Similarity Explorer

Compare poem or verse similarities across the Finnic runosong corpus (292,092 poems, 4.36M verses) using multiple algorithms.
Two modes: Poem Similarity compares entire poems; Verse Similarity compares individual verse lines.

TF-IDF Lemma (green headers): Groups word forms by lemma (dictionary form), then measures cosine distance. Rarer shared lemmas contribute more than common ones. Works within one language.
Jaccard Wordform (blue headers): Compares exact word forms via set intersection (|A∩B| / |A∪B|). Stricter than TF-IDF because morphological variants don't match.
Translation-Pivot (teal headers): Maps verses through their English translations, enabling cross-lingual Estonian↔Finnish matches. A verse in Estonian and a verse in Finnish can match if they share translated meanings.
Alignment (indigo headers): Aligns the verse sequences of two poems using character n-gram similarity and dynamic programming (Wagner-Fischer algorithm). From the FILTER project. Captures structural similarity — poems that follow the same verse order score high even if individual words differ.
Verse Match (purple headers): Runs four verse-level algorithms (TF-IDF, Jaccard, Translation, CharBigram) on individual verse pairs, then aggregates to poem level via RRF. Captures fine-grained verse-by-verse similarity.
RRF (Reciprocal Rank Fusion): Combines TF-IDF, Jaccard, Translation, and Alignment at poem level. If a match ranks high in multiple algorithms, its combined score is boosted. Shown as a percentage (0-100%).

Search by archive reference (e.g., "H II 1, 389") to load a poem and see its most similar poems.
Two-column comparison: pick any two of six algorithms to view side-by-side. Each column shows the top 50 most similar poems. Or use Combined (RRF) for a single fused ranking.
Score threshold slider filters out weak matches. Cross-lingual filter (on Translation, Alignment) limits to Estonian↔Finnish matches.
Click a poem ID to chain-navigate (load that poem's similarities). Breadcrumbs track your exploration path.
Shared items show what the poems have in common: lemmas (TF-IDF), wordforms (Jaccard), translations (Translation-pivot), or aligned verse pairs (Alignment).

Search for a poem, then click any verse line to see its similar verses across the corpus.
Four algorithm columns (TF-IDF, Jaccard, Translation-pivot, CharBigram) show the top verse matches each, plus an RRF combined view.
Cross-lingual badge (ET↔FI) appears when a translation-pivot match crosses the language boundary.
Click verse IDs to navigate to that verse in the Poem Reader.

In verse mode, the sidebar shows the top 200 formulaic verse clusters discovered across the corpus.
A formula cluster groups verses that are nearly identical across many poems (high-frequency recurring lines).
Click a cluster to load its representative verse and explore its similar verses.

Shows a map of the strongest place-to-place connections based on verse-level similarity across the entire corpus.
Each line connects two places that share similar verse pairs. Line thickness and opacity reflect connection strength (number of shared verse pairs).
Green lines = same-language connections (ET↔ET or FI↔FI). Gold lines = cross-lingual connections (ET↔FI).
Circle size reflects how many verse-pair connections involve that place.
Click a line to see the top matching verse pairs with links to the Poem Reader.
Use "Cross-lingual only" to see only Estonian↔Finnish geographic bridges (uses a dedicated cross-lingual index of 5,000 connections).
Data: 243,118 unique place connections from 59.4M verse-pair matches across 4 algorithms (Jaccard, TF-IDF, Translation, CharBigram).

Select a specific place to see its connections radiate outward on the map. Search by name using the autocomplete dropdown.
Circle size at connected places reflects connection strength using the selected metric.
Each place shows its top 50 connections (by strength) plus top 10 cross-lingual connections separately.
Click a connection line for details: verse pair count, strength score, surprise score, and top matching verse pairs with Poem Reader links.
Click "Focus on this place" in a marker popup to chain-navigate (explore from that place). Use the browser back button to return.

Tradition Strength: For each pair of places, sums the best similarity score (across 4 algorithms) for every unique verse pair between them. Rewards both quantity (many shared verses) and quality (high-scoring matches). Larger circles = more high-quality shared poetic material. Naturally highlights neighboring places with extensive shared traditions.
Surprising Connections: Normalizes the tradition strength by the geometric mean of both places' verse counts: strength / √(verses_A × verses_B). This controls for corpus size — a small place sharing many strong verse pairs with a distant place scores higher than two large neighboring places sharing the same number. Highlights unexpected long-range and cross-lingual connections. Warning: unreliable for places with fewer than 200 verses.

Click poem/verse IDs to chain-navigate (explore from that poem/verse).
Click "explore" links to view a poem in the standalone Similarity Explorer or Poem Reader.
Breadcrumbs at the top track your navigation history. Click any breadcrumb to go back.
Use the URL parameters ?poem=ID or ?verse=ID for deep linking.
Geography deep links: ?mode=geography&submode=placefocus&place=Kuusalu&viz=tradition

Poem Similarity Explorer

Compare poems across two similarity algorithms side-by-side. Search for any poem above, or use ?poem=ID in the URL.

Six algorithms: TF-IDF Lemma, Jaccard Wordform, Translation-Pivot, Alignment, Verse Match, and Combined RRF.

Formulaic Patterns (curated entry points)

Loading formulaic patterns...