/// <summary>
/// Gets synset shells from a word index line. A synset shell is an instance of SynSet with only the POS and Offset
/// members initialized. These members are enough to look up the full synset within the corresponding data file. This
/// method is static to prevent inadvertent references to a current WordNetEngine, which should be passed via the
/// corresponding parameter.
/// </summary>
/// <param name="wordIndexLine">Word index line from which to get synset shells</param>
/// <param name="pos">POS of the given index line</param>
/// <param name="mostCommonSynSet">Returns the most common synset for the word</param>
/// <param name="wordNetEngine">WordNetEngine to pass to the constructor of each synset shell</param>
/// <returns>Synset shells for the given index line</returns>
private static Set <SynSet> GetSynSetShells(string wordIndexLine, POS pos, out SynSet mostCommonSynSet, WordNetEngine wordNetEngine)
{
Set <SynSet> synsets = new Set <SynSet>();
mostCommonSynSet = null;
// get number of synsets
string[] parts = wordIndexLine.Split(new char[] { ' ' }, StringSplitOptions.RemoveEmptyEntries);
int numSynSets = int.Parse(parts[2]);
// grab each synset shell, from last to first
int firstOffsetIndex = parts.Length - numSynSets;
for (int i = parts.Length - 1; i >= firstOffsetIndex; --i)
{
// create synset
int offset = int.Parse(parts[i]);
// add synset to collection
SynSet synset = new SynSet(pos, offset, wordNetEngine);
synsets.Add(synset);
// if this is the last synset offset to get (since we grabbed them in reverse order), record it as the most common synset
if (i == firstOffsetIndex)
{
mostCommonSynSet = synset;
}
}
if (mostCommonSynSet == null)
{
throw new Exception("Failed to get most common synset");
}
return(synsets);
}
/// <summary>
/// Gets the shortest path from the current synset to another, following the given synset relations.
/// </summary>
/// <param name="destination">Destination synset</param>
/// <param name="relations">Relations to follow, or null for all relations.</param>
/// <returns>Synset path, or null if none exists.</returns>
public List <SynSet> GetShortestPathTo(SynSet destination, IEnumerable <WordNetEngine.SynSetRelation> relations)
{
if (relations == null)
{
relations = Enum.GetValues(typeof(WordNetEngine.SynSetRelation)) as WordNetEngine.SynSetRelation[];
}
// make sure the backpointer on the current synset is null - can't predict what other functions might do
_searchBackPointer = null;
// avoid cycles
Set <SynSet> synsetsEncountered = new Set <SynSet>();
synsetsEncountered.Add(this);
// start search queue
Queue <SynSet> searchQueue = new Queue <SynSet>();
searchQueue.Enqueue(this);
// run search
List <SynSet> path = null;
while (searchQueue.Count > 0 && path == null)
{
SynSet currSynSet = searchQueue.Dequeue();
// see if we've finished the search
if (currSynSet == destination)
{
// gather synsets along path
path = new List <SynSet>();
while (currSynSet != null)
{
path.Add(currSynSet);
currSynSet = currSynSet.SearchBackPointer;
}
// reverse for the correct order
path.Reverse();
}
// expand the search one level
else
{
foreach (SynSet synset in currSynSet.GetRelatedSynSets(relations, false))
{
if (!synsetsEncountered.Contains(synset))
{
synset.SearchBackPointer = currSynSet;
searchQueue.Enqueue(synset);
synsetsEncountered.Add(synset);
}
}
}
}
// null-out all search backpointers
foreach (SynSet synset in synsetsEncountered)
{
synset.SearchBackPointer = null;
}
return(path);
}
/// <summary>
/// Instantiates the current synset. If idSynset is non-null, related synsets references are set to those from
/// idSynset; otherwise, related synsets are created as shells.
