shortcuts_provider.cc revision eb525c5499e34cc9c4b825d6d9e75bb07cc06ace
1// Copyright (c) 2012 The Chromium Authors. All rights reserved.
2// Use of this source code is governed by a BSD-style license that can be
3// found in the LICENSE file.
4
5#include "chrome/browser/autocomplete/shortcuts_provider.h"
6
7#include <algorithm>
8#include <cmath>
9#include <map>
10#include <vector>
11
12#include "base/i18n/break_iterator.h"
13#include "base/i18n/case_conversion.h"
14#include "base/logging.h"
15#include "base/metrics/histogram.h"
16#include "base/prefs/pref_service.h"
17#include "base/strings/string_number_conversions.h"
18#include "base/strings/string_util.h"
19#include "base/strings/utf_string_conversions.h"
20#include "base/time/time.h"
21#include "chrome/browser/autocomplete/autocomplete_input.h"
22#include "chrome/browser/autocomplete/autocomplete_provider_listener.h"
23#include "chrome/browser/autocomplete/autocomplete_result.h"
24#include "chrome/browser/history/history_notifications.h"
25#include "chrome/browser/history/history_service.h"
26#include "chrome/browser/history/history_service_factory.h"
27#include "chrome/browser/history/shortcuts_backend_factory.h"
28#include "chrome/browser/omnibox/omnibox_field_trial.h"
29#include "chrome/browser/profiles/profile.h"
30#include "chrome/common/pref_names.h"
31#include "chrome/common/url_constants.h"
32#include "url/url_parse.h"
33
34namespace {
35
36class RemoveMatchPredicate {
37 public:
38  explicit RemoveMatchPredicate(const std::set<GURL>& urls)
39      : urls_(urls) {
40  }
41  bool operator()(const AutocompleteMatch& match) {
42    return urls_.find(match.destination_url) != urls_.end();
43  }
44 private:
45  // Lifetime of the object is less than the lifetime of passed |urls|, so
46  // it is safe to store reference.
47  const std::set<GURL>& urls_;
48};
49
50}  // namespace
51
52ShortcutsProvider::ShortcutsProvider(AutocompleteProviderListener* listener,
53                                     Profile* profile)
54    : AutocompleteProvider(listener, profile,
55          AutocompleteProvider::TYPE_SHORTCUTS),
56      languages_(profile_->GetPrefs()->GetString(prefs::kAcceptLanguages)),
57      initialized_(false),
58      max_relevance_(AutocompleteResult::kLowestDefaultScore - 1) {
59  scoped_refptr<history::ShortcutsBackend> backend =
60      ShortcutsBackendFactory::GetForProfile(profile_);
61  if (backend.get()) {
62    backend->AddObserver(this);
63    if (backend->initialized())
64      initialized_ = true;
65  }
66  int max_relevance;
67  if (OmniboxFieldTrial::ShortcutsScoringMaxRelevance(&max_relevance))
68    max_relevance_ = max_relevance;
69}
70
71void ShortcutsProvider::Start(const AutocompleteInput& input,
72                              bool minimal_changes) {
73  matches_.clear();
74
75  if ((input.type() == AutocompleteInput::INVALID) ||
76      (input.type() == AutocompleteInput::FORCED_QUERY))
77    return;
78
79  // None of our results are applicable for best match.
80  if (input.matches_requested() == AutocompleteInput::BEST_MATCH)
81    return;
82
83  if (input.text().empty())
84    return;
85
86  if (!initialized_)
87    return;
88
89  base::TimeTicks start_time = base::TimeTicks::Now();
90  GetMatches(input);
91  if (input.text().length() < 6) {
92    base::TimeTicks end_time = base::TimeTicks::Now();
93    std::string name = "ShortcutsProvider.QueryIndexTime." +
94        base::IntToString(input.text().size());
95    base::HistogramBase* counter = base::Histogram::FactoryGet(
96        name, 1, 1000, 50, base::Histogram::kUmaTargetedHistogramFlag);
97    counter->Add(static_cast<int>((end_time - start_time).InMilliseconds()));
98  }
99  UpdateStarredStateOfMatches();
100}
101
102void ShortcutsProvider::DeleteMatch(const AutocompleteMatch& match) {
103  // Copy the URL since DeleteMatchesWithURLs() will invalidate |match|.
