diff --git a/.gitignore b/.gitignore
index 3d20afb1..0b9c4f9f 100644
--- a/.gitignore
+++ b/.gitignore
@@ -40,3 +40,4 @@ nosetests.xml
 .project
 .pydevproject
 venv
+Pipfile*
diff --git a/newspaper/extractors.py b/newspaper/extractors.py
index 96255401..ed5964cc 100644
--- a/newspaper/extractors.py
+++ b/newspaper/extractors.py
@@ -820,13 +820,22 @@ def calculate_best_node(self, doc):
             if parent_node not in parent_nodes:
                 parent_nodes.append(parent_node)
 
-            # Parent of parent node
-            parent_parent_node = self.parser.getParent(parent_node)
-            if parent_parent_node is not None:
-                self.update_node_count(parent_parent_node, 1)
-                self.update_score(parent_parent_node, upscore / 2)
-                if parent_parent_node not in parent_nodes:
-                    parent_nodes.append(parent_parent_node)
+            # Issue 645, 776 - Some text was left out due to "deep" nodes
+            # containing content not being accounted for.  This should result
+            # in the "heaviest" content subtree as the "top node."
+            #
+            # Percolate weights up through the tree.  Not really sure
+            # what the upscore is, but seems that it should decay, so
+            # decaying linearly.
+            ancestor_node = self.parser.getParent(parent_node)
+            j = 2
+            while ancestor_node is not None:
+                self.update_score(ancestor_node, upscore / (float(j)))
+                if ancestor_node not in parent_nodes:
+                    parent_nodes.append(ancestor_node)
+                ancestor_node = self.parser.getParent(ancestor_node)
+                j += 1
+
             cnt += 1
             i += 1
 
diff --git a/newspaper/parsers.py b/newspaper/parsers.py
index a4d030c0..1c554a8b 100644
--- a/newspaper/parsers.py
+++ b/newspaper/parsers.py
@@ -7,8 +7,11 @@
 or query an lxml or soup dom object generated from an article's html.
 """
 import logging
+import traceback
+
 import lxml.etree
 import lxml.html
+import lxml.html.soupparser
 import lxml.html.clean
 import re
 from html import unescape
@@ -65,10 +68,17 @@ def fromstring(cls, html):
             # lxml does not play well with <? ?> encoding tags
             if html.startswith('<?'):
                 html = re.sub(r'^\<\?.*?\?\>', '', html, flags=re.DOTALL)
-            cls.doc = lxml.html.fromstring(html)
+            root = lxml.html.fromstring(html)
+            try:
+                ignore = lxml.html.tostring(root, encoding='unicode')
+                cls.doc = root
+            except UnicodeDecodeError:
+                cls.doc = lxml.html.soupparser.fromstring(html, None, None, features='html5lib')
             return cls.doc
-        except Exception:
-            log.warn('fromstring() returned an invalid string: %s...', html[:20])
+        except Exception as e:
+            tb = traceback.format_exc()
+            log.warn('fromstring() returned an invalid string (%s): %s...', str(e), html[:20])
+            log.warn(tb)
             return
 
     @classmethod
diff --git a/requirements.txt b/requirements.txt
index 61974601..b225d10b 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -11,4 +11,5 @@ python-dateutil>=2.5.3
 PyYAML>=3.11
 requests>=2.10.0
 tinysegmenter==0.3  # TODO(codelucas): Investigate making this >=0.3
-tldextract>=2.0.1
\ No newline at end of file
+tldextract>=2.0.1
+html5lib>=1.1