/// </summary>
/// <param name="definition">Definition line of synset from data file</param>
/// <param name="idSynset">Lookup for related synsets. If null, all related synsets will be created as shells.</param>
internal void Instantiate(string definition, Dictionary <string, SynSet> idSynset)
{
// don't re-instantiate
if (_instantiated)
{
throw new Exception("Synset has already been instantiated");
}
/* get lexicographer file name...the enumeration lines up precisely with the wordnet spec (see the lexnames file) except that
* it starts with None, so we need to add 1 to the definition line's value to get the correct file name */
int lexicographerFileNumber = int.Parse(GetField(definition, 1)) + 1;
if (lexicographerFileNumber <= 0)
{
throw new Exception("Invalid lexicographer file name number. Should be >= 1.");
}
_lexicographerFileName = (WordNetEngine.LexicographerFileName)lexicographerFileNumber;
// get number of words in the synset and the start character of the word list
int wordStart;
int numWords = int.Parse(GetField(definition, 3, out wordStart), NumberStyles.HexNumber);
wordStart = definition.IndexOf(' ', wordStart) + 1;
// get words in synset
_words = new List <string>(numWords);
for (int i = 0; i < numWords; ++i)
{
int wordEnd = definition.IndexOf(' ', wordStart + 1) - 1;
int wordLen = wordEnd - wordStart + 1;
string word = definition.Substring(wordStart, wordLen);
if (word.Contains(' '))
{
throw new Exception("Unexpected space in word: " + word);
}
_words.Add(word);
// skip lex_id field
wordStart = definition.IndexOf(' ', wordEnd + 2) + 1;
}
// get gloss
_gloss = definition.Substring(definition.IndexOf('|') + 1).Trim();
if (_gloss.Contains('|'))
{
throw new Exception("Unexpected pipe in gloss");
}
// get number and start of relations
int relationCountField = 3 + (_words.Count * 2) + 1;
int relationFieldStart;
int numRelations = int.Parse(GetField(definition, relationCountField, out relationFieldStart));
relationFieldStart = definition.IndexOf(' ', relationFieldStart) + 1;
// grab each related synset
_relationSynSets = new Dictionary <WordNetEngine.SynSetRelation, Set <SynSet> >();
_lexicalRelations = new Dictionary <WordNetEngine.SynSetRelation, Dictionary <SynSet, Dictionary <int, Set <int> > > >();
for (int relationNum = 0; relationNum < numRelations; ++relationNum)
{
string relationSymbol = null;
int relatedSynSetOffset = -1;
WordNetEngine.POS relatedSynSetPOS = WordNetEngine.POS.None;
int sourceWordIndex = -1;
int targetWordIndex = -1;
// each relation has four columns
for (int relationField = 0; relationField <= 3; ++relationField)
{
int fieldEnd = definition.IndexOf(' ', relationFieldStart + 1) - 1;
int fieldLen = fieldEnd - relationFieldStart + 1;
string fieldValue = definition.Substring(relationFieldStart, fieldLen);
// relation symbol
if (relationField == 0)
{
relationSymbol = fieldValue;
}
// related synset offset
else if (relationField == 1)
{
relatedSynSetOffset = int.Parse(fieldValue);
}
// related synset POS
else if (relationField == 2)
{
relatedSynSetPOS = GetPOS(fieldValue);
}
// source/target word for lexical relation
else if (relationField == 3)
{
sourceWordIndex = int.Parse(fieldValue.Substring(0, 2), NumberStyles.HexNumber);
targetWordIndex = int.Parse(fieldValue.Substring(2), NumberStyles.HexNumber);
}
else
{
throw new Exception();
}
relationFieldStart = definition.IndexOf(' ', relationFieldStart + 1) + 1;
}
// get related synset...create shell if we don't have a lookup
SynSet relatedSynSet;
if (idSynset == null)
{
relatedSynSet = new SynSet(relatedSynSetPOS, relatedSynSetOffset, _wordNetEngine);
}
// look up related synset directly
else
{
relatedSynSet = idSynset[relatedSynSetPOS + ":" + relatedSynSetOffset];
}
// get relation
WordNetEngine.SynSetRelation relation = WordNetEngine.GetSynSetRelation(_pos, relationSymbol);
// add semantic relation if we have neither a source nor a target word index
if (sourceWordIndex == 0 && targetWordIndex == 0)
{
_relationSynSets.EnsureContainsKey(relation, typeof(Set <SynSet>));
_relationSynSets[relation].Add(relatedSynSet);
}
// add lexical relation
else
{
_lexicalRelations.EnsureContainsKey(relation, typeof(Dictionary <SynSet, Dictionary <int, Set <int> > >));
_lexicalRelations[relation].EnsureContainsKey(relatedSynSet, typeof(Dictionary <int, Set <int> >));
_lexicalRelations[relation][relatedSynSet].EnsureContainsKey(sourceWordIndex, typeof(Set <int>));
if (!_lexicalRelations[relation][relatedSynSet][sourceWordIndex].Contains(targetWordIndex))
{
_lexicalRelations[relation][relatedSynSet][sourceWordIndex].Add(targetWordIndex);
}
}
}
// release the wordnet engine if we have one...don't need it anymore
if (_wordNetEngine != null)
{
_wordNetEngine = null;
}
_instantiated = true;
}
/// <summary>
/// Gets a synset
/// </summary>
/// <param name="synsetID">ID of synset in the format returned by SynSet.ID (i.e., POS:Offset)</param>
/// <returns>SynSet</returns>
public SynSet GetSynSet(string synsetID)