104  GURL url(match.destination_url);
105
106  // When a user deletes a match, he probably means for the URL to disappear out
107  // of history entirely. So nuke all shortcuts that map to this URL.
108  std::set<GURL> urls;
109  urls.insert(url);
110  // Immediately delete matches and shortcuts with the URL, so we can update the
111  // controller synchronously.
112  DeleteShortcutsWithURLs(urls);
113  DeleteMatchesWithURLs(urls);  // NOTE: |match| is now dead!
114
115  // Delete the match from the history DB. This will eventually result in a
116  // second call to DeleteShortcutsWithURLs(), which is harmless.
117  HistoryService* const history_service =
118      HistoryServiceFactory::GetForProfile(profile_, Profile::EXPLICIT_ACCESS);
119
120  DCHECK(history_service && url.is_valid());
121  history_service->DeleteURL(url);
122}
123
124ShortcutsProvider::~ShortcutsProvider() {
125  scoped_refptr<history::ShortcutsBackend> backend =
126      ShortcutsBackendFactory::GetForProfileIfExists(profile_);
127  if (backend.get())
128    backend->RemoveObserver(this);
129}
130
131void ShortcutsProvider::OnShortcutsLoaded() {
132  initialized_ = true;
133}
134
135void ShortcutsProvider::DeleteMatchesWithURLs(const std::set<GURL>& urls) {
136  std::remove_if(matches_.begin(), matches_.end(), RemoveMatchPredicate(urls));
137  listener_->OnProviderUpdate(true);
138}
139
140void ShortcutsProvider::DeleteShortcutsWithURLs(const std::set<GURL>& urls) {
141  scoped_refptr<history::ShortcutsBackend> backend =
142      ShortcutsBackendFactory::GetForProfileIfExists(profile_);
143  if (!backend.get())
144    return;  // We are off the record.
145  for (std::set<GURL>::const_iterator url = urls.begin(); url != urls.end();
146       ++url)
147    backend->DeleteShortcutsWithUrl(*url);
148}
149
150void ShortcutsProvider::GetMatches(const AutocompleteInput& input) {
151  scoped_refptr<history::ShortcutsBackend> backend =
152      ShortcutsBackendFactory::GetForProfileIfExists(profile_);
153  if (!backend.get())
154    return;
155  // Get the URLs from the shortcuts database with keys that partially or
156  // completely match the search term.
157  string16 term_string(base::i18n::ToLower(input.text()));
158  DCHECK(!term_string.empty());
159
160  for (history::ShortcutsBackend::ShortcutMap::const_iterator it =
161           FindFirstMatch(term_string, backend.get());
162       it != backend->shortcuts_map().end() &&
163           StartsWith(it->first, term_string, true); ++it) {
164    // Don't return shortcuts with zero relevance.
165    int relevance = CalculateScore(term_string, it->second);
166    if (relevance)
167      matches_.push_back(ShortcutToACMatch(relevance, term_string, it->second));
168  }
169  std::partial_sort(matches_.begin(),
170      matches_.begin() +
171          std::min(AutocompleteProvider::kMaxMatches, matches_.size()),
172      matches_.end(), &AutocompleteMatch::MoreRelevant);
173  if (matches_.size() > AutocompleteProvider::kMaxMatches) {
174    matches_.erase(matches_.begin() + AutocompleteProvider::kMaxMatches,
175                   matches_.end());
176  }
177  // Reset relevance scores to guarantee no results are given an
178  // inlineable score and all scores are decreasing (but not do assign
179  // any scores below 1).
