/*
* Read conversations from file
*/
private void Read_conversations(string target_file_path)
{
this.convsList = new ArrayList();
int numberofTweets = 0;
int word_count_a_tweet = 0;
string[] line_arr = null;
SDTM_v1_Conversation one_conv = null;
SDTM_v1_Tweet one_tweet = null;
SDTM_v1_Word one_word = null;
Dictionary <SDTM_v1_Word, int> word_count = null;
string one_word_count = null;
string[] one_word_count_arr = null;
int conv_idx = 0;
try
{
using (StreamReader sr = new StreamReader(target_file_path))
{
string line = null;
while ((line = sr.ReadLine()) != null)
{
// Conversation name, userid1_userid2_convid
one_conv = new SDTM_v1_Conversation(conv_idx);
conv_idx++;
line_arr = line.Split('_');
one_conv.set_users(Convert.ToInt32(line_arr[0]), Convert.ToInt32(line_arr[1]));
// number of tweets in the conversation
line = sr.ReadLine();
numberofTweets = Convert.ToInt32(line);
// Each tweet in a conversation
for (int tweet_idx = 0; tweet_idx < numberofTweets; tweet_idx++)
{
// Line format is
// user_id lambda_0 lambda_1 numberofuniquewords BagofWordsFormat
line = sr.ReadLine();
line_arr = line.Split(' ');
one_tweet = new SDTM_v1_Tweet();
one_tweet.set_tweet_id(Convert.ToInt32(line_arr[0]));
one_tweet.set_max_ent_prob(Convert.ToDouble(line_arr[1]), Convert.ToDouble(line_arr[2]));
word_count = new Dictionary <SDTM_v1_Word, int>();
//word_count_a_tweet = Convert.ToInt32(line_arr[3]) + 4; // At the end of line_arr
word_count_a_tweet = line_arr.Length;
for (int word_idx = 4; word_idx < word_count_a_tweet; word_idx++)
{
one_word_count = line_arr[word_idx];
one_word_count_arr = one_word_count.Split(':');
one_word = new SDTM_v1_Word(Convert.ToInt32(one_word_count_arr[0]));
word_count.Add(one_word, Convert.ToInt32(one_word_count_arr[1]));
}
one_tweet.set_word_count(word_count);
// insert tweet to conversation
one_conv.insert_tweet(one_tweet);
}
// insert conversation to list
this.convsList.Add(one_conv);
}
}
}
catch (Exception e)
{
Console.WriteLine(e);
Environment.Exit(1);
}
this.numConvs = this.convsList.Count;
}
/*
Read conversations from file
*/
private void Read_conversations(string target_file_path)
{
this.convsList = new ArrayList();
int numberofTweets = 0;
int word_count_a_tweet = 0;
string[] line_arr = null;
SDTM_v1_Conversation one_conv = null;
SDTM_v1_Tweet one_tweet = null;
SDTM_v1_Word one_word = null;
Dictionary<SDTM_v1_Word, int> word_count = null;
string one_word_count = null;
string[] one_word_count_arr = null;
int conv_idx = 0;
try
{
using (StreamReader sr = new StreamReader(target_file_path))
{
string line = null;
while ((line = sr.ReadLine()) != null)
{
// Conversation name, userid1_userid2_convid
one_conv = new SDTM_v1_Conversation(conv_idx);
conv_idx++;
line_arr = line.Split('_');
one_conv.set_users(Convert.ToInt32(line_arr[0]), Convert.ToInt32(line_arr[1]));
// number of tweets in the conversation
line = sr.ReadLine();
numberofTweets = Convert.ToInt32(line);
// Each tweet in a conversation
for (int tweet_idx = 0; tweet_idx < numberofTweets; tweet_idx++)
{
// Line format is
// user_id lambda_0 lambda_1 numberofuniquewords BagofWordsFormat
line = sr.ReadLine();
line_arr = line.Split(' ');
one_tweet = new SDTM_v1_Tweet();
one_tweet.set_tweet_id(Convert.ToInt32(line_arr[0]));
one_tweet.set_max_ent_prob(Convert.ToDouble(line_arr[1]), Convert.ToDouble(line_arr[2]));
word_count = new Dictionary<SDTM_v1_Word, int>();
//word_count_a_tweet = Convert.ToInt32(line_arr[3]) + 4; // At the end of line_arr
