Add calculation of derived features for distillable page model

components/dom_distiller/core/page_features.cc

+// Copyright 2015 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "components/dom_distiller/core/page_features.h"
+
+#include <string>
+
+#include "third_party/re2/re2/re2.h"
+
+namespace dom_distiller {
+/* This code needs to derive features in the same way and order in which they
+ * are derived when training the model. Parts of that code are reproduced in the
+ * comments below.
+ */
+
+namespace {
+std::string GetLastSegment(const std::string& path) {
+  // return re.search('[^/]*\/?$', path).group(0)
+  if (path.size() == 0)
+    return "";
+  size_t start = path.rfind("/", path.size() - 1);
+  return start == std::string::npos ? "" : path.substr(start + 1);
+}
+
+int CountMatches(const std::string& s, const std::string& p) {
+  // return len(re.findall(p, s))
+  re2::StringPiece sp(s);
+  re2::RE2 regexp(p);
+  int count = 0;
+  while (re2::RE2::FindAndConsume(&sp, regexp))
+    count++;
+  return count;
+}
+
+int GetWordCount(const std::string& s) {
+  return CountMatches(s, "\\w+");
+}
+
+bool Contains(const std::string& n, const std::string& h) {
+  return h.find(n) != std::string::npos;
+}
+
+bool EndsWith(const std::string& t, const std::string& s) {
+  return s.size() >= t.size() &&
+         s.compare(s.size() - t.size(), std::string::npos, t) == 0;
+}
+}
+
+std::vector<double> CalculateDerivedFeatures(bool isOGArticle,
+                                             const GURL& url,
+                                             double numElements,
+                                             double numAnchors,
+                                             double numForms,
+                                             const std::string& innerText,
+                                             const std::string& textContent,
+                                             const std::string& innerHTML) {
+  // In the training pipeline, the strings are explicitly encoded in utf-8 (as
+  // they are here).
+  const std::string& path = url.path();
+  int innerTextWords = GetWordCount(innerText);
+  int textContentWords = GetWordCount(textContent);
+  int innerHTMLWords = GetWordCount(innerHTML);
+  std::vector<double> features;
+  // 'opengraph', opengraph,
+  features.push_back(isOGArticle);
+  // 'forum', 'forum' in path,
+  features.push_back(Contains("forum", path));
+  // 'index', 'index' in path,
+  features.push_back(Contains("index", path));
+  // 'view', 'view' in path,
+  features.push_back(Contains("view", path));
+  // 'asp', '.asp' in path,
+  features.push_back(Contains(".asp", path));
+  // 'phpbb', 'phpbb' in path,
+  features.push_back(Contains(".phpbb", path));

[nyquist, 2015/04/01 01:19:14]

The contains-statement here starts with '.', but the python does not.

[cjhopman, 2015/04/01 20:12:13]

Done.

+  // 'php', path.endswith('.php'),
+  features.push_back(EndsWith(".php", path));
+  // 'pathlength', len(path),
+  features.push_back(path.size());
+  // 'domain', len(path) < 2,
+  features.push_back(path.size() < 2);
+  // 'pathcomponents', CountMatches(path, r'\/.'),
+  features.push_back(CountMatches(path, "\\/."));
+  // 'slugdetector', CountMatches(path, r'[^\w/]'),
+  features.push_back(CountMatches(path, "[^\\w/]"));
+  // 'pathnumbers', CountMatches(path, r'\d+'),
+  features.push_back(CountMatches(path, "\\d+"));
+  // 'lastSegmentLength', len(GetLastSegment(path)),
+  features.push_back(GetLastSegment(path).size());
+  // 'formcount', numForms,
+  features.push_back(numForms);
+  // 'anchorcount', numAnchors,
+  features.push_back(numAnchors);
+  // 'elementcount', numElements,
+  features.push_back(numElements);
+  // 'anchorratio', float(numAnchors) / max(1, numElements),
+  features.push_back(double(numAnchors) / std::max<double>(1, numElements));
+  // 'innertextlength', len(innerText),
+  features.push_back(innerText.size());
+  // 'textcontentlength', len(textContent),
+  features.push_back(textContent.size());
+  // 'innerhtmllength', len(innerHTML),
+  features.push_back(innerHTML.size());
+  // 'innertextlengthratio', float(len(innerText)) / max(1, len(innerHTML)),
+  features.push_back(double(innerText.size()) /
+                     std::max<double>(1.0, innerHTML.size()));
+  // 'textcontentlengthratio', float(len(textContent)) / max(1, len(innerHTML)),
+  features.push_back(double(textContent.size()) /
+                     std::max<double>(1.0, innerHTML.size()));
+  // 'innertexttextcontentlengthratio',
+  // float(len(innerText)) / max(1, len(textContent)),
+  features.push_back(double(innerText.size()) /
+                     std::max<double>(1.0, textContent.size()));
+  // 'innertextwordcount', innerTextWords,
+  features.push_back(innerTextWords);
+  // 'textcontentwordcount', textContentWords,
+  features.push_back(textContentWords);
+  // 'innerhtmlwordcount', innerHTMLWords,
+  features.push_back(innerHTMLWords);
+  // 'innertextwordcountratio', float(innerTextWords) / max(1, innerHTMLWords),
+  features.push_back(double(innerTextWords) /
+                     std::max<int>(1.0, innerHTMLWords));
+  // 'textcontentwordcountratio',
+  // float(textContentWords) / max(1, innerHTMLWords),
+  features.push_back(double(textContentWords) /
+                     std::max<int>(1.0, innerHTMLWords));
+  // 'innertexttextcontentwordcountratio',
+  // float(innerTextWords) / max(1, textContentWords),
+  features.push_back(double(innerTextWords) /
+                     std::max<int>(1.0, textContentWords));
+  return features;
+}
+
+std::vector<double> CalculateDerivedFeaturesFromJSON(const base::Value* json) {
+  const base::DictionaryValue* dict;
+  if (!json->GetAsDictionary(&dict)) {
+    return std::vector<double>();
+  }
+
+  bool isOGArticle;
+  std::string url, innerText, textContent, innerHTML;
+  double numElements, numAnchors, numForms;
+
+  if (!(dict->GetBoolean("opengraph", &isOGArticle) &&
+        dict->GetString("url", &url) &&
+        dict->GetDouble("numElements", &numElements) &&
+        dict->GetDouble("numAnchors", &numAnchors) &&
+        dict->GetDouble("numForms", &numForms) &&
+        dict->GetString("innerText", &innerText) &&
+        dict->GetString("textContent", &textContent) &&
+        dict->GetString("innerHTML", &innerHTML))) {
+    return std::vector<double>();
+  }
+
+  GURL parsed_url(url);
+  if (!parsed_url.is_valid()) {
+    return std::vector<double>();
+  }
+
+  return CalculateDerivedFeatures(isOGArticle, parsed_url, numElements,
+                                  numAnchors, numForms, innerText, textContent,
+                                  innerHTML);
+}
+}