{
SynSet synset;
if (_inMemory)
synset = _idSynset[synsetID];
else
{
// get POS and offset
int colonLoc = synsetID.IndexOf(':');
POS pos = (POS)Enum.Parse(typeof(POS), synsetID.Substring(0, colonLoc));
int offset = int.Parse(synsetID.Substring(colonLoc + 1));
// create shell and then instantiate
synset = new SynSet(pos, offset, this);
synset.Instantiate();
}
return synset;
}
/// <summary>
/// Constructor
/// </summary>
/// <param name="wordNetDirectory">Path to WorNet directory (the one with the data and index files in it)</param>
/// <param name="inMemory">Whether or not to store all data in memory. In-memory storage requires quite a bit of space
/// but it is also very quick. The alternative (false) will cause the data to be searched on-disk with an efficient
/// binary search algorithm.</param>
public WordNetEngine(string wordNetDirectory, bool inMemory)
{
_wordNetDirectory = wordNetDirectory;
_inMemory = inMemory;
_posIndexWordSearchStream = null;
_posSynSetDataFile = null;
if (!System.IO.Directory.Exists(_wordNetDirectory))
throw new DirectoryNotFoundException("Отсутствует WordNet директория: " + _wordNetDirectory);
// get data and index paths
string[] dataPaths = new string[]
{
Path.Combine(_wordNetDirectory, "data.adj"),
Path.Combine(_wordNetDirectory, "data.adv"),
Path.Combine(_wordNetDirectory, "data.noun"),
Path.Combine(_wordNetDirectory, "data.verb")
};
string[] indexPaths = new string[]
{
Path.Combine(_wordNetDirectory, "index.adj"),
Path.Combine(_wordNetDirectory, "index.adv"),
Path.Combine(_wordNetDirectory, "index.noun"),
Path.Combine(_wordNetDirectory, "index.verb")
};
// make sure all files exist
foreach (string path in dataPaths.Union(indexPaths))
if (!System.IO.File.Exists(path))
throw new FileNotFoundException("Failed to find WordNet file: " + path);
#region index file sorting
string sortFlagPath = Path.Combine(_wordNetDirectory, ".sorted_for_dot_net");
if (!System.IO.File.Exists(sortFlagPath))
{
/* make sure the index files are sorted according to the current sort order. the index files in the
* wordnet distribution are sorted in the order needed for (presumably) the java api, which uses
* a different sort order than the .net runtime. thus, unless we resort the lines in the index
* files, we won't be able to do a proper binary search over the data. */
foreach (string indexPath in indexPaths)
{
// create temporary file for sorted lines
string tempPath = Path.GetTempFileName();
StreamWriter tempFile = new StreamWriter(tempPath);
// get number of words (lines) in file
int numWords = 0;
StreamReader indexFile = new StreamReader(indexPath);
string line;
while (indexFile.TryReadLine(out line))
if (!line.StartsWith(" "))
++numWords;
// get lines in file, sorted by first column (i.e., the word)
Dictionary<string, string> wordLine = new Dictionary<string, string>(numWords);
indexFile = new StreamReader(indexPath);
while (indexFile.TryReadLine(out line))
// write header lines to temp file immediately
if (line.StartsWith(" "))
tempFile.WriteLine(line);
else
{
// trim useless blank spaces from line and map line to first column
line = line.Trim();
wordLine.Add(line.Substring(0, line.IndexOf(' ')), line);
}
// get sorted words
List<string> sortedWords = new List<string>(wordLine.Count);
sortedWords.AddRange(wordLine.Keys);
sortedWords.Sort();
// write lines sorted by word
foreach (string word in sortedWords)
tempFile.WriteLine(wordLine[word]);
tempFile.Close();
// replace original index file with properly sorted one
System.IO.File.Delete(indexPath);
System.IO.File.Move(tempPath, indexPath);
}
// create flag file, indicating that we've sorted the data
StreamWriter sortFlagFile = new StreamWriter(sortFlagPath);
sortFlagFile.WriteLine("This file serves no purpose other than to indicate that the WordNet distribution data in the current directory has been sorted for use by the .NET API.");
sortFlagFile.Close();
}
#endregion
#region engine init
if (inMemory)
{
// pass 1: get total number of synsets
int totalSynsets = 0;
foreach (string dataPath in dataPaths)
{
// scan synset data file for lines that don't start with a space...these are synset definition lines
StreamReader dataFile = new StreamReader(dataPath);
string line;
while (dataFile.TryReadLine(out line))
{
int firstSpace = line.IndexOf(' ');
if (firstSpace > 0)
++totalSynsets;
}
}
// pass 2: create synset shells (pos and offset only)
_idSynset = new Dictionary<string, SynSet>(totalSynsets);
foreach (string dataPath in dataPaths)
{
POS pos = GetFilePOS(dataPath);
// scan synset data file
StreamReader dataFile = new StreamReader(dataPath);
string line;
while (dataFile.TryReadLine(out line))
{
int firstSpace = line.IndexOf(' ');
if (firstSpace > 0)
{
// get offset and create synset shell
int offset = int.Parse(line.Substring(0, firstSpace));
SynSet synset = new SynSet(pos, offset, null);
_idSynset.Add(synset.ID, synset);
}
}
}
// pass 3: instantiate synsets (hooks up relations, set glosses, etc.)
