zz

mongo/mongocrawler.py

@@ -32,6 +32,10 @@ MINTEXTSIZE=200
 CHECK_PARAGRAPH_SIZE=150
 TEXT_TRASH_SIZE=200
 TEXT_TRASH_RATIO=0.6
+DISCOVER_LINK_RATIO = 0.3
+SAMPLE_SET_SIZE =10000
+CLASSIFIER_SET_SIZE = 200
+STOP_PATHS=["xml","rss","login","admin"]
 
 def split_train(res):
     trainset = []
@@ -83,6 +87,11 @@ def is_link_good(link):
     if r is None:
         return None
     llink,lhostname = r
+    paths = set(llink.split("/"))
+    for item in STOP_PATHS:
+        if item in paths:
+            return None
+
     #print(llink,lhostname)
     # hostname rules
     if not lhostname.endswith(DOMAIN):
@@ -122,12 +131,13 @@ def fetch_page(link:str)->(str,str):
             LOGGER.error('too large: length %s for URL %s', len(response.data), link)
         if good:
             html = trafilatura.utils.decode_response(response) 
-            final_link = response.url
         if html is not None:
             html, final_link = trafilatura.spider.refresh_detection(html, final_link)
             # is there a meta-refresh on the page?
             if final_link is None:  # malformed or malicious content
                 html = None
+
+        final_link = courlan.normalize_url(final_link)
     return final_link,html
 
 def fetch_robot(base_url:str)->urllib.robotparser.RobotFileParser:
@@ -170,6 +180,7 @@ def index_pages(db,hostname,extracted_pages):
         state = "good"
         link = original_link
         if original_link != final_link:
+            print(original_link,final_link)
             linkcol.update_one({"url":original_link},{"$set":{"status":"redirect"}})
             link = final_link
         if html is None:
@@ -215,7 +226,7 @@ def index_pages(db,hostname,extracted_pages):
             htmlcol.update_one({"url":link},{"$set":htdoc},upsert=True)
             doc.update(get_link_doc(link,"good"))
             # todo extract links
-            print(doc)
+            print(link,doc)
             del doc["url"]
             contentcol.update_one({"url":link},{"$set":doc},upsert=True)
             for chs in doc["paragraph_checksums"]:
@@ -223,11 +234,13 @@ def index_pages(db,hostname,extracted_pages):
                     checkcol.insert_one({"_id":chs})
                 except pymongo.errors.DuplicateKeyError as err:
                     pass
-        linkcol.update_one({"url":link},{"$set":{"status":state}})
+
+        linkdoc = get_link_doc(link,state)
+        del linkdoc["url"]
+        linkcol.update_one({"url":link},{"$set":linkdoc})
 
 from bs4 import BeautifulSoup
 import urllib.parse
-import w3lib.url
 import os.path
 
 def get_bs_links(link,html):
@@ -236,7 +249,8 @@ def get_bs_links(link,html):
     base = link
     if bs.base is not None and "href" in bs.base.attrs:
         base = bs.base["href"]
-    base = urllib.parse.urlparse(w3lib.url.canonicalize_url(base))
+    base = urllib.parse.urlparse(courlan.normalize_url(base))
+
     external_links = set()
     internal_links = set()
     # Normalizacia linkov
@@ -245,12 +259,10 @@ def get_bs_links(link,html):
             continue
         href = l["href"]
         try:
-            parsed = urllib.parse.urlparse(w3lib.url.canonicalize_url(href))
+            parsed = urllib.parse.urlparse(courlan.normalize_url(href))
             netloc = parsed.netloc
             path = os.path.normpath(parsed.path)
             scheme = parsed.scheme
-            query = w3lib.url.url_query_cleaner(parsed.query,["id","aid","p","page","pid"])
-            print(parsed)
             if parsed.netloc == "":
                 scheme = base.scheme
                 if parsed.path == "/":
@@ -262,12 +274,14 @@ def get_bs_links(link,html):
                 continue
             if path.startswith("/"):
                 path = path[1:]
+            if path.endswith(")"):
+                # javascript
+                continue
             external = True
             if parsed.netloc == base.netloc:
                 external = False
             href = urllib.parse.urlunparse((scheme,netloc,path,"","",""))
-            href = w3lib.url.canonicalize_url(href)
-            print(href)
+            href = courlan.normalize_url(href)
             if external:
                 external_links.add(href)
             else:
@@ -275,7 +289,6 @@ def get_bs_links(link,html):
         except ValueError as err:
             print(err)
             pass
-    print(internal_links,external_links)
     return internal_links,external_links
 
 def extract_links(link_batch:list,responses:list,hostname:str,rules,default_status="frontlink")->list:
@@ -283,6 +296,8 @@ def extract_links(link_batch:list,responses:list,hostname:str,rules,default_stat
     badrobot = 0
     for original_link,(final_link,html) in zip(link_batch,responses):
         status = default_status
+        if html is None or len(html) < 256:
+            continue
         internal_links, external_links = get_bs_links(final_link,html)
         #external_links = courlan.extract_links(html,final_link,external_bool=True,language=LANGUAGE)
         for link in external_links:
@@ -516,19 +531,19 @@ def link_summary(db,hostname):
     print(res)
 
 def sample_links(db,hostname,status,batch_size):
-    print("Getting backlinks")
+    print("Sampling links")
     linkcol = db["links"]
     res = linkcol.find({"host":hostname,"status": {"$not":{"$in":["frontlink","backlink"]}}})
     cl = LinkClassifier()
     crawled_links = list(res)
     crawled_count = len(crawled_links)
     prediction_accuracy = 0
-    if crawled_count > 200:
+    if crawled_count > CLASSIFIER_SET_SIZE:
         # train on crawled links
         trainset,testset = split_train(crawled_links)
         cl.train(trainset)
         prediction_accuracy = cl.test(testset)
-    sample_set_size = 10000
+    sample_set_size = SAMPLE_SET_SIZE
     res = linkcol.find({"host":hostname,"status": status})
     sample_links = []
     predicted_good = 0
@@ -546,7 +561,7 @@ def sample_links(db,hostname,status,batch_size):
             continue
         for feature in features:
             visitcounter[feature] += 1
-    mls = int(min(batch_size/2,len(good_links)))
+    mls = int(min(batch_size*(1- DISCOVER_LINK_RATIO),len(good_links)))
     random.shuffle(good_links)
     links = good_links[0:mls]
     numdiscover = len(discover_links)
@@ -556,11 +571,12 @@ def sample_links(db,hostname,status,batch_size):
         prob = 0
         if features is not None:
             for feature in features:
-                prob += math.log(visitcounter[feature] / numdiscover)
+                c = visitcounter[feature]
+                prob -= math.log(c) / c
         eval_discover_links.append((link,prob))
     eval_discover_links.sort(key=lambda x: x[1],reverse=True)
-    print(eval_discover_links)
-    mls = int(min(batch_size/2,len(discover_links)))
+    #print(eval_discover_links)
+    mls = int(min(batch_size * DISCOVER_LINK_RATIO,len(eval_discover_links)))
     links += [l[0] for l in eval_discover_links[0:mls]]
     return links