180  int max_relevance = AutocompleteResult::kLowestDefaultScore - 1;
181  for (ACMatches::iterator it = matches_.begin(); it != matches_.end(); ++it) {
182    max_relevance = std::min(max_relevance, it->relevance);
183    it->relevance = max_relevance;
184    if (max_relevance > 1)
185      --max_relevance;
186  }
187}
188
189AutocompleteMatch ShortcutsProvider::ShortcutToACMatch(
190    int relevance,
191    const string16& term_string,
192    const history::ShortcutsBackend::Shortcut& shortcut) {
193  DCHECK(!term_string.empty());
194  AutocompleteMatch match(this, relevance, true,
195                          AutocompleteMatchType::HISTORY_TITLE);
196  match.destination_url = shortcut.url;
197  DCHECK(match.destination_url.is_valid());
198  match.fill_into_edit = UTF8ToUTF16(shortcut.url.spec());
199  match.contents = shortcut.contents;
200  match.contents_class = shortcut.contents_class;
201  match.description = shortcut.description;
202  match.description_class = shortcut.description_class;
203  match.RecordAdditionalInfo("number of hits", shortcut.number_of_hits);
204  match.RecordAdditionalInfo("last access time", shortcut.last_access_time);
205  match.RecordAdditionalInfo("original input text", UTF16ToUTF8(shortcut.text));
206
207  // Try to mark pieces of the contents and description as matches if they
208  // appear in |term_string|.
209  WordMap terms_map(CreateWordMapForString(term_string));
210  if (!terms_map.empty()) {
211    match.contents_class = ClassifyAllMatchesInString(term_string, terms_map,
212        match.contents, match.contents_class);
213    match.description_class = ClassifyAllMatchesInString(term_string, terms_map,
214        match.description, match.description_class);
215  }
216  return match;
217}
218
219// static
220ShortcutsProvider::WordMap ShortcutsProvider::CreateWordMapForString(
221    const string16& text) {
222  // First, convert |text| to a vector of the unique words in it.
223  WordMap word_map;
224  base::i18n::BreakIterator word_iter(text,
225                                      base::i18n::BreakIterator::BREAK_WORD);
226  if (!word_iter.Init())
227    return word_map;
228  std::vector<string16> words;
229  while (word_iter.Advance()) {
230    if (word_iter.IsWord())
231      words.push_back(word_iter.GetString());
232  }
233  if (words.empty())
234    return word_map;
235  std::sort(words.begin(), words.end());
236  words.erase(std::unique(words.begin(), words.end()), words.end());
237
238  // Now create a map from (first character) to (words beginning with that
239  // character).  We insert in reverse lexicographical order and rely on the
240  // multimap preserving insertion order for values with the same key.  (This
241  // is mandated in C++11, and part of that decision was based on a survey of
242  // existing implementations that found that it was already true everywhere.)
243  std::reverse(words.begin(), words.end());
244  for (std::vector<string16>::const_iterator i(words.begin()); i != words.end();
245       ++i)
246    word_map.insert(std::make_pair((*i)[0], *i));
247  return word_map;
248}
249
250// static
251ACMatchClassifications ShortcutsProvider::ClassifyAllMatchesInString(
252    const string16& find_text,
253    const WordMap& find_words,
254    const string16& text,
255    const ACMatchClassifications& original_class) {
256  DCHECK(!find_text.empty());
257  DCHECK(!find_words.empty());
258
259  // The code below assumes |text| is nonempty and therefore the resulting
260  // classification vector should always be nonempty as well.  Returning early
261  // if |text| is empty assures we'll return the (correct) empty vector rather
262  // than a vector with a single (0, NONE) match.
263  if (text.empty())
264    return original_class;
265
266  // First check whether |text| begins with |find_text| and mark that whole
267  // section as a match if so.
268  string16 text_lowercase(base::i18n::ToLower(text));
269  ACMatchClassifications match_class;
270  size_t last_position = 0;
271  if (StartsWith(text_lowercase, find_text, true)) {
272    match_class.push_back(
273        ACMatchClassification(0, ACMatchClassification::MATCH));
274    last_position = find_text.length();
275    // If |text_lowercase| is actually equal to |find_text|, we don't need to
276    // (and in fact shouldn't) put a trailing NONE classification after the end
277    // of the string.
278    if (last_position < text_lowercase.length()) {
279      match_class.push_back(
280          ACMatchClassification(last_position, ACMatchClassification::NONE));
281    }
282  } else {
283    // |match_class| should start at position 0.  If the first matching word is
284    // found at position 0, this will be popped from the vector further down.