foreach (string dataPath in dataPaths)
{
POS pos = GetFilePOS(dataPath);
// scan synset data file
StreamReader dataFile = new StreamReader(dataPath);
string line;
while (dataFile.TryReadLine(out line))
{
int firstSpace = line.IndexOf(' ');
if (firstSpace > 0)
// instantiate synset defined on current line, using the instantiated synsets for all references
_idSynset[pos + ":" + int.Parse(line.Substring(0, firstSpace))].Instantiate(line, _idSynset);
}
}
// organize synsets by pos and words...also set most common synset for word-pos pairs that have multiple synsets
_posWordSynSets = new Dictionary<POS, Dictionary<string, Set<SynSet>>>();
foreach (string indexPath in indexPaths)
{
POS pos = GetFilePOS(indexPath);
_posWordSynSets.EnsureContainsKey(pos, typeof(Dictionary<string, Set<SynSet>>));
// scan word index file, skipping header lines
StreamReader indexFile = new StreamReader(indexPath);
string line;
while (indexFile.TryReadLine(out line))
{
int firstSpace = line.IndexOf(' ');
if (firstSpace > 0)
{
// grab word and synset shells, along with the most common synset
string word = line.Substring(0, firstSpace);
SynSet mostCommonSynSet;
Set<SynSet> synsets = GetSynSetShells(line, pos, out mostCommonSynSet, null);
// set flag on most common synset if it's ambiguous
if (synsets.Count > 1)
_idSynset[mostCommonSynSet.ID].SetAsMostCommonSynsetFor(word);
// use reference to the synsets that we instantiated in our three-pass routine above
_posWordSynSets[pos].Add(word, new Set<SynSet>(synsets.Count));
foreach (SynSet synset in synsets)
_posWordSynSets[pos][word].Add(_idSynset[synset.ID]);
}
}
}
}
else
{
// open binary search streams for index files
_posIndexWordSearchStream = new Dictionary<POS, BinarySearchTextStream>();
foreach (string indexPath in indexPaths)
{
// create binary search stream for index file
BinarySearchTextStream searchStream = new BinarySearchTextStream(indexPath, new BinarySearchTextStream.SearchComparisonDelegate(delegate(object searchWord, string currentLine)
{
// if we landed on the header text, search further down
if (currentLine[0] == ' ')
return 1;
// get word on current line
string currentWord = currentLine.Substring(0, currentLine.IndexOf(' '));
// compare searched-for word to the current word
return ((string)searchWord).CompareTo(currentWord);
}));
// add search stream for current POS
_posIndexWordSearchStream.Add(GetFilePOS(indexPath), searchStream);
}
// open readers for synset data files
_posSynSetDataFile = new Dictionary<POS, StreamReader>();
foreach (string dataPath in dataPaths)
_posSynSetDataFile.Add(GetFilePOS(dataPath), new StreamReader(dataPath));
}
#endregion
}
/// <summary>
/// Gets synset shells from a word index line. A synset shell is an instance of SynSet with only the POS and Offset
/// members initialized. These members are enough to look up the full synset within the corresponding data file. This
/// method is static to prevent inadvertent references to a current WordNetEngine, which should be passed via the
/// corresponding parameter.
/// </summary>
/// <param name="wordIndexLine">Word index line from which to get synset shells</param>
/// <param name="pos">POS of the given index line</param>
/// <param name="mostCommonSynSet">Returns the most common synset for the word</param>
/// <param name="wordNetEngine">WordNetEngine to pass to the constructor of each synset shell</param>
/// <returns>Synset shells for the given index line</returns>
private static Set<SynSet> GetSynSetShells(string wordIndexLine, POS pos, out SynSet mostCommonSynSet, WordNetEngine wordNetEngine)
{
Set<SynSet> synsets = new Set<SynSet>();
mostCommonSynSet = null;
// get number of synsets
string[] parts = wordIndexLine.Split(new char[] { ' ' }, StringSplitOptions.RemoveEmptyEntries);
int numSynSets = int.Parse(parts[2]);
// grab each synset shell, from last to first
int firstOffsetIndex = parts.Length - numSynSets;
for (int i = parts.Length - 1; i >= firstOffsetIndex; --i)
{
// create synset
int offset = int.Parse(parts[i]);
// add synset to collection
SynSet synset = new SynSet(pos, offset, wordNetEngine);
synsets.Add(synset);
// if this is the last synset offset to get (since we grabbed them in reverse order), record it as the most common synset
if (i == firstOffsetIndex)
mostCommonSynSet = synset;
}
if (mostCommonSynSet == null)
throw new Exception("Failed to get most common synset");
return synsets;
}
/// <summary>
/// Constructor
/// </summary>
/// <param name="wordNetDirectory">Path to WorNet directory (the one with the data and index files in it)</param>
