IFDB's home page "IFDB Recommends" shouldn't randomize

[dfabulich]

The IFDB home page has an "IFDB Recommends" section wayyy down at the bottom.

These are a few randomly-selected games with high average member ratings.

IMO, this section shouldn't randomize. Instead, it should just show the best-ranked games on IFDB, as a leaderboard.

I see it as similar to the "Reviewer Trophy Room" section on the home page, which shows the top 5 reviewers on IFDB. It wouldn't make sense to show five random reviewers. In this setting, we want the best.

[dfabulich]

I've filed a separate issue iftechfoundation/ifdb#863 that IFDB shouldn't just sort by average rating when determining the best games, but should instead incorporate the number of ratings.

But even if this section did incorporate the number of ratings, I still wouldn't want it to show a random sample from the IFDB Top 100… it should just show a few of the best games.

[qdacsvx]

-1 won't it be boring to see the same game list every time I visit IFDB? What if I already played all the games on the list?

[huftis]

-1 won't it be boring to see the same game list every time I visit IFDB? What if I already played all the games on the list?

The recommended list should only show games you haven’t played (when you’re logged in).

Also, I recommend that instead of the raw average scores, a naïve Bayesian estimator is used, just like for the IMDb. This takes the number of ratings into account. (Otherwise, a game with a single rating will be the top game if that rating is a 5.)

Basically, the rating ‘estimate’ is a weighted average of the 1) the game’s average rating, and 2) the average rating of all the games on IFDb. For games with a high number of ratings, the game’s average rating will get more heavily weighted. (There also exists more advanced estimators, which takes the rating patterns of the users into account, e.g., a, in general, ‘generous’ rater giving a high rating won’t raise a game’s rating by very much. But I don’t think this is necessary. A naïve Bayes estimator is good enough.)

You can see a list of the current top IFDb games based on a Bayesian estimator at: https://ifdb.tads.org/viewlist?id=k7rrytlz3wihmx2o

[dfabulich]

FYI, improving the "sort by rating" is filed as issue iftechfoundation/ifdb#863.

[qdacsvx]

I reviewed the live Top 100 List on IFDB. Only 3 works were published in the past year. This does nothing to allay the concern that the table will be static with the greatest hits and new games (and new authors) will have a negligible chance to appear.

@dfabulich I appreciate your efforts to improve searches.

[huftis]

I reviewed the live Top 100 List on IFDB. Only 3 works were published in the past year. This does nothing to allay the concern that the table will be static with the greatest hits and new games (and new authors) will have a negligible chance to appear.

OK, how about using the “Top 100” feature as a general “view” of the currently selected set of games? If you select “Top 100” in general, you would basically get the top 100 games list I linked to, i.e., the “all-time greatest hits”. But if you’re browsing the “games released in 2020” category, you would get a list of the top 100 games from 2020 (ranked using the fancy algorithm). And if you select the “science fiction” genre, you would get a list of the top 100 science fiction games. Etc. There could even by a couple of default top lists, e.g., the top 100 (or 10?) games released in the last two years, or a top 100 (or 10?) list of games in a certain (random) “genre” (= a “popular” keyword).