/or last names is being analyzed. Fan fiction represents a safe haven for readers to explore new directions with known characters and modify representations to be more similar to themselves, which could help fan fiction writers accept themselves and educate others by diversifying available representations. It can be seen as a special way of feedback as it shows the readers’ engagement with the original story. Studying fan fiction can help discover the relationship between authors and readers and how this engagement affects published fiction. Examining why readers become emotionally invested in a story and what provokes such imaginative engagement has yet to be explored further.

Here, the sentiment is defined as the sentiment of the opinion holder, in this case, the writer. Writers tend to bring more attention to characters that they like in fan fiction. The author’s sentiment is expressed through how a character is portrayed and developed, which reflects the sentiment a reader experiences.

In addition, the sentiment analysis was performed on the sentences in which the character is the grammatical subject as this assures that the sentence’s sentiment does not affect any other character that is present in the sentence. The original alignment of characters corresponds to the original author’s sentiment towards characters and the predicted alignment corresponds to the fan fiction writer’s sentiment. The neutral class has been added as some characters have no clear alignment or have switched alignment during their development in the story.

The sentiment scoring is calculated using a compound score calculated with the Valence Aware Dictionary for sEntiment Reasoning (VADER) model. As VADER is designed for social media, it is sensitive to polarity and intensity of emotion, and it works best on short documents like sentences. VADER seems applicable as the text used are sentences of fan fiction summaries. The output of VADER makes sense as the polarity of emotion matches the assumed sentiment about characters, i.e., the assumption that heroes tend to have positive sentiment scores and villains tend to have negative sentiment scores. This is used to classify and determine if characters are indeed perceived as intended by the original author. The sentiment scores are averaged over each character and then classified using thresholds.

It has been found that the sentiment for a character’s first and last name are different from each other, as the way a character is called reflects the caller’s affection. Moreover, it has been found that the characters’ average scores were not explicitly positive or negative, and they contained a high variance. This suggests that fan fiction reflects a version of the characters, which does not always align with the original way characters are portrayed.