Skip to content

Sentiment · 2026-27

Headlines

How fans and media talk about players and teams. The two are shown side by side, never merged into one number.

Data as of Oct 10, 2026Headlines· Oct 102,646 headlinesFans· Oct 1047,800 fan posts

A language model rates the tone of headlines, fan posts and comments toward the players they name. Team scores and fan posts from before the model still use a word list, so read those as rough.

Sources and method
  • Headlines· Oct 10Media lanes. 2,646 NBA headlines from 23 publishers, Jul 24 to Oct 10. A player needs 3 headlines in the last 7 days to get a score. A language model rates each headline as positive, neutral or negative toward each player it names. Checked against 127 labeled headlines, it matched the label about 7 times in 10 and never rated one the opposite way. Players named only in the summary, and team scores, still use a word list. 73% of this week's player mentions have a model rating.
  • Fans· Oct 10Fan lanes. 47,800 fan posts: 1,561 headlines from 30 team fan blogs, 3,871 social posts and 42,368 video comments, Jun 25 to Oct 10. A player or team needs 5 posts in the last 7 days to get a score. Comments under the same video share weight, so 100 comments on one video count for less than 100 posts spread across many places. Fan blog headlines get the same model rating as news headlines. Social posts and video comments that name a player are rated by the same model when they are collected, up to a nightly limit. Checked against 143 labeled fan posts, it matched the label about 3 times in 4 and rated 2 the opposite way. The word list matched about half and rated 18 the opposite way. Older posts, posts past the nightly limit and team scores keep the word list. 66% of this week's player mentions in posts and comments have a model rating.
How often the tone rating matches a hand labelSame as the labelOne step offOpposite side

News headlines 63 headlines

Language model71% match0 opp.
Word list52% match2 opp.

Social posts 72 posts

Language model76% match0 opp.
Word list44% match6 opp.

Video comments 71 comments

Language model72% match2 opp.
Word list54% match12 opp.

A coding agent labeled each one positive, neutral or negative toward one named player, not a human panel. One step off means a neutral rating where the label picked a side, or the reverse. The headline set leaves out the half used to tune the word list.

Fan and media tone are kept separate from performance metrics and Movement Center evidence. The privacy page lists what is kept from each source.

Latest headlines

The newest headlines behind the media lanes, with each one's tone score. Expect some misreads.

How this works

When a headline names one player, the label is a language model's rating of its tone toward that player. Other headlines show a word-list score from the headline and the first lines of the summary, where 50% is neutral. 2,646 headlines stored since Jul 24, 2026.