Answers are grounded in this paper and cite their sources.
Sentiment lexicons such as VADER were built for general and social media English, and they misread the vocabulary of video game reviews. They miss slang and jargon such as cashgrab, unplayable or goated, and they score descriptive words such as kill, scary or brutal as strongly negative. This study presents GamerLex, a sentiment lexicon of gaming slang and game jargon for player reviews, built from 4.69 million Steam reviews. Candidate terms came from embedding-based expansion of slang seeds, two rounds of manual review and an audit against a glossary of game terminology. Each term receives a valence on the VADER scale from its association with players’ recommendations. This association is estimated on a training split as a Mantel-Haenszel log odds ratio stratified by review length and calibrated on 1,449 VADER words used as anchors. The release contains 4,223 hand-audited entries in three tiers and an optional automatic tier of 3,823 frequent review words. On 99,389 held-out reviews, adding GamerLex to VADER raised the length-stratified AUC for predicting recommendations from 0.825 to 0.893. On 1,462 review sentences rated by three raters, the Spearman correlation with the median rating rose from 0.579 to 0.664, above two uncurated lexicons induced from the same data (0.634 and 0.620). The gain over VADER held when the review label was controlled for, in an approximately unselected subset of sentences and in sentences without terms added during the lexicon audit. The lexicon, its corpus statistics and the evaluation code are openly available.
No takes yet. Share an insight, caveat, or question.
Tibor Guzsvinecz (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: