public Lexicon1Lang(String filename) throws IOException {
captionToRawQuery = new LinkedHashMap<String, String>();
captionToExpandedQuery = new LinkedHashMap<String, String>();
List<String> lines =
Util.getLinesFromInputStream(
new FileInputStream(filename),
false /* ignore comment lines = false, we'll strip comments here */);
for (String line : lines) {
int idx = line.indexOf('#'); // strip everything after the comment char
if (idx >= 0) line = line.substring(0, idx);
line = line.trim();
if (line.length() == 0) continue; // ignore blank lines
StringTokenizer st = new StringTokenizer(line, ":");
if (st.countTokens() != 2) {
log.warn("line ignored: " + line);
continue;
}
String caption = st.nextToken().trim();
String query = st.nextToken().trim();
String existingQuery = captionToRawQuery.get(caption);
if (!Util.nullOrEmpty(existingQuery)) query = existingQuery + "|" + query;
captionToRawQuery.put(caption, query);
}
expandQueries();
}
private void expandQueries() {
captionToExpandedQuery
.clear(); // clear the expanded query map first, we don't want any residue from the
// previous state
for (String caption : captionToRawQuery.keySet()) {
String query = captionToRawQuery.get(caption);
List<String> orTerms = Util.tokenize(query, "|");
String expandedQuery = "";
for (int i = 0; i < orTerms.size(); i++) {
String t = orTerms.get(i).trim();
if (t.length() == 0) continue;
if (t.startsWith("{") && t.endsWith("}")) {
String c = t.substring(1, t.length() - 1);
String exp =
captionToExpandedQuery.get(
c); // note: expanded map, not rawmap, to allow multi-level expansion
if (exp == null) {
t = captionToRawQuery.get(c);
if (t == null) {
log.warn("ERROR: no prev. caption: " + c + " in query " + query);
continue;
}
} else t = exp;
usedInOtherCaptions.add(c);
}
expandedQuery += t;
// there is no point adding or(|), as the query is treated just as a text string and is
// not handled specially in Indexer.lookupDocsAsId
// however, adding a non-word, non-special character will enable tokenization at that
// index and will be appended as many "or" terms.
if (i < orTerms.size() - 1) expandedQuery += "|";
}
if (caption.length() > 0 && expandedQuery.length() > 0) {
// if caption already exists, just add to it
String existingQuery = captionToExpandedQuery.get(caption);
if (!Util.nullOrEmpty(existingQuery)) expandedQuery = existingQuery + "|" + expandedQuery;
captionToExpandedQuery.put(caption, expandedQuery);
}
}
// remove the non top-level captions
for (String caption : usedInOtherCaptions) captionToExpandedQuery.remove(caption);
}
/**
* main entry point: returns a category -> docs map for each (non-zero) category in the current
* captionToQueryMap.
*
* @indexer must already have run
* @docs results are restrictes to these docs. assumes all docs if docs is null or empty.
* @captions (null/none = all)
* <p>vihari This is a weird name for a method that returns documents with emotions instead
* of emotions.
