feat(deepseek): normalize model labels and unwrap redirect URLs

- Normalize provider_model so UI shows DeepSeek/DeepSeek-R1 instead of raw toggle text (联网搜索 etc.) - Unwrap DeepSeek redirect URLs (chat.deepseek.com/api/redirect?url=...) to canonical sources - Expand reasoning sections and separate reasoning search pages from answer-body citations - Improve webpages-trigger click heuristics and auxiliary link filtering in answer rendering Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-24 23:37:16 +08:00
parent bc2e23cfcb
commit 4f3c39f5f5
3 changed files with 940 additions and 123 deletions
@@ -220,6 +220,39 @@ function statusLabel(status: string): string {
  return t(`tracking.status.${status}`);
 }

+function resolveDeepSeekProviderModelLabel(value: string | null | undefined): string {
+  const normalized = String(value ?? "").trim();
+  if (!normalized || normalized === "--") {
+    return "DeepSeek";
+  }
+
+  const compact = normalized.replace(/\s+/g, "").toLowerCase();
+  if (/r1|reasoner|deepthink|深度思考/.test(compact)) {
+    return "DeepSeek";
+  }
+
+  const versionMatch =
+    normalized.match(/deepseek[-_\s]*(r1|v\d+(?:\.\d+)?)/i)
+    ?? normalized.match(/\b(r1|v\d+(?:\.\d+)?)\b/i);
+  if (versionMatch?.[1]) {
+    return `DeepSeek-${versionMatch[1].toUpperCase()}`;
+  }
+
+  if (/search|web|联网|搜索/.test(compact)) {
+    return "DeepSeek";
+  }
+
+  return normalized;
+}
+
+function resolveAnswerProviderModel(platform: MonitoringQuestionDetailPlatform | null): string {
+  if (isDeepSeekInlineCitationPlatform(platform?.ai_platform_id)) {
+    return resolveDeepSeekProviderModelLabel(platform?.provider_model);
+  }
+
+  return String(platform?.provider_model ?? "").trim() || "--";
+}
+
 function resolveCitationTitle(citation: MonitoringQuestionDetailCitation): string {
  return citation.cited_title || citation.article_title || citation.cited_url;
 }
@@ -374,6 +407,78 @@ function buildCitationSourceLookup(platform: MonitoringQuestionDetailPlatform |
  return lookup;
 }

+function buildDeepSeekAnswerLinkSourceLookup(
+  platform: MonitoringQuestionDetailPlatform | null,
+): Map<string, { title: string | null; siteName: string | null }> {
+  const lookup = new Map<string, { title: string | null; siteName: string | null }>();
+
+  for (const item of platform?.inline_citations ?? []) {
+    const normalizedURL = normalizeInlineCitationUrl(item.normalized_url ?? item.url);
+    if (!normalizedURL || lookup.has(normalizedURL)) {
+      continue;
+    }
+    lookup.set(normalizedURL, {
+      title: item.title?.trim() || null,
+      siteName: item.site_name?.trim() || null,
+    });
+  }
+
+  for (const citation of platform?.citations ?? []) {
+    const normalizedURL = normalizeInlineCitationUrl(citation.cited_url);
+    if (!normalizedURL || lookup.has(normalizedURL)) {
+      continue;
+    }
+    lookup.set(normalizedURL, {
+      title: resolveCitationTitle(citation),
+      siteName: citation.site_name?.trim() || null,
+    });
+  }
+
+  return lookup;
+}
+
+function isHiddenDeepSeekAnswerNode(element: Element): boolean {
+  let current: Element | null = element;
+  for (let depth = 0; current && depth < 8; depth += 1) {
+    if (current.hasAttribute("hidden") || current.getAttribute("aria-hidden") === "true") {
+      return true;
+    }
+    const style = current.getAttribute("style") ?? "";
+    if (/(display\s*:\s*none|visibility\s*:\s*hidden|opacity\s*:\s*0)/i.test(style)) {
+      return true;
+    }
+    current = current.parentElement;
+  }
+  return false;
+}
+
+function isInsideDeepSeekAuxiliaryLinkContainer(anchor: HTMLAnchorElement): boolean {
+  let current = anchor.parentElement;
+  for (let depth = 0; current && depth < 8; depth += 1) {
+    const descriptor = `${current.className || ""} ${current.getAttribute("data-testid") || ""} ${current.getAttribute("role") || ""}`;
+    if (/(search-view-card|search-result|result-card|source-card|reference-card|citation-card|popover|tooltip|dropdown|floating|hover-card|preview-card)/i.test(descriptor)) {
+      return true;
+    }
+    current = current.parentElement;
+  }
+  return false;
+}
+
+function shouldCountDeepSeekAnswerLink(anchor: HTMLAnchorElement): boolean {
+  if (isHiddenDeepSeekAnswerNode(anchor) || isInsideDeepSeekAuxiliaryLinkContainer(anchor)) {
+    return false;
+  }
+
+  const text = String(anchor.textContent ?? "").trim();
+  if (normalizeDeepSeekCitationLabel(text)) {
+    return true;
+  }
+  if (shouldStripDeepSeekOrphanAnchor(anchor)) {
+    return false;
+  }
+  return /[A-Za-z0-9\u3400-\u9fff]/.test(text);
+}
+
 function normalizeDeepSeekCitationLabel(value: string | null | undefined): string | null {
  const normalized = String(value ?? "").trim().replace(/\s+/g, " ");
  if (!normalized) {
@@ -416,13 +521,16 @@ function pushUniqueDeepSeekCitationLabel(labels: string[], label: string): void