/// <param name="inMemory">Whether or not to store all data in memory. In-memory storage requires quite a bit of space
/// but it is also very quick. The alternative (false) will cause the data to be searched on-disk with an efficient
/// binary search algorithm.</param>
public WordNetEngine(string wordNetDirectory, bool inMemory)
{
_wordNetDirectory = wordNetDirectory;
_inMemory = inMemory;
_posIndexWordSearchStream = null;
_posSynSetDataFile = null;
if (!System.IO.Directory.Exists(_wordNetDirectory))
{
throw new DirectoryNotFoundException("Отсутствует WordNet директория: " + _wordNetDirectory);
}
// get data and index paths
string[] dataPaths = new string[]
{
Path.Combine(_wordNetDirectory, "data.adj"),
Path.Combine(_wordNetDirectory, "data.adv"),
Path.Combine(_wordNetDirectory, "data.noun"),
Path.Combine(_wordNetDirectory, "data.verb")
};
string[] indexPaths = new string[]
{
Path.Combine(_wordNetDirectory, "index.adj"),
Path.Combine(_wordNetDirectory, "index.adv"),
Path.Combine(_wordNetDirectory, "index.noun"),
Path.Combine(_wordNetDirectory, "index.verb")
};
// make sure all files exist
foreach (string path in dataPaths.Union(indexPaths))
{
if (!System.IO.File.Exists(path))
{
throw new FileNotFoundException("Failed to find WordNet file: " + path);
}
}
#region index file sorting
string sortFlagPath = Path.Combine(_wordNetDirectory, ".sorted_for_dot_net");
if (!System.IO.File.Exists(sortFlagPath))
{
/* make sure the index files are sorted according to the current sort order. the index files in the
* wordnet distribution are sorted in the order needed for (presumably) the java api, which uses
* a different sort order than the .net runtime. thus, unless we resort the lines in the index
* files, we won't be able to do a proper binary search over the data. */
foreach (string indexPath in indexPaths)
{
// create temporary file for sorted lines
string tempPath = Path.GetTempFileName();
StreamWriter tempFile = new StreamWriter(tempPath);
// get number of words (lines) in file
int numWords = 0;
StreamReader indexFile = new StreamReader(indexPath);
string line;
while (indexFile.TryReadLine(out line))
{
if (!line.StartsWith(" "))
{
++numWords;
}
}
// get lines in file, sorted by first column (i.e., the word)
Dictionary <string, string> wordLine = new Dictionary <string, string>(numWords);
indexFile = new StreamReader(indexPath);
while (indexFile.TryReadLine(out line))
{
// write header lines to temp file immediately
if (line.StartsWith(" "))
{
tempFile.WriteLine(line);
}
else
{
// trim useless blank spaces from line and map line to first column
line = line.Trim();
wordLine.Add(line.Substring(0, line.IndexOf(' ')), line);
}
}
// get sorted words
List <string> sortedWords = new List <string>(wordLine.Count);
sortedWords.AddRange(wordLine.Keys);
sortedWords.Sort();
// write lines sorted by word
foreach (string word in sortedWords)
{
tempFile.WriteLine(wordLine[word]);
}
tempFile.Close();
// replace original index file with properly sorted one
System.IO.File.Delete(indexPath);
System.IO.File.Move(tempPath, indexPath);
}
// create flag file, indicating that we've sorted the data
StreamWriter sortFlagFile = new StreamWriter(sortFlagPath);
sortFlagFile.WriteLine("This file serves no purpose other than to indicate that the WordNet distribution data in the current directory has been sorted for use by the .NET API.");
sortFlagFile.Close();
}
#endregion
#region engine init
if (inMemory)
{
// pass 1: get total number of synsets
int totalSynsets = 0;
foreach (string dataPath in dataPaths)
{
// scan synset data file for lines that don't start with a space...these are synset definition lines
StreamReader dataFile = new StreamReader(dataPath);
string line;
while (dataFile.TryReadLine(out line))
{
int firstSpace = line.IndexOf(' ');
if (firstSpace > 0)
{
++totalSynsets;
}
}
}
// pass 2: create synset shells (pos and offset only)
_idSynset = new Dictionary <string, SynSet>(totalSynsets);
foreach (string dataPath in dataPaths)
{
POS pos = GetFilePOS(dataPath);
// scan synset data file
StreamReader dataFile = new StreamReader(dataPath);
string line;
while (dataFile.TryReadLine(out line))
{
int firstSpace = line.IndexOf(' ');
if (firstSpace > 0)
{
// get offset and create synset shell
int offset = int.Parse(line.Substring(0, firstSpace));
SynSet synset = new SynSet(pos, offset, null);
_idSynset.Add(synset.ID, synset);
}
}
}
// pass 3: instantiate synsets (hooks up relations, set glosses, etc.)
