implemented weighting tag format.

web-ui-image-search-lsi.py

@@ -25,9 +25,37 @@ class Arguments(Protocol):
 
 args: Optional[Arguments] = None
 
+def normalize_and_apply_weight_to_quey_bow(query_bow: List[Tuple[int, int]], new_doc: str) -> List[Tuple[int, float]]:
+    tags: List[str] = new_doc.split(" ")
+
+    # parse tag:weight format
+    tag_and_weight_list: List[Tuple[str, float]] = []
+    for tag in tags:
+        tag_splited: List[str] = tag.split(":")
+        if len(tag_splited) == 2:
+            tag_and_weight_list.append((tag_splited[0], int(tag_splited[1])))
+        else:
+            tag_and_weight_list.append((tag_splited[0], 1))
+
+    # apply weight to query_bow
+    for tag, weight in tag_and_weight_list:
+        tag_id: int = dictionary.token2id[tag]
+        for ii in range(len(query_bow)):
+            if query_bow[ii][0] == tag_id:
+                query_bow[ii] = (query_bow[ii][0], query_bow[ii][1]*weight)
+                break
+
+    query_lsi: List[Tuple[int, float]] = model[query_bow]
+
+    # normalize query with tag num
+    tag_num: int = len(tags)
+    query_lsi = [(tag_id, tag_value / tag_num) for tag_id, tag_value in query_lsi]
+    return query_lsi
+
 def find_similar_documents(model: LsiModel, new_doc: str, topn: int = 50) -> List[Tuple[int, float]]:
     query_bow: List[Tuple[int, int]] = dictionary.doc2bow(simple_preprocess(new_doc))
-    query_lsi: List[Tuple[int, float]] = model[query_bow]
+    query_lsi = normalize_and_apply_weight_to_quey_bow(query_bow, new_doc)
+    #query_lsi: List[Tuple[int, float]] = model[query_bow]
 
     sims: List[Tuple[int, float]] = index[query_lsi]