*/
public Map<String, Collection<Document>> getEmotions(
Indexer indexer,
Collection<Document> docs,
boolean originalContentOnly,
String... captions) {
Map<String, Collection<Document>> result = new LinkedHashMap<String, Collection<Document>>();
Set<Document> docs_set = Util.castOrCloneAsSet(docs);
// for (String[] emotion: emotionsData)
String[] selected_captions =
captions.length > 0 ? captions : captionToExpandedQuery.keySet().toArray(new String[0]);
for (String caption : selected_captions) {
String query = captionToExpandedQuery.get(caption);
if (query == null) {
log.warn("Skipping unknown caption '" + caption + "'");
continue;
}
// query is simply word1|word2|word3 etc for that sentiment
// the -1 indicates that we want all docs in the indexer that match the query
int threshold = 1;
Indexer.QueryOptions options = new Indexer.QueryOptions();
options.setThreshold(threshold);
options.setQueryType(Indexer.QueryType.ORIGINAL);
Collection<Document> docsForCaption = indexer.docsForQuery(query, options);
/*
log.info (docsForCaption.size() + " before");
threshold = 2;
docsForCaption = indexer.docsForQuery(query, -1, threshold);
log.info (docsForCaption.size() + " after");
*/
// Set<Document> docs = indexer.docsWithPhraseThreshold(query, -1, 2); // in future, we
// may have a higher threshold for sentiment matching
// if @param docs is present, retain only those docs that match, otherwise retain all
if (!Util.nullOrEmpty(docs_set))
// docsForCaption.retainAll(docs_set);
docsForCaption = Util.listIntersection(docsForCaption, docs_set);
// put it in the result only if at least 1 doc matches
if (docsForCaption.size() > 0) result.put(caption, docsForCaption);
}
return result;
}
/**
* Generates list of questions and stores it in the current instance of MemoryStudy We handle two
* kinds of questions namely, person names tests and non-person name tests. Non-person name test
* is a fill in the blank kind where the blank is to be filled with the correct non-person entity
* to complete the sentence person name test is to guess the person in correspondent list based on
* some distinctive sentences in the mail
*
* @param maxInt - max. number of questions from a interval
* @throws IOException
*/
public void generateQuestions(
Archive archive,
NERModel nerModel,
Collection<EmailDocument> allDocs,
Lexicon lex,
int maxInt,
boolean personTest)
throws IOException, GeneralSecurityException, ClassNotFoundException, ReadContentsException,
ParseException {
this.archive = archive;
if (allDocs == null) allDocs = (Collection) archive.getAllDocs();
questions = new ArrayList<>();
ArchiveCluer cluer = new ArchiveCluer(null, archive, nerModel, null, lex);
Short[] itypes =
new Short[] {
FeatureDictionary.BUILDING,
FeatureDictionary.PLACE,
FeatureDictionary.RIVER,
FeatureDictionary.ROAD,
FeatureDictionary.UNIVERSITY,
FeatureDictionary.MOUNTAIN,
FeatureDictionary.AIRPORT,
FeatureDictionary.ISLAND,
FeatureDictionary.MUSEUM,
FeatureDictionary.BRIDGE,
FeatureDictionary.AIRLINE,
FeatureDictionary.THEATRE,
FeatureDictionary.LIBRARY,
FeatureDictionary.LAWFIRM,
FeatureDictionary.GOVAGENCY
};
double CUTOFF = 0.001;
tabooCluesSet = new LinkedHashSet<>();
archive.assignThreadIds();
List<Document> docs = archive.getAllDocs();
Map<String, Date> entityToLastDate = new LinkedHashMap<>();
Multimap<String, EmailDocument> entityToMessages = LinkedHashMultimap.create();
Multimap<String, Long> entityToThreads = LinkedHashMultimap.create();
Multimap<String, String> ceToDisplayEntity = LinkedHashMultimap.create();
int di = 0;
// sort by date
Collections.sort(docs);
Set<String> ownerNames = archive.ownerNames;
Date earliestDate = null, latestDate = null;
Set<String> allEntities = new LinkedHashSet<>();
for (Document doc : docs) {
EmailDocument ed = (EmailDocument) doc;
if (earliestDate == null || ed.date.before(earliestDate)) earliestDate = ed.date;
if (latestDate == null || ed.date.after(latestDate)) latestDate = ed.date;
List<String> entities = new ArrayList<>();
if (!personTest) {
entities.addAll(
Arrays.asList(archive.getAllNamesInDoc(doc, true))
.stream()
.filter(n -> n.typeScore > CUTOFF)
.map(n -> n.text)
.collect(Collectors.toList()));
} else {
// do not consider mailing lists
if (ed.sentToMailingLists != null && ed.sentToMailingLists.length > 0) continue;
// discard doc if it is not a sent mail
if ((ed.sentOrReceived(archive.addressBook) & EmailDocument.SENT_MASK) == 0) continue;
List<Address> addrs = new ArrayList<>();
if (ed.to != null) for (Address addr : ed.to) addrs.add(addr);
List<String> names = new ArrayList<>();
for (Address addr : addrs) {
Contact c = archive.addressBook.lookupByAddress(addr);
names.add(c.pickBestName());
}
for (String name : names) {
if (!ownerNames.contains(name) && !DictUtils.hasDictionaryWord(name)) {
entities.add(name);
}
}
}
allEntities.addAll(entities);
// get entities
for (String e : entities) {
if (Util.nullOrEmpty(e)) continue;
e = e.replaceAll("^\\W+|\\W+$", "");
if (e.length() > 10 && e.toUpperCase().equals(e))
continue; // all upper case, more than 10 letters, you're out.