word_count_a_tweet = line_arr.Length;
for (int word_idx = 4; word_idx < word_count_a_tweet; word_idx++)
{
one_word_count = line_arr[word_idx];
one_word_count_arr = one_word_count.Split(':');
one_word = new SDTM_v1_Word(Convert.ToInt32(one_word_count_arr[0]));
word_count.Add(one_word, Convert.ToInt32(one_word_count_arr[1]));
}
one_tweet.set_word_count(word_count);
// insert tweet to conversation
one_conv.insert_tweet(one_tweet);
}
// insert conversation to list
this.convsList.Add(one_conv);
}
}
}
catch (Exception e)
{
Console.WriteLine(e);
Environment.Exit(1);
}
this.numConvs = this.convsList.Count;
}
private void GibbsSampling_Each_conv(SDTM_v1_Conversation one_conv)
{
// Preparing
int oldLevel = 0;
int oldTopic = 0;
int newLevel = 0;
int newTopic = 0;
int numTopics_target_level = 0;
double prob_part_senti_value = 0.0;
double target_sumBeta = 0.0;
double beta0, m0, expectLTW, beta, betaw;
double prob_table_temp = 0.0;
double sumProb = 0.0;
// Each tweet
foreach (SDTM_v1_Tweet one_tweet in one_conv.tweet_list)
{
sumProb = 0.0;
// Decrease current one_tweet value
oldLevel = one_tweet.sd_level;
oldTopic = one_tweet.topic;
foreach (KeyValuePair <SDTM_v1_Word, int> one_entry in one_tweet.word_count_table)
{
this.matrixLTW[oldLevel, oldTopic, one_entry.Key.wordidx]--;
}
this.sumLTW[oldLevel, oldTopic] -= one_tweet.word_count_table.Count;
one_conv.CLT[oldLevel, oldTopic]--;
one_conv.sumCLT[oldLevel]--;
// Fill probability table
// Level 0
numTopics_target_level = numTopics_arr[0];
prob_part_senti_value = one_tweet.max_ent_prob[0] / (one_conv.sumCLT[0] + this.sumAlpha[0]);
target_sumBeta = this.sumBeta[0];
for (int ti = 0; ti < numTopics_target_level; ti++)
{
beta0 = this.sumLTW[0, ti] + target_sumBeta;
m0 = 0;
expectLTW = 1.0;
foreach (KeyValuePair <SDTM_v1_Word, int> one_entry in one_tweet.word_count_table)
{
if (-1 == one_entry.Key.seed_word_level)
{
beta = this.betas[0];
}
else
{
beta = this.betas[2];
}
betaw = this.matrixLTW[0, ti, one_entry.Key.wordidx] + beta;
for (int m = 0; m < (int)one_entry.Value; m++)
{
expectLTW *= (betaw + m) / (beta0 + m0);
m0++;
}
}
prob_table_temp = (one_conv.CLT[0, ti] + this.alpha)
* prob_part_senti_value
* expectLTW;
sumProb += prob_table_temp;
this.probTable[ti, 0] = prob_table_temp;
}
// Level 1 and 2
for (int level_idx = 1; level_idx < SDTM_v1.numLevels; level_idx++)
{
numTopics_target_level = numTopics_arr[level_idx];
prob_part_senti_value = one_tweet.max_ent_prob[1] / (one_conv.sumCLT[level_idx] + this.sumAlpha[level_idx]);
target_sumBeta = this.sumBeta[level_idx];
for (int ti = 0; ti < numTopics_target_level; ti++)
{
beta0 = this.sumLTW[level_idx, ti] + target_sumBeta;
m0 = 0;
expectLTW = 1.0;
foreach (KeyValuePair <SDTM_v1_Word, int> one_entry in one_tweet.word_count_table)
{
if (-1 == one_entry.Key.seed_word_level)
{
beta = this.betas[0];
}
else if (level_idx == one_entry.Key.seed_word_level)
{
beta = this.betas[1];
}
else
{
beta = this.betas[2];
}
betaw = this.matrixLTW[level_idx, ti, one_entry.Key.wordidx] + beta;
for (int m = 0; m < (int)one_entry.Value; m++)
{
expectLTW *= (betaw + m) / (beta0 + m0);
m0++;
}
}
prob_table_temp = (one_conv.CLT[level_idx, ti] + this.alpha)
* prob_part_senti_value
* expectLTW;
sumProb += prob_table_temp;
this.probTable[ti, level_idx] = prob_table_temp;
}
}
// Multinomial sampling
Multinomial_sampling(sumProb, out newLevel, out newTopic);
// Assign and increase with new value
one_tweet.set_sd_level(newLevel);
