Spaces:
Running
Running
| // A result set can miss the page the query actually asked for: an official | |
| // site's schedule or ticket page often ranks below aggregators and fan pages, | |
| // and a promotion of what the search did return cannot summon a page that was | |
| // never in it. This lane asks for that page directly, with a second SearXNG | |
| // query scoped to `site:`, and merges what comes back. | |
| // | |
| // Two conditions have to hold before the second query is worth its latency. | |
| // The query must carry navigation intent ("官网", "排片", "hours", "showtimes"), | |
| // and one of the returned results must already name the site ("disney" and | |
| // disney.com), so the domain is evidence from the search itself rather than a | |
| // guess. No domain is derived from the query alone: probing `<token>.com` would | |
| // turn a search box into a scanner. | |
| import { namedDomain, registrableDomainOf } from "./authorityRanking.ts"; | |
| /** How many results one lane may add; the lane is a supplement, not a feed. */ | |
| export const MAX_OFFICIAL_LANE_RESULTS = 3; | |
| // The query reaches the server with this ceiling already applied | |
| // (`server/searchEndpointServerHook.ts`), and the `site:` prefix must fit too. | |
| const MAX_QUERY_LENGTH = 2000; | |
| // Words that ask for an official page rather than reporting on one. Chinese has | |
| // no word boundaries to split on, so those phrases are matched in the raw query. | |
| const INTENT_PHRASES = [ | |
| "官网", | |
| "官方网站", | |
| "官方站", | |
| "官方网页", | |
| "官方页面", | |
| "片单", | |
| "排片", | |
| "场次", | |
| "票价", | |
| "营业时间", | |
| "上映时间", | |
| "购票", | |
| "订票", | |
| ]; | |
| const INTENT_WORDS = new Set([ | |
| "hours", | |
| "official", | |
| "opening", | |
| "pricing", | |
| "prices", | |
| "schedule", | |
| "showtime", | |
| "showtimes", | |
| "ticket", | |
| "tickets", | |
| "timetable", | |
| ]); | |
| export interface OfficialLaneRequest { | |
| /** The registrable domain the query named, as the second query scopes it. */ | |
| domain: string; | |
| /** The scoped query sent to the search engines. */ | |
| query: string; | |
| } | |
| /** | |
| * The second query this search warrants, or null when it warrants none. The | |
| * domain is taken from the results (`namedDomain`), so this only fires when the | |
| * official site already appeared somewhere in the set. | |
| */ | |
| export function officialLaneRequest( | |
| query: string, | |
| results: [title: string, content: string, url: string][], | |
| ): OfficialLaneRequest | null { | |
| if (!hasNavigationIntent(query)) return null; | |
| for (const [, , url] of results) { | |
| const domain = namedDomain(query, url, INTENT_WORDS); | |
| if (!domain) continue; | |
| const scopedQuery = buildScopedQuery(domain, query); | |
| if (scopedQuery) return { domain, query: scopedQuery }; | |
| } | |
| return null; | |
| } | |
| /** | |
| * Merges the lane's results into the main set, by URL, with the lane's pages | |
| * placed behind the pinned first result. The position matters: the ranking | |
| * keeps the first result where it is and re-sorts only what follows, so a page | |
| * appended past that window would never be re-sorted at all. | |
| */ | |
| export function mergeOfficialLaneResults( | |
| mainResults: [title: string, content: string, url: string][], | |
| laneResults: [title: string, content: string, url: string][], | |
| domain: string, | |
| ): [title: string, content: string, url: string][] { | |
| if (mainResults.length === 0) return mainResults; | |
| const seen = new Set(mainResults.map(([, , url]) => normalizeUrl(url))); | |
| const additions: [title: string, content: string, url: string][] = []; | |
| for (const result of laneResults) { | |
| if (additions.length >= MAX_OFFICIAL_LANE_RESULTS) break; | |
| const [, , url] = result; | |
| // An engine that ignores `site:` must not smuggle an unrelated host in. | |
| if (registrableDomainOf(url) !== domain) continue; | |
| const normalized = normalizeUrl(url); | |
| if (seen.has(normalized)) continue; | |
| seen.add(normalized); | |
| additions.push(result); | |
| } | |
| if (additions.length === 0) return mainResults; | |
| const [firstResult, ...rest] = mainResults; | |
| return [firstResult, ...additions, ...rest]; | |
| } | |
| /** | |
| * Off by default, because the lane sends a second request to the engines, and | |
| * an instance whose engines already rate-limit it should not have to opt out of | |
| * that. `SEARCH_OFFICIAL_LANE_ENABLED=true` turns it on. | |
| */ | |
| export function isOfficialLaneEnabled(): boolean { | |
| const value = (process.env.SEARCH_OFFICIAL_LANE_ENABLED ?? "") | |
| .trim() | |
| .toLowerCase(); | |
| return value === "true" || value === "1"; | |
| } | |
| function hasNavigationIntent(query: string): boolean { | |
| const lowered = query.toLowerCase(); | |
| if (INTENT_PHRASES.some((phrase) => lowered.includes(phrase))) return true; | |
| return lowered | |
| .split(/[^\p{L}\p{N}]+/u) | |
| .some((token) => token.length > 0 && INTENT_WORDS.has(token)); | |
| } | |
| function buildScopedQuery(domain: string, query: string): string | null { | |
| const scopedQuery = `site:${domain} ${query}`; | |
| return scopedQuery.length <= MAX_QUERY_LENGTH ? scopedQuery : null; | |
| } | |
| /** The scheme, `www.` and trailing slashes do not make two spellings of one page. */ | |
| function normalizeUrl(url: string): string { | |
| return url | |
| .replace(/^https?:\/\//i, "") | |
| .replace(/^www\./i, "") | |
| .replace(/\/+$/, "") | |
| .toLowerCase(); | |
| } | |