foreach (string dataPath in dataPaths)
{
POS pos = GetFilePOS(dataPath);
// scan synset data file
StreamReader dataFile = new StreamReader(dataPath);
string line;
while (dataFile.TryReadLine(out line))
{
int firstSpace = line.IndexOf(' ');
if (firstSpace > 0)
{
// instantiate synset defined on current line, using the instantiated synsets for all references
_idSynset[pos + ":" + int.Parse(line.Substring(0, firstSpace))].Instantiate(line, _idSynset);
}
}
}
// organize synsets by pos and words...also set most common synset for word-pos pairs that have multiple synsets
_posWordSynSets = new Dictionary <POS, Dictionary <string, Set <SynSet> > >();
foreach (string indexPath in indexPaths)
{
POS pos = GetFilePOS(indexPath);
_posWordSynSets.EnsureContainsKey(pos, typeof(Dictionary <string, Set <SynSet> >));
// scan word index file, skipping header lines
StreamReader indexFile = new StreamReader(indexPath);
string line;
while (indexFile.TryReadLine(out line))
{
int firstSpace = line.IndexOf(' ');
if (firstSpace > 0)
{
// grab word and synset shells, along with the most common synset
string word = line.Substring(0, firstSpace);
SynSet mostCommonSynSet;
Set <SynSet> synsets = GetSynSetShells(line, pos, out mostCommonSynSet, null);
// set flag on most common synset if it's ambiguous
if (synsets.Count > 1)
{
_idSynset[mostCommonSynSet.ID].SetAsMostCommonSynsetFor(word);
}
// use reference to the synsets that we instantiated in our three-pass routine above
_posWordSynSets[pos].Add(word, new Set <SynSet>(synsets.Count));
foreach (SynSet synset in synsets)
{
_posWordSynSets[pos][word].Add(_idSynset[synset.ID]);
}
}
}
}
}
else
{
// open binary search streams for index files
_posIndexWordSearchStream = new Dictionary <POS, BinarySearchTextStream>();
foreach (string indexPath in indexPaths)
{
// create binary search stream for index file
BinarySearchTextStream searchStream = new BinarySearchTextStream(indexPath, new BinarySearchTextStream.SearchComparisonDelegate(delegate(object searchWord, string currentLine)
{
// if we landed on the header text, search further down
if (currentLine[0] == ' ')
{
return(1);
}
// get word on current line
string currentWord = currentLine.Substring(0, currentLine.IndexOf(' '));
// compare searched-for word to the current word
return(((string)searchWord).CompareTo(currentWord));
}));
// add search stream for current POS
_posIndexWordSearchStream.Add(GetFilePOS(indexPath), searchStream);
}
// open readers for synset data files
_posSynSetDataFile = new Dictionary <POS, StreamReader>();
foreach (string dataPath in dataPaths)
{
_posSynSetDataFile.Add(GetFilePOS(dataPath), new StreamReader(dataPath));
}
}
#endregion
}
/// <summary>
/// Gets the closest synset that is reachable from the current and another synset along the given relations. For example,
/// given two synsets and the Hypernym relation, this will return the lowest synset that is a hypernym of both synsets. If
/// the hypernym hierarchy forms a tree, this will be the lowest common ancestor.
/// </summary>
/// <param name="synset">Other synset</param>
/// <param name="relations">Relations to follow</param>
/// <returns>Closest mutually reachable synset</returns>
public SynSet GetClosestMutuallyReachableSynset(SynSet synset, IEnumerable<WordNetEngine.SynSetRelation> relations)
{
// avoid cycles
Set<SynSet> synsetsEncountered = new Set<SynSet>();
synsetsEncountered.Add(this);
// start search queue
Queue<SynSet> searchQueue = new Queue<SynSet>();
searchQueue.Enqueue(this);
// run search
SynSet closest = null;
while (searchQueue.Count > 0 && closest == null)
{
SynSet currSynSet = searchQueue.Dequeue();
/* check for a path between the given synset and the current one. if such a path exists, the current
* synset is the closest mutually reachable synset. */
if (synset.GetShortestPathTo(currSynSet, relations) != null)
closest = currSynSet;
// otherwise, expand the search along the given relations
else
foreach (SynSet relatedSynset in currSynSet.GetRelatedSynSets(relations, false))
if (!synsetsEncountered.Contains(relatedSynset))
{
searchQueue.Enqueue(relatedSynset);
synsetsEncountered.Add(relatedSynset);
}
}
return closest;
}
/// <summary>
/// Gets the shortest path from the current synset to another, following the given synset relations.
/// </summary>
/// <param name="destination">Destination synset</param>
/// <param name="relations">Relations to follow, or null for all relations.</param>
/// <returns>Synset path, or null if none exists.</returns>
public List<SynSet> GetShortestPathTo(SynSet destination, IEnumerable<WordNetEngine.SynSetRelation> relations)
{
if (relations == null)
relations = Enum.GetValues(typeof(WordNetEngine.SynSetRelation)) as WordNetEngine.SynSetRelation[];
// make sure the backpointer on the current synset is null - can't predict what other functions might do
_searchBackPointer = null;
// avoid cycles
Set<SynSet> synsetsEncountered = new Set<SynSet>();
synsetsEncountered.Add(this);
// start search queue
Queue<SynSet> searchQueue = new Queue<SynSet>();
searchQueue.Enqueue(this);
// run search
List<SynSet> path = null;
while (searchQueue.Count > 0 && path == null)
{
SynSet currSynSet = searchQueue.Dequeue();
// see if we've finished the search
if (currSynSet == destination)
{
// gather synsets along path
path = new List<SynSet>();
while (currSynSet != null)
{
path.Add(currSynSet);
currSynSet = currSynSet.SearchBackPointer;
}
// reverse for the correct order
path.Reverse();
}
// expand the search one level
else
foreach (SynSet synset in currSynSet.GetRelatedSynSets(relations, false))
if (!synsetsEncountered.Contains(synset))
{
synset.SearchBackPointer = currSynSet;
searchQueue.Enqueue(synset);
synsetsEncountered.Add(synset);
}
}
// null-out all search backpointers
foreach (SynSet synset in synsetsEncountered)
synset.SearchBackPointer = null;
return path;
}
/// <summary>
/// Instantiates the current synset. If idSynset is non-null, related synsets references are set to those from
/// idSynset; otherwise, related synsets are created as shells.