 function shouldStripDeepSeekOrphanAnchor(anchor: HTMLAnchorElement): boolean {
  const text = String(anchor.textContent ?? "").trim().replace(/\s+/g, "");
+  if (/^(link|source|citation|reference|open|原文|链接|来源|引用)$/i.test(text)) {
+    return true;
+  }
  if (/[A-Za-z0-9\u3400-\u9fff]/.test(text)) {
    return false;
  }
  if (anchor.querySelector("img, svg")) {
    return true;
  }
-  return text.length === 0 || /^[\[\](){}<>#.,:;!?+\-_/\\|~`'"“”‘’·•:：，。！？；、🔗]+$/.test(text);
+  return text.length === 0 || /^[\[\](){}<>#.,:;!?+\-_/\\|~`'"“”‘’·•:：，。！？；、🔗↗↘↙↖↪⤴]+$/.test(text);
 }

 function collectDeepSeekCitationReferenceLabels(
@@ -489,6 +597,17 @@ function resolveCitationIndexDisplayLabels(
  return labels;
 }

+function shouldShowCitationSourceListIndex(
+  citationIndex: number,
+  platform: MonitoringQuestionDetailPlatform | null,
+): boolean {
+  if (!isDeepSeekInlineCitationPlatform(platform?.ai_platform_id)) {
+    return true;
+  }
+
+  return resolveCitationIndexDisplayLabels(citationIndex, platform).length === 0;
+}
+
 function countHanCharacters(value: string): number {
  return value.match(/[\u3400-\u9fff]/g)?.length ?? 0;
 }
@@ -575,6 +694,8 @@ function renderDeepSeekAnswerContent(
    if (!source && !citationSource) {
      if (shouldStripDeepSeekOrphanAnchor(anchor)) {
        anchor.remove();
+      } else {
+        anchor.replaceWith(document.createTextNode(anchor.textContent ?? ""));
      }
      continue;
    }
@@ -657,28 +778,26 @@ function collectDeepSeekInlineCitationOccurrences(
    return [];
  }

-  const inlineCitationLookup = buildInlineCitationSourceLookup(platform);
-  if (!inlineCitationLookup.size) {
-    return [];
-  }
+  const sourceLookup = buildDeepSeekAnswerLinkSourceLookup(platform);

  const counts = new Map<string, { label: string | null; siteName: string | null; citationCount: number }>();
  for (const anchor of Array.from(root.querySelectorAll<HTMLAnchorElement>("a[href]"))) {
    const normalizedURL = normalizeInlineCitationUrl(anchor.getAttribute("href") || anchor.href || "");
-    if (!normalizedURL || !inlineCitationLookup.has(normalizedURL)) {
+    if (!normalizedURL || !shouldCountDeepSeekAnswerLink(anchor)) {
      continue;
    }

-    const source = inlineCitationLookup.get(normalizedURL) ?? null;
+    const source = sourceLookup.get(normalizedURL) ?? null;
    const existing = counts.get(normalizedURL);
    if (existing) {
+      // One row per URL, but every visible answer occurrence contributes to the count.
      existing.citationCount += 1;
      continue;
    }

    counts.set(normalizedURL, {
-      label: source?.title?.trim() || null,
-      siteName: source?.site_name?.trim() || null,
+      label: source?.title || anchor.getAttribute("title")?.trim() || null,
+      siteName: source?.siteName || null,
      citationCount: 1,
    });
  }
@@ -996,7 +1115,7 @@ const contentCitationColumns = computed(() => [

