Add recompute button to bypass cache for analysis

- Add force_recompute param to compute_similar_talks/compute_topic_clusters
- Add ?recompute=1 query param support to compute-analysis endpoint
- Show "Recompute (ignore cache)" button after initial results load
- Replace globals with functools.cache, simplify _get_submission_languages
- Remove unused language mappings, keep only English and Italian

Author: patrick91

backend/reviews/admin.py

@@ -447,13 +447,20 @@ def review_recap_compute_analysis_view(self, request, review_session_id):
 
         conference = review_session.conference
         accepted_submissions = self._get_accepted_submissions(conference)
+        force_recompute = request.GET.get("recompute") == "1"
 
         similar_talks = compute_similar_talks(
-            accepted_submissions, top_n=5, conference_id=conference.id
+            accepted_submissions,
+            top_n=5,
+            conference_id=conference.id,
+            force_recompute=force_recompute,
         )
 
         topic_clusters = compute_topic_clusters(
-            accepted_submissions, min_topic_size=3, conference_id=conference.id
+            accepted_submissions,
+            min_topic_size=3,
+            conference_id=conference.id,
+            force_recompute=force_recompute,
         )
 
         # Build submissions list with similar talks, sorted by highest similarity

backend/reviews/similar_talks.py

@@ -253,14 +253,17 @@ def _get_cache_key(prefix: str, conference_id: int, submissions) -> str:
     return f"{prefix}:conf_{conference_id}:{content_hash.hexdigest()}"
 
 
-def compute_similar_talks(submissions, top_n=5, conference_id=None):
+def compute_similar_talks(
+    submissions, top_n=5, conference_id=None, force_recompute=False
+):
     """
     Compute similar talks for each submission based on embeddings.
 
     Args:
         submissions: A list of Submission objects
         top_n: Number of similar talks to return for each submission
         conference_id: Optional conference ID for cache key
+        force_recompute: Skip cache and recompute fresh results
 
     Returns:
         A dictionary mapping submission IDs to lists of similar submission IDs
@@ -276,9 +279,10 @@ def compute_similar_talks(submissions, top_n=5, conference_id=None):
     cache_key = None
     if conference_id:
         cache_key = _get_cache_key("similar_talks", conference_id, submissions_list)
-        cached_result = cache.get(cache_key)
-        if cached_result is not None:
-            return cached_result
+        if not force_recompute:
+            cached_result = cache.get(cache_key)
+            if cached_result is not None:
+                return cached_result
 
     model = get_embedding_model()
 
@@ -309,14 +313,17 @@ def compute_similar_talks(submissions, top_n=5, conference_id=None):
     return similar_talks
 
 
-def compute_topic_clusters(submissions, min_topic_size=3, conference_id=None):
+def compute_topic_clusters(
+    submissions, min_topic_size=3, conference_id=None, force_recompute=False
+):
     """
     Compute topic clusters for submissions using BERTopic.
 
     Args:
         submissions: A list of Submission objects
         min_topic_size: Minimum number of talks per cluster
         conference_id: Optional conference ID for cache key
+        force_recompute: Skip cache and recompute fresh results
 
     Returns:
         A dictionary with:
@@ -334,9 +341,10 @@ def compute_topic_clusters(submissions, min_topic_size=3, conference_id=None):
     cache_key = None
     if conference_id:
         cache_key = _get_cache_key("topic_clusters_v4", conference_id, submissions_list)
-        cached_result = cache.get(cache_key)
-        if cached_result is not None:
-            return cached_result
+        if not force_recompute:
+            cached_result = cache.get(cache_key)
+            if cached_result is not None:
+                return cached_result
 
     model = get_embedding_model()
     texts = [get_embedding_text(s) for s in submissions_list]

backend/reviews/templates/reviews-recap.html

@@ -434,12 +434,15 @@ <h3>🏷️ Tags</h3>
   {% endfor %}
 
   <!-- Compute Analysis Button -->
-  <div class="recap-section" style="text-align: center;">
+  <div id="analysis-controls" class="recap-section" style="text-align: center;">
     <button type="button" id="compute-analysis-btn" class="button" style="padding: 10px 30px; font-size: 16px; cursor: pointer;">
       Compute Topic Clusters &amp; Similar Talks
     </button>
+    <button type="button" id="recompute-analysis-btn" class="button" style="padding: 10px 30px; font-size: 16px; cursor: pointer; display: none;">
+      Recompute (ignore cache)
+    </button>
     <div id="analysis-loading" style="display: none; margin-top: 15px;">
-      <span style="color: var(--body-quiet-color, #666);">Computing analysis (this may take a moment on first run)...</span>
+      <span style="color: var(--body-quiet-color, #666);">Computing analysis (this may take a moment)...</span>
     </div>
     <div id="analysis-error" style="display: none; margin-top: 15px; color: #c0392b;"></div>
   </div>
@@ -470,6 +473,7 @@ <h2 class="recap-section-title">🔗 Similar Talks</h2>
 <script>
 (function() {
   const btn = document.getElementById('compute-analysis-btn');
+  const recomputeBtn = document.getElementById('recompute-analysis-btn');
   const loading = document.getElementById('analysis-loading');
   const errorDiv = document.getElementById('analysis-error');
   const computeUrl = '{{ compute_analysis_url }}';
@@ -566,13 +570,16 @@ <h2 class="recap-section-title">🔗 Similar Talks</h2>
     section.style.display = '';
   }
 
-  btn.addEventListener('click', function() {
-    btn.disabled = true;
-    btn.textContent = 'Computing...';
+  function fetchAnalysis(recompute) {
+    var url = recompute ? computeUrl + '?recompute=1' : computeUrl;
+    var activeBtn = recompute ? recomputeBtn : btn;
+
+    activeBtn.disabled = true;
+    activeBtn.textContent = recompute ? 'Recomputing...' : 'Computing...';
     loading.style.display = '';
     errorDiv.style.display = 'none';
 
-    fetch(computeUrl, {
+    fetch(url, {
       headers: { 'X-Requested-With': 'XMLHttpRequest' }
     })
     .then(function(response) {
@@ -582,18 +589,24 @@ <h2 class="recap-section-title">🔗 Similar Talks</h2>
     .then(function(data) {
       loading.style.display = 'none';
       btn.style.display = 'none';
+      recomputeBtn.style.display = '';
+      recomputeBtn.disabled = false;
+      recomputeBtn.textContent = 'Recompute (ignore cache)';
 
       renderTopicClusters(data.topic_clusters);
       renderSimilarTalks(data.submissions_list);
     })
     .catch(function(err) {
       loading.style.display = 'none';
-      btn.disabled = false;
-      btn.textContent = 'Compute Topic Clusters & Similar Talks';
+      activeBtn.disabled = false;
+      activeBtn.textContent = recompute ? 'Recompute (ignore cache)' : 'Compute Topic Clusters & Similar Talks';
       errorDiv.textContent = 'Failed to compute analysis: ' + err.message;
       errorDiv.style.display = '';
     });
-  });
+  }
+
+  btn.addEventListener('click', function() { fetchAnalysis(false); });
+  recomputeBtn.addEventListener('click', function() { fetchAnalysis(true); });
 })();
 </script>
 {% endblock %}