Hi all,

So I've been spending some time searching for new series and found something fairly frustrating. I would stumble upon a series with both a high average and Bayesian score, but the recent comments/reviews were all negative. Case in point Kimi no Iru Machi.

Some series start out strong and fail over time, others seem very cliché at the outset but improve if you give it a chance. With that it mind I thought I would suggest adding a time series representation of scores.

It seems like the largest difficulty with this would be aggregating the data so that it is manageable, but because the time scale varies by series, up to several years of data, x-axis groupings could quickly become overwhelming or meaningless. A scaling binning system where data below a certain threshold is averaged by month and above that threshold averaged by quarter would probably be sufficient to keep even long running series trends readable.

It would also likely be helpful to remove outliers. For series like Naruto, eliminating outliers leaves a credible score range from [10,3). You can then report the data as "High, Average, Low," averaging within categories and reporting the average for each month. High and Low might be the number of scores above and below 1 standard deviation added in a month. There is a problem here as the scores are not normally distributed and are heavily skewed toward ~10 so the high line in many cases would just be a count of the 10s added in a month. The average line would just be the change in the average score with respect to time and is likely to be the most useful part of this.

Anyway, I just thought I'd add this and see what people thought.