285    match_class.push_back(
286        ACMatchClassification(0, ACMatchClassification::NONE));
287  }
288
289  // Now, starting with |last_position|, check each character in
290  // |text_lowercase| to see if we have words starting with that character in
291  // |find_words|.  If so, check each of them to see if they match the portion
292  // of |text_lowercase| beginning with |last_position|.  Accept the first
293  // matching word found (which should be the longest possible match at this
294  // location, given the construction of |find_words|) and add a MATCH region to
295  // |match_class|, moving |last_position| to be after the matching word.  If we
296  // found no matching words, move to the next character and repeat.
297  while (last_position < text_lowercase.length()) {
298    std::pair<WordMap::const_iterator, WordMap::const_iterator> range(
299        find_words.equal_range(text_lowercase[last_position]));
300    size_t next_character = last_position + 1;
301    for (WordMap::const_iterator i(range.first); i != range.second; ++i) {
302      const string16& word = i->second;
303      size_t word_end = last_position + word.length();
304      if ((word_end <= text_lowercase.length()) &&
305          !text_lowercase.compare(last_position, word.length(), word)) {
306        // Collapse adjacent ranges into one.
307        if (match_class.back().offset == last_position)
308          match_class.pop_back();
309
310        AutocompleteMatch::AddLastClassificationIfNecessary(&match_class,
311            last_position, ACMatchClassification::MATCH);
312        if (word_end < text_lowercase.length()) {
313          match_class.push_back(
314              ACMatchClassification(word_end, ACMatchClassification::NONE));
315        }
316        last_position = word_end;
317        break;
318      }
319    }
320    last_position = std::max(last_position, next_character);
321  }
322
323  return AutocompleteMatch::MergeClassifications(original_class, match_class);
324}
325
326history::ShortcutsBackend::ShortcutMap::const_iterator
327    ShortcutsProvider::FindFirstMatch(const string16& keyword,
328                                      history::ShortcutsBackend* backend) {
329  DCHECK(backend);
330  history::ShortcutsBackend::ShortcutMap::const_iterator it =
331      backend->shortcuts_map().lower_bound(keyword);
332  // Lower bound not necessarily matches the keyword, check for item pointed by
333  // the lower bound iterator to at least start with keyword.
334  return ((it == backend->shortcuts_map().end()) ||
335    StartsWith(it->first, keyword, true)) ? it :
336    backend->shortcuts_map().end();
337}
338
339int ShortcutsProvider::CalculateScore(
340    const string16& terms,
341    const history::ShortcutsBackend::Shortcut& shortcut) {
342  DCHECK(!terms.empty());
343  DCHECK_LE(terms.length(), shortcut.text.length());
344
345  // The initial score is based on how much of the shortcut the user has typed.
346  // Using the square root of the typed fraction boosts the base score rapidly
347  // as characters are typed, compared with simply using the typed fraction
348  // directly. This makes sense since the first characters typed are much more
349  // important for determining how likely it is a user wants a particular
350  // shortcut than are the remaining continued characters.
351  double base_score = max_relevance_ *
352      sqrt(static_cast<double>(terms.length()) / shortcut.text.length());
353
354  // Then we decay this by half each week.
355  const double kLn2 = 0.6931471805599453;
356  base::TimeDelta time_passed = base::Time::Now() - shortcut.last_access_time;
357  // Clamp to 0 in case time jumps backwards (e.g. due to DST).
358  double decay_exponent = std::max(0.0, kLn2 * static_cast<double>(
359      time_passed.InMicroseconds()) / base::Time::kMicrosecondsPerWeek);
360
361  // We modulate the decay factor based on how many times the shortcut has been
362  // used. Newly created shortcuts decay at full speed; otherwise, decaying by
363  // half takes |n| times as much time, where n increases by
364  // (1.0 / each 5 additional hits), up to a maximum of 5x as long.
365  const double kMaxDecaySpeedDivisor = 5.0;
366  const double kNumUsesPerDecaySpeedDivisorIncrement = 5.0;
367  double decay_divisor = std::min(kMaxDecaySpeedDivisor,
368      (shortcut.number_of_hits + kNumUsesPerDecaySpeedDivisorIncrement - 1) /
369      kNumUsesPerDecaySpeedDivisorIncrement);
370
371  return static_cast<int>((base_score / exp(decay_exponent / decay_divisor)) +
372      0.5);
373}
374