String ce = DictUtils.canonicalize(e); // canonicalize
if (ce == null) {
JSPHelper.log.info("Dropping entity: " + e);
continue;
}
ceToDisplayEntity.put(ce, e);
entityToLastDate.put(ce, ed.date);
entityToMessages.put(ce, ed);
entityToThreads.put(ce, ed.threadID);
}
if ((++di) % 1000 == 0) log.info(di + " of " + docs.size() + " messages processed...<br/>");
}
log.info(
"Considered #"
+ allEntities.size()
+ " unique entities and #"
+ ceToDisplayEntity.size()
+ " good ones in #"
+ docs.size()
+ " docs<br>");
log.info("Owner Names: " + ownerNames);
JSPHelper.log.info(
"Considered #"
+ allEntities.size()
+ " unique entities and #"
+ ceToDisplayEntity.size()
+ " good ones in #"
+ docs.size()
+ "docs");
JSPHelper.log.info(
"earliest date = "
+ edu.stanford.muse.email.CalendarUtil.formatDateForDisplay(earliestDate));
JSPHelper.log.info(
"latest date = " + edu.stanford.muse.email.CalendarUtil.formatDateForDisplay(latestDate));
Multimap<String, String> tokenToCE = LinkedHashMultimap.create();
for (String ce : ceToDisplayEntity.keySet()) {
List<String> tokens = Util.tokenize(ce);
for (String t : tokens) tokenToCE.put(t, ce);
}
// Compute date intervals
int DAYS_PER_INTERVAL = 30;
List<Pair<Date, Date>> intervals = new ArrayList<Pair<Date, Date>>();
{
JSPHelper.log.info("computing time intervals");
Date closingDate = latestDate;
JSPHelper.log.info(
"closing = " + edu.stanford.muse.email.CalendarUtil.formatDateForDisplay(closingDate));
while (earliestDate.before(closingDate)) {
Calendar cal = new GregorianCalendar();
cal.setTime(closingDate); // this is the time of the last sighting of the term
// scroll to the beginning of this month
cal.set(Calendar.HOUR_OF_DAY, 23);
cal.set(Calendar.MINUTE, 59);
cal.set(Calendar.SECOND, 59);
Date endDate = cal.getTime();
cal.add(
Calendar.DATE,
(1
- DAYS_PER_INTERVAL)); // 1- because we want from 0:00 of first date to 23:59 of
// last date
cal.set(Calendar.HOUR_OF_DAY, 0);
cal.set(Calendar.MINUTE, 0);
cal.set(Calendar.SECOND, 0);
Date startDate = cal.getTime();
intervals.add(new Pair<Date, Date>(startDate, endDate));
// ok we got an interval
// closing date for the next interval is 1 day before endDate
cal.add(Calendar.DATE, -1);
closingDate = cal.getTime();
}
JSPHelper.log.info("done computing intervals, #time intervals: " + intervals.size());
for (Pair<Date, Date> p : intervals)
JSPHelper.log.info(
"Interval: "
+ edu.stanford.muse.email.CalendarUtil.formatDateForDisplay(p.getFirst())
+ " - "
+ edu.stanford.muse.email.CalendarUtil.formatDateForDisplay(p.getSecond()));
}
// initialize clueInfos to empty lists
List<ClueInfo> clueInfos[] = new ArrayList[intervals.size()];
for (int i = 0; i < intervals.size(); i++) {
clueInfos[i] = new ArrayList<ClueInfo>();
}
Map<Integer, Integer> intervalCount = new LinkedHashMap<>();
// nSent is the number of sentences allowed in a clue text
int nvalidclues = 0, nSent = 2;
// generate clueInfos for each entity
for (String ce : entityToLastDate.keySet()) {
Date lastSeenDate = entityToLastDate.get(ce);
// compute displayEntity (which has red for core words) and fullAnswer, which is a simple
// string
String fullAnswer = "";
{
List<String> tokens = Util.tokenize(ceToDisplayEntity.get(ce).iterator().next());
for (String t : tokens) {
if (EnglishDictionary.stopWords.contains(t.toLowerCase())) continue;
fullAnswer += t + " ";
}
fullAnswer = fullAnswer.trim();
}
// dont want the answer to be scored low just because it has extra non-word chars in the begin
// or end
fullAnswer = fullAnswer.replaceAll("^\\W+|\\W+$", "");
// which interval does this date belong to?
