/// <summary>
/// get tagged list
/// </summary>
/// <param name="inputWordList"></param>
/// <param name="isSuccessful"></param>
/// <param name="message"></param>
/// <returns></returns>
public List<Tag> GetTagList(List<Word> inputWordList, out bool isSuccessful, out string message)
{
//make Viterbi table as array
var viterbiArr = new List<ViterbiObject>[inputWordList.Count];
#region initilization
//add into first position
viterbiArr[0] = inputWordList[0].AssociatedTagDict.Keys.Select(tag3 => new ViterbiObject()
{
Tag1 = DumpTag,
Tag2 = DumpTag,
Tag3 = tag3,
Pi = Math.Log(GetProbabilitySequence3Tag(DumpTag, DumpTag, tag3))
+ Math.Log(GetProbabilityWordGivenTag(inputWordList[0], tag3))//use log to store probabilities
}).ToList();
//add into second position
if (inputWordList.Count >= 2)
{
viterbiArr[1] = (from tag2 in inputWordList[0].AssociatedTagDict.Keys
from tag3 in inputWordList[1].AssociatedTagDict.Keys
let prevViterbi = viterbiArr[0].FirstOrDefault(m => m.Tag3 == tag2)
where prevViterbi != null
select new ViterbiObject()
{
Tag1 = DumpTag,
Tag2 = tag2,
Tag3 = tag3,
Pi = prevViterbi.Pi
+ Math.Log(GetProbabilitySequence2Tag(tag2, tag3))
+ Math.Log(GetProbabilityWordGivenTag(inputWordList[1], tag3))
}).ToList();
}
#endregion
#region calculate Pi and generate viterbi table
//add from 3 to n
for (var i = 2; i < inputWordList.Count; i++)
{
var viterbiList = new List<ViterbiObject>();
foreach (var tag3 in inputWordList[i].AssociatedTagDict.Keys)
{
foreach (var tag2 in inputWordList[i - 1].AssociatedTagDict.Keys)
{
ViterbiObject maxValue = null;
foreach (var tag1 in inputWordList[i - 2].AssociatedTagDict.Keys)
{
var prevViterbi = viterbiArr[i - 1].FirstOrDefault(m => m.Tag3 == tag2 && m.Tag2 == tag1);
if (prevViterbi == null) continue;
var currentPi = prevViterbi.Pi
+ Math.Log(GetProbabilitySequence3Tag(tag1, tag2, tag3))
+ Math.Log(GetProbabilityWordGivenTag(inputWordList[i], tag3));
if (maxValue == null || maxValue.Pi < currentPi)
{
maxValue = new ViterbiObject()
{
Tag1 = tag1,
Tag2 = tag2,
Tag3 = tag3,
Pi = currentPi
};
}
}
viterbiList.Add(maxValue);
}
}
viterbiArr[i] = viterbiList;
}
#endregion
#region backtrack to find sequence tag
var tagArr = new Tag[inputWordList.Count];
//find last tag (n)th
tagArr[inputWordList.Count - 1] = viterbiArr[inputWordList.Count - 1].MaxBy(m => m.Pi).Tag3;
//find (n-1)th tag
if (inputWordList.Count > 1)
{
tagArr[inputWordList.Count - 2] = viterbiArr[inputWordList.Count - 1].MaxBy(m => m.Pi).Tag2;
}
//find (n-2)th tag
if (inputWordList.Count > 2)
{
tagArr[inputWordList.Count - 3] = viterbiArr[inputWordList.Count - 1].MaxBy(m => m.Pi).Tag1;
}
//find remaing tag
for (var i = inputWordList.Count - 4; i >= 0; i--)
{
var viterbiObj =
viterbiArr[i + 2].FirstOrDefault(m => m.Tag3 == tagArr[i + 2] && m.Tag2 == tagArr[i + 1]);
if (viterbiObj != null)
{
tagArr[i] = viterbiObj.Tag1;
}
}
#endregion
isSuccessful = true;
message = "";
return tagArr.ToList();
}