one_tweet.set_topic(newTopic);
foreach (KeyValuePair <SDTM_v1_Word, int> one_entry in one_tweet.word_count_table)
{
this.matrixLTW[newLevel, newTopic, one_entry.Key.wordidx]++;
}
this.sumLTW[newLevel, newTopic] += one_tweet.word_count_table.Count;
one_conv.CLT[newLevel, newTopic]++;
one_conv.sumCLT[newLevel]++;
}
}
private void GibbsSampling_Each_conv(SDTM_v1_Conversation one_conv)
{
// Preparing
int oldLevel = 0;
int oldTopic = 0;
int newLevel = 0;
int newTopic = 0;
int numTopics_target_level = 0;
double prob_part_senti_value = 0.0;
double target_sumBeta = 0.0;
double beta0, m0, expectLTW, beta, betaw;
double prob_table_temp = 0.0;
double sumProb = 0.0;
// Each tweet
foreach (SDTM_v1_Tweet one_tweet in one_conv.tweet_list)
{
sumProb = 0.0;
// Decrease current one_tweet value
oldLevel = one_tweet.sd_level;
oldTopic = one_tweet.topic;
foreach (KeyValuePair<SDTM_v1_Word, int> one_entry in one_tweet.word_count_table)
{
this.matrixLTW[oldLevel, oldTopic, one_entry.Key.wordidx]--;
}
this.sumLTW[oldLevel, oldTopic] -= one_tweet.word_count_table.Count;
one_conv.CLT[oldLevel, oldTopic]--;
one_conv.sumCLT[oldLevel]--;
// Fill probability table
// Level 0
numTopics_target_level = numTopics_arr[0];
prob_part_senti_value = one_tweet.max_ent_prob[0] / (one_conv.sumCLT[0] + this.sumAlpha[0]);
target_sumBeta = this.sumBeta[0];
for (int ti = 0; ti < numTopics_target_level; ti++)
{
beta0 = this.sumLTW[0, ti] + target_sumBeta;
m0 = 0;
expectLTW = 1.0;
foreach (KeyValuePair<SDTM_v1_Word, int> one_entry in one_tweet.word_count_table)
{
if (-1 == one_entry.Key.seed_word_level)
{
beta = this.betas[0];
}
else
{
beta = this.betas[2];
}
betaw = this.matrixLTW[0, ti, one_entry.Key.wordidx] + beta;
for (int m = 0; m < (int) one_entry.Value; m++)
{
expectLTW *= (betaw + m) / (beta0 + m0);
m0++;
}
}
prob_table_temp = (one_conv.CLT[0, ti] + this.alpha)
* prob_part_senti_value
* expectLTW;
sumProb += prob_table_temp;
this.probTable[ti, 0] = prob_table_temp;
}
// Level 1 and 2
for (int level_idx = 1; level_idx < SDTM_v1.numLevels; level_idx++)
{
numTopics_target_level = numTopics_arr[level_idx];
prob_part_senti_value = one_tweet.max_ent_prob[1] / (one_conv.sumCLT[level_idx] + this.sumAlpha[level_idx]);
target_sumBeta = this.sumBeta[level_idx];
for (int ti = 0; ti < numTopics_target_level; ti++)
{
beta0 = this.sumLTW[level_idx, ti] + target_sumBeta;
m0 = 0;
expectLTW = 1.0;
foreach (KeyValuePair<SDTM_v1_Word, int> one_entry in one_tweet.word_count_table)
{
if (-1 == one_entry.Key.seed_word_level)
{
beta = this.betas[0];
}
else if (level_idx == one_entry.Key.seed_word_level)
{
beta = this.betas[1];
}
else
{
beta = this.betas[2];
}
betaw = this.matrixLTW[level_idx, ti, one_entry.Key.wordidx] + beta;
for (int m = 0; m < (int)one_entry.Value; m++)
{
expectLTW *= (betaw + m) / (beta0 + m0);
m0++;
}
}
prob_table_temp = (one_conv.CLT[level_idx, ti] + this.alpha)
* prob_part_senti_value
* expectLTW;
sumProb += prob_table_temp;
this.probTable[ti, level_idx] = prob_table_temp;
}
}
// Multinomial sampling
Multinomial_sampling(sumProb, out newLevel, out newTopic);
// Assign and increase with new value
one_tweet.set_sd_level(newLevel);
one_tweet.set_topic(newTopic);
foreach (KeyValuePair<SDTM_v1_Word, int> one_entry in one_tweet.word_count_table)
{
this.matrixLTW[newLevel, newTopic, one_entry.Key.wordidx]++;
}
this.sumLTW[newLevel, newTopic] += one_tweet.word_count_table.Count;
one_conv.CLT[newLevel, newTopic]++;
one_conv.sumCLT[newLevel]++;
}
}