int interval = -1;
Date intervalStart = null, intervalEnd = null;
{
int i = 0;
for (Pair<Date, Date> p : intervals) {
intervalStart = p.getFirst();
intervalEnd = p.getSecond();
if ((intervalStart.before(lastSeenDate) && intervalEnd.after(lastSeenDate))
|| intervalStart.equals(lastSeenDate)
|| intervalEnd.equals(lastSeenDate)) {
interval = i;
break;
}
i++;
}
}
if (interval < 0 || interval == intervals.size())
JSPHelper.log.info(
"What, no interval!? for "
+ edu.stanford.muse.email.CalendarUtil.formatDateForDisplay(lastSeenDate));
if (!intervalCount.containsKey(interval)) intervalCount.put(interval, 0);
if (intervalCount.get(interval) > maxInt) continue;
intervalCount.put(interval, intervalCount.get(interval) + 1);
List<Integer> lengthList = Crossword.convertToWord(fullAnswer).getSecond();
String lengthDescr = "";
if (lengthList.size() > 1) lengthDescr += Integer.toString(lengthList.size()) + " words: ";
for (Integer i : lengthList) {
lengthDescr += Util.pluralize(i, "letter") + ", ";
}
lengthDescr = lengthDescr.substring(0, lengthDescr.length() - 2); // subtract the extra comma.
ClueInfo ci = new ClueInfo();
ci.link = "../browse?term=\"" + fullAnswer + "\"&sort_by=recent&searchType=original";
ci.lastSeenDate = lastSeenDate;
ci.nMessages = entityToMessages.get(ce).size();
ci.nThreads = entityToThreads.get(ce).size();
// TODO: we are doing default initialisation of evaluators by setting it to null below, it is
// more appropriate to consider it as an argument for this method
Clue clue =
cluer.createClue(
fullAnswer,
(personTest
? ArchiveCluer.QuestionType.GUESS_CORRESPONDENT
: ArchiveCluer.QuestionType.FILL_IN_THE_BLANK),
null,
tabooCluesSet,
null,
intervalStart,
intervalEnd,
nSent,
archive);
if (clue != null) ci.clues = new Clue[] {clue};
if (ci.clues == null || ci.clues.length == 0 || clue == null) {
JSPHelper.log.warn("Did not find any clue for: " + fullAnswer);
} else {
// is the times value of the clue important?
questions.add(new MemoryQuestion(this, fullAnswer, clue, 1, lengthDescr));
nvalidclues++;
// makes sure that the clue with the same statement is not generated again
tabooCluesSet.add(clue.clue);
}
clueInfos[interval].add(ci);
}
log.info("Found valid clues for " + nvalidclues + " answers");
JSPHelper.log.info("Found valid clues for " + nvalidclues + " answers");
log.info("Top candidates are:");
for (MemoryQuestion mq : questions)
log.info(mq.correctAnswer + " times=" + mq.stats.nMessagesWithAnswer);
// sort q's by clue score
Collections.sort(questions);
// log.info("Based on clue score, top answers:");
// for (MemoryQuestion mq: questions)
// log.info (mq.correctAnswer + " times= clue=" + mq.clue.clue);
// now we have up to 2*N questions, sorted by cluescore.
// drop ones that are prefix/suffix of another, and cap to N
int prev_size = questions.size();
int new_size = questions.size();
// log.info ("#questions before prefix-suffix elim: " + prev_size + " after: " + new_size);
int count = 0;
for (MemoryQuestion mq : questions) {
mq.setQuestionNum(count++);
}
// log the questions as well, just in case we don't get to the final point due to user fatigue
// or crashes
logStats("questions.final", false);
}