/// <summary>
/// Classifies the sentiment of the input text corpus
/// </summary>
/// <param name="contents">Text contents that needs to be classified. Works best for 1000+ words</param>
/// <param name="wordsToIgnore"></param>
/// <returns></returns>
public IDictionary <string, double> Classify(string contents, HashSet <string> wordsToIgnore)
{
if (_positiveClass == null || _negativeClass == null)
{
throw new Exception("One of the evidences were not properly defined");
}
if (string.IsNullOrWhiteSpace(_positiveClass.Name()))
{
throw new Exception("Evidence name not defined on the first one");
}
if (string.IsNullOrWhiteSpace(_negativeClass.Name()))
{
throw new Exception("Evidence name not defined on the second one");
}
var words = Tokenizer.TokenizeNow(contents).ToList();
var chunkSize = Math.Ceiling(words.Count() / (double)WORDS_IN_CHUNK);
var index = 0;
#region Classify in chunks
var scores = new List <Dictionary <string, decimal> >();
for (var i = 0; i < chunkSize; i++)
{
var score = new Dictionary <string, decimal>
{
{ _positiveClass.Name(), (decimal)0.0 },
{ _negativeClass.Name(), (decimal)0.0 }
};
scores.Add(score);
}
foreach (var wordsChunk in words.Chunk(WORDS_IN_CHUNK))
{
foreach (
var word in
wordsChunk.Where(word => !string.IsNullOrWhiteSpace(word) && !wordsToIgnore.Contains(word)))
{
//First Class
var classAEvidence = _positiveClass.GetVocabulary();
double wordCountInClassA;
wordCountInClassA = classAEvidence.TryGetValue(word, out wordCountInClassA)
? wordCountInClassA
: 1; //ToDo - Make this 1 or 0.01
var scoreClassA = (decimal)Math.Log(wordCountInClassA / _positiveClass.TotalWords());
scores[index][_positiveClass.Name()] += scoreClassA;
//Second Class
var classBEvidence = _negativeClass.GetVocabulary();
double wordCountInClassB;
wordCountInClassB = classBEvidence.TryGetValue(word, out wordCountInClassB)
? wordCountInClassB
: 1; //ToDo - Make this 1 or 0.01
var scoreClassB = (decimal)Math.Log(wordCountInClassB / _negativeClass.TotalWords());
scores[index][_negativeClass.Name()] += scoreClassB;
}
var totalWordsAllCategories = _positiveClass.TotalWords() + _negativeClass.TotalWords();
scores[index][_positiveClass.Name()] += (decimal)Math.Log(_positiveClass.TotalWords() / totalWordsAllCategories);
scores[index][_negativeClass.Name()] += (decimal)Math.Log(_negativeClass.TotalWords() / totalWordsAllCategories);
var scoreA = Math.Exp((double)scores[index][_positiveClass.Name()]);
var scoreB = Math.Exp((double)scores[index][_negativeClass.Name()]);
var totalScore = scoreA + scoreB;
try
{
scores[index][_positiveClass.Name()] = (decimal)(100 * scoreA / totalScore);
scores[index][_negativeClass.Name()] = (decimal)(100 * scoreB / totalScore);
}
catch (OverflowException overflow)
{
throw;
}
index++;
}
//Coumpute the average
var results = new Dictionary <string, double>
{
{ _positiveClass.Name(), 0.0 },
{ _negativeClass.Name(), 0.0 }
};
foreach (var score in scores)
{
results[_positiveClass.Name()] += (double)score[_positiveClass.Name()];
results[_negativeClass.Name()] += (double)score[_negativeClass.Name()];
}
results[_positiveClass.Name()] = results[_positiveClass.Name()] / scores.Count;
results[_negativeClass.Name()] = results[_negativeClass.Name()] / scores.Count;
return(results);
#endregion Classify in chunks
}