/// </summary>
/// <param name="definition">Definition line of synset from data file</param>
/// <param name="idSynset">Lookup for related synsets. If null, all related synsets will be created as shells.</param>
internal void Instantiate(string definition, Dictionary<string, SynSet> idSynset)
{
// don't re-instantiate
if (_instantiated)
throw new Exception("Synset has already been instantiated");
/* get lexicographer file name...the enumeration lines up precisely with the wordnet spec (see the lexnames file) except that
* it starts with None, so we need to add 1 to the definition line's value to get the correct file name */
int lexicographerFileNumber = int.Parse(GetField(definition, 1)) + 1;
if (lexicographerFileNumber <= 0)
throw new Exception("Invalid lexicographer file name number. Should be >= 1.");
_lexicographerFileName = (WordNetEngine.LexicographerFileName)lexicographerFileNumber;
// get number of words in the synset and the start character of the word list
int wordStart;
int numWords = int.Parse(GetField(definition, 3, out wordStart), NumberStyles.HexNumber);
wordStart = definition.IndexOf(' ', wordStart) + 1;
// get words in synset
_words = new List<string>(numWords);
for (int i = 0; i < numWords; ++i)
{
int wordEnd = definition.IndexOf(' ', wordStart + 1) - 1;
int wordLen = wordEnd - wordStart + 1;
string word = definition.Substring(wordStart, wordLen);
if (word.Contains(' '))
throw new Exception("Unexpected space in word: " + word);
_words.Add(word);
// skip lex_id field
wordStart = definition.IndexOf(' ', wordEnd + 2) + 1;
}
// get gloss
_gloss = definition.Substring(definition.IndexOf('|') + 1).Trim();
if (_gloss.Contains('|'))
throw new Exception("Unexpected pipe in gloss");
// get number and start of relations
int relationCountField = 3 + (_words.Count * 2) + 1;
int relationFieldStart;
int numRelations = int.Parse(GetField(definition, relationCountField, out relationFieldStart));
relationFieldStart = definition.IndexOf(' ', relationFieldStart) + 1;
// grab each related synset
_relationSynSets = new Dictionary<WordNetEngine.SynSetRelation, Set<SynSet>>();
_lexicalRelations = new Dictionary<WordNetEngine.SynSetRelation, Dictionary<SynSet, Dictionary<int, Set<int>>>>();
for (int relationNum = 0; relationNum < numRelations; ++relationNum)
{
string relationSymbol = null;
int relatedSynSetOffset = -1;
WordNetEngine.POS relatedSynSetPOS = WordNetEngine.POS.None;
int sourceWordIndex = -1;
int targetWordIndex = -1;
// each relation has four columns
for (int relationField = 0; relationField <= 3; ++relationField)
{
int fieldEnd = definition.IndexOf(' ', relationFieldStart + 1) - 1;
int fieldLen = fieldEnd - relationFieldStart + 1;
string fieldValue = definition.Substring(relationFieldStart, fieldLen);
// relation symbol
if (relationField == 0)
relationSymbol = fieldValue;
// related synset offset
else if (relationField == 1)
relatedSynSetOffset = int.Parse(fieldValue);
// related synset POS
else if (relationField == 2)
relatedSynSetPOS = GetPOS(fieldValue);
// source/target word for lexical relation
else if (relationField == 3)
{
sourceWordIndex = int.Parse(fieldValue.Substring(0, 2), NumberStyles.HexNumber);
targetWordIndex = int.Parse(fieldValue.Substring(2), NumberStyles.HexNumber);
}
else
throw new Exception();
relationFieldStart = definition.IndexOf(' ', relationFieldStart + 1) + 1;
}
// get related synset...create shell if we don't have a lookup
SynSet relatedSynSet;
if (idSynset == null)
relatedSynSet = new SynSet(relatedSynSetPOS, relatedSynSetOffset, _wordNetEngine);
// look up related synset directly
else
relatedSynSet = idSynset[relatedSynSetPOS + ":" + relatedSynSetOffset];
// get relation
WordNetEngine.SynSetRelation relation = WordNetEngine.GetSynSetRelation(_pos, relationSymbol);
// add semantic relation if we have neither a source nor a target word index
if (sourceWordIndex == 0 && targetWordIndex == 0)
{
_relationSynSets.EnsureContainsKey(relation, typeof(Set<SynSet>));
_relationSynSets[relation].Add(relatedSynSet);
}
// add lexical relation
else
{
_lexicalRelations.EnsureContainsKey(relation, typeof(Dictionary<SynSet, Dictionary<int, Set<int>>>));
_lexicalRelations[relation].EnsureContainsKey(relatedSynSet, typeof(Dictionary<int, Set<int>>));
_lexicalRelations[relation][relatedSynSet].EnsureContainsKey(sourceWordIndex, typeof(Set<int>));
if (!_lexicalRelations[relation][relatedSynSet][sourceWordIndex].Contains(targetWordIndex))
_lexicalRelations[relation][relatedSynSet][sourceWordIndex].Add(targetWordIndex);
}
}
// release the wordnet engine if we have one...don't need it anymore
if (_wordNetEngine != null)