            <div class="tracking-question-answer__meta">
              <span>{{ activePlatform?.sampled_at ? formatDateTime(activePlatform.sampled_at) : "--" }}</span>
-              <span>{{ activePlatform?.provider_model ?? "--" }}</span>
+              <span>{{ resolveAnswerProviderModel(activePlatform) }}</span>
            </div>
          </div>
        </section>
@@ -1031,6 +1150,7 @@ const contentCitationColumns = computed(() => [
                      {{ label }}
                    </span>
                    <span
+                      v-if="shouldShowCitationSourceListIndex(index, activePlatform)"
                      class="tracking-question-source-card__index"
                      :class="{ 'tracking-question-source-card__index--secondary': resolveCitationIndexDisplayLabels(index, activePlatform).length > 0 }"
                    >
@@ -97,6 +97,40 @@ describe("deepseek adapter helpers", () => {
    expect(classified.searchResults).toHaveLength(1);
  });

+  it("keeps reasoning search pages as citation sources separate from answer content citations", () => {
+    const classified = classifyDeepseekSources({
+      inlineLinks: [
+        {
+          url: "https://example.com/body-citation",
+          title: "Answer body citation",
+          text: "[15]",
+          siteName: "Body Site",
+          kind: "inline",
+        },
+      ],
+      sourcePanelLinks: [],
+      searchPanelLinks: [
+        {
+          url: "https://example.com/reasoning-search",
+          title: "Reasoning searched page",
+          text: "浏览页面",
+          siteName: "Search Site",
+          kind: "search",
+        },
+      ],
+    });
+
+    expect(classified.inlineCitationCandidates.map((item) => item.url)).toEqual([
+      "https://example.com/body-citation",
+    ]);
+    expect(classified.citations.map((item) => item.url)).toEqual([
+      "https://example.com/reasoning-search",
+    ]);
+    expect(classified.searchResults.map((item) => item.url)).toEqual([
+      "https://example.com/reasoning-search",
+    ]);
+  });
+
  it("retains all revealed DeepSeek search results instead of truncating after 32 items", () => {
    const searchPanelLinks = Array.from({ length: 78 }, (_, index) => ({
      url: `https://example.com/search-${index + 1}`,
@@ -173,6 +207,22 @@ describe("deepseek adapter helpers", () => {
    }));
  });

+  it("unwraps DeepSeek redirect URLs before storing source items", () => {
+    const item = buildSourceItem({
+      url: "https://chat.deepseek.com/api/redirect?url=https%3A%2F%2Fexample.com%2Fsource%23ref",
+      title: "Redirected source",
+      text: null,
+      siteName: null,
+      kind: "search",
+    });
+
+    expect(item).toEqual(expect.objectContaining({
+      url: "https://example.com/source",
+      normalized_url: "https://example.com/source",
+      host: "example.com",
+    }));
+  });
+
  it("does not use numeric citation markers as source titles", () => {
    const item = buildSourceItem({
      url: "https://example.com/report#cite-1",
@@ -217,6 +267,45 @@ describe("deepseek adapter helpers", () => {
    expect(isDeepSeekWebpagesTriggerText("搜索到 39 个网页")).toBe(false);
  });

+  it("normalizes DeepSeek model labels without mistaking search toggles for models", () => {
+    const snapshot = {
+      url: "https://chat.deepseek.com/",
+      title: "DeepSeek",
+      loginRequired: false,
+      loginReason: null,
+      challengeRequired: false,
+      challengeReason: null,
+      busy: false,
+      busySignals: [],
+      composerValue: null,
+      sendDisabled: false,
+      answer: "Final answer",
+      answerText: "Final answer",
+      reasoning: null,
+      signature: "sig-1",
+      currentModelLabel: "深度思考 (R1)",
+      providerRequestID: null,
+      requestID: null,
+      inlineLinks: [],
+      sourcePanelLinks: [],
+      searchPanelLinks: [],
+    };
+    const sources = {
+      citations: [],
+      inlineCitationCandidates: [],
+      searchResults: [],
+    };
+
+    const r1Payload = buildDeepSeekPayload(snapshot, sources, "Final answer");
+    const searchPayload = buildDeepSeekPayload({
+      ...snapshot,
+      currentModelLabel: "联网搜索",
+    }, sources, "Final answer");
+
+    expect(r1Payload.provider_model).toBe("DeepSeek-R1");
+    expect(searchPayload.provider_model).toBeNull();
+  });
+
  it("builds json-safe payloads with null placeholders instead of undefined", () => {
    const payload = buildDeepSeekPayload({
      url: "https://chat.deepseek.com/",