How WriterMate analyses trends

We collect publicly available web-serial ranking pages in each country. Nothing behind a login or a paywall is used.

The originals are kept as collected. Rankings, titles and covers on screen are each platform's own; shares and changes are calculated by WriterMate from those originals.

Once a week, with weeks divided on Monday in Korea Standard Time. The “week covered” shown on screen runs Monday to Sunday of that week.

If the same platform was collected twice in one week, only the last collection of that week is used.

Week = KST Monday 00:00 → the following Monday 00:00 (exclusive)

The top 20 titles by rank in each snapshot. The cap is fixed so shares do not swing when the collected volume differs week to week.

Long-run trajectories accumulate up to 12 weeks.

• Title (original and Korean)
• Author
• Genre as assigned by the platform
• Synopsis
• Tags and keywords
• Rank, and rank movement against last week

Genre names differ by platform. Counting them as-is makes cross-market comparison meaningless, so everything is normalised once into WriterMate's taxonomy after collection. Chinese “xianxia” and Korean “wuxia” are grouped as one lineage; “urban fantasy / urban-fantasy” spellings are merged.

A genre that matches no normalisation rule keeps its original name. We never force it into a different genre.

A title carrying several genres counts once for each, so the shares can add up to more than 100%.

Genre share (%) = (titles carrying that genre ÷ titles counted that week) × 100

Tropes, themes and settings are taken only from expressions stated outright in the title, synopsis or tags. We do not read the body text, and we do not infer settings that were never stated.

Translation sometimes returns a different name for the same concept from week to week (“reincarnation” one week, “reincarnator” the next). Counted as-is, a renamed concept would register as a new entry or a collapse, so names are unified before counting.

Expressions longer than 20 characters are treated as description rather than a label and excluded; a trope repeated within one title counts once. A trope carried by fewer than 2 titles is treated as noise and dropped.

• Theme: the spine of the story — revenge, growth, survival, politics, management
• Trope: the device — regression, possession, system, hunter, awakening
• Setting: the stage — wulin, modern Korea, apocalypse, isekai, dungeon
• Spacing- or punctuation-only variants are treated as the same label
• Distinct words with the same meaning are merged only when listed in a fixed table
• The displayed name is whichever variant actually occurred most often in the data

Every judgement comes from the formulas below. The same data always yields the same result.

The unit of judgement is titles, not percent. With a sample of 10–20 titles a week, one title is already 5–10 percentage points. A percentage threshold alone would flag every single swap as meaningful, leaving no real noise floor.

Share (%) = (titles carrying the item ÷ titles counted that week) × 100

Change (pp) = this week's share − last observed week's share

• Rise or fall: only when the item gained or lost at least 2 titles against the last observed week. A difference of one is shown as holding steady, and no direction is claimed.
• Streak length: single-title gains are counted here. One week alone is noise, but the same direction repeating is itself a signal. To avoid an artefact from a different collection volume, a week counts only when share and title count move the same way.
• New entry: absent for the last 4 weeks, appearing this week, and carried by at least 2 titles.
• Co-occurring trope: another trope appearing in 40% or more of the titles that carry the target trope.

The four markets' changes for the same week are placed side by side, and only the direction is read.

The reference market follows the language you are reading in — Korean puts Korea first, Japanese puts Japan first. The underlying counts are identical either way; only the ordering changes.

Because ranking criteria differ by platform, direction is the only comparable thing. We never read absolute figures as “this market is bigger”.

• Aligned rise: rising in 2 or more markets in the same week (each gaining at least 2 titles)
• Divergent: rising in one market while falling in another
• Absent at home: zero titles on the reference market's chart while rising elsewhere

What AI does and does not do is separated explicitly.

AI does not produce figures such as shares or changes. Those come only from the formulas in section 6, and the already-calculated values are handed to AI as input it cannot alter.

The block labelled “story pattern read from the synopses” is AI reading synopses. Three machine checks are applied to it.

That is why the source titles and their count are always shown next to it — so you can check what the sentence was based on.

• If a title not present in the input is cited, that citation is removed; if no evidence remains, the whole item is withheld.
• If a sentence invents a figure (percent, percentage points, multiples), that sentence is removed. The only quantity allowed is counting the titles we supplied, as in “5 of 8 titles”.
• If nothing survives the checks, the item does not appear on screen at all.

Below is what this data cannot tell you. It is published alongside so the figures are not misread.

• Ranking criteria differ by platform (views, rating scores, paid votes). These figures therefore cannot be used to compare market size or revenue across countries. Only “which way it moved in the same week” is comparable.
• Only the top of each chart is observed. Movement outside the ranked range is invisible here.
• Tropes and genres come only from expressions in the title, synopsis and tags. We do not read body text, so settings never stated in the synopsis are not counted. Where a platform publishes no synopsis, no story pattern is produced.
• These are once-a-week snapshots, so rank movement within a single week is not reflected.
• The sample is roughly 10–20 titles a week (chart depth differs by country). One title is 5–10 percentage points, so small percentage differences should not be read as meaningful.
• Collection can fail, or a platform's markup can change, leaving a week empty. In that case the empty week is skipped and the comparison uses the last observed week, which is shown on screen.