_wordNetEngine = null;
_instantiated = true;
}
/// <summary>
/// Gets similarity of two strings using the most common synset for given string/pos pairs
/// </summary>
/// <param name="string1">First string</param>
/// <param name="pos1">First POS</param>
/// <param name="pos2">Second POS</param>
/// <param name="string2">Second string</param>
/// <param name="strategy">Similarity strategy to use</param>
/// <param name="relations">Relations to use when computing similarity</param>
/// <returns>Similarity</returns>
public float GetSimilarity(string string1, WordNetEngine.POS pos1, string string2, WordNetEngine.POS pos2, Strategy strategy, params WordNetEngine.SynSetRelation[] relations)
{
float similarity = 0;
if (strategy == Strategy.WuPalmer1994Average)
{
// get average similarity across all synsets
int numScores = 0;
foreach (SynSet synset1 in _wordNetEngine.GetSynSets(string1, pos1))
{
foreach (SynSet synset2 in _wordNetEngine.GetSynSets(string2, pos2))
{
similarity += GetSimilarity(synset1, synset2, strategy, relations);
++numScores;
}
}
if (numScores > 0)
{
similarity = similarity / (float)numScores;
}
}
else if (strategy == Strategy.WuPalmer1994Maximum)
{
// get maximum similarity across all synsets
foreach (SynSet synset1 in _wordNetEngine.GetSynSets(string1, pos1))
{
foreach (SynSet synset2 in _wordNetEngine.GetSynSets(string2, pos2))
{
float currSim = GetSimilarity(synset1, synset2, strategy, relations);
if (currSim > similarity)
{
similarity = currSim;
}
}
}
}
else if (strategy == Strategy.WuPalmer1994Minimum)
{
// get minimum similarity across all synsets
similarity = -1;
foreach (SynSet synset1 in _wordNetEngine.GetSynSets(string1, pos1))
{
foreach (SynSet synset2 in _wordNetEngine.GetSynSets(string2, pos2))
{
float currSim = GetSimilarity(synset1, synset2, strategy, relations);
if (similarity == -1 || currSim < similarity)
{
similarity = currSim;
}
}
}
// if we didn't find any synsets, similarity is zero
if (similarity == -1)
{
similarity = 0;
}
}
else if (strategy == Strategy.WuPalmer1994MostCommon)
{
// use most common synsets
SynSet synset1 = _wordNetEngine.GetMostCommonSynSet(string1, pos1);
SynSet synset2 = _wordNetEngine.GetMostCommonSynSet(string2, pos2);
if (synset1 != null && synset2 != null)
{
similarity = GetSimilarity(synset1, synset2, strategy, relations);
}
}
else
{
throw new NotImplementedException("Unimplemented strategy: " + strategy);
}
if (similarity < 0 || similarity > 1)
{
throw new Exception("Invalid similarity: " + similarity);
}
return(similarity);
}
private void UpdateWordWeight(string word, string stemmedWord, SynSet requiredSynset)
{
foreach (var synword in requiredSynset.Words)
{
string stemmedSynword = string.Empty;
//delete underscore symbols from synword if it consists of several words and not stemm it
stemmedSynword = !synword.Contains("_") ? Stemm(synword) : synword.Replace("_", " ");
if (stemmedSynword == stemmedWord)
continue;
if (ReferringManager.Instance.OriginalText.ToLower().Contains(stemmedSynword))
{
//add weight to word from synword
var usingCount = CalculateUsingCount(synword);
if (usingCount > 0)
{
AddWordWeight(word, usingCount);
}
}
}
}
/// <summary>
/// Gets similarity of two synsets
/// </summary>
/// <param name="synset1">First synset</param>
/// <param name="synset2">Second synset</param>
/// <param name="strategy">Strategy to use. All strategies named WuPalmer1994* will produce the same result since only two synsets
/// are available.</param>
/// <param name="relations">Synset relations to follow when computing similarity (pass null for all relations)</param>
/// <returns>Similarity</returns>
public float GetSimilarity(SynSet synset1, SynSet synset2, Strategy strategy, params WordNetEngine.SynSetRelation[] relations)
{
if (relations == null)
relations = Enum.GetValues(typeof(WordNetEngine.SynSetRelation)).Cast<WordNetEngine.SynSetRelation>().ToArray();
float similarity = 0;
if (strategy.ToString().StartsWith("WuPalmer1994"))
{
// get the LCS along the similarity relations
SynSet lcs = synset1.GetClosestMutuallyReachableSynset(synset2, relations);
if (lcs == null)
similarity = 0;
else
{
// get depth of synsets
int lcsDepth = lcs.GetDepth(relations) + 1;
int synset1Depth = synset1.GetShortestPathTo(lcs, relations).Count - 1 + lcsDepth;
int synset2Depth = synset2.GetShortestPathTo(lcs, relations).Count - 1 + lcsDepth;
// get similarity
similarity = 2 * lcsDepth / (float)(synset1Depth + synset2Depth);
}
}
else
throw new NotImplementedException("Unrecognized strategy");
if (similarity < 0 || similarity > 1)
throw new Exception("Invalid similarity");
return similarity;
}