/// <summary>
/// reference:
/// https://stanfordnlp.github.io/CoreNLP/api.html
/// 生成情景
/// 解析每组中的sentence dep
/// </summary>
/// <param name="manager"></param>
public List <Scenario> GenerateScenarios(ScenarioTimeManager manager)
{
string id = "S0";
List <Scenario> scenarios = new List <Scenario>();
manager.Group.Keys.ToList().ForEach(key =>
{
//each key has a scenario depict
List <edu.stanford.nlp.util.CoreMap> group = manager.Group[key];
Scenario scenario = new Scenario(_net, id);
//逐句子处理，拼凑情景元素值
group.ForEach(sentence => ElementExtractByDependencyPrase(sentence, scenario));
scenarios.Add(scenario);
//更新事件链id
id += "1";
});
return(scenarios);
}
/// <summary>
/// 分别转换情景成词向量
/// </summary>
/// <param name="manager"></param>
/// <returns></returns>
public List <double[]> ToScenarioWordVector(ScenarioTimeManager manager)
{
List <double[]> list = new List <double[]>();
manager.Group.Keys.ToList().ForEach(key =>
{
//each key has a scenario depict
List <edu.stanford.nlp.util.CoreMap> group = manager.Group[key];
group.ForEach(sentence =>
{
java.util.AbstractList tokens = sentence.get(tokensAnnotationClass) as java.util.AbstractList;
foreach (edu.stanford.nlp.ling.CoreLabel lable in tokens)
{
string word = lable.value();
double[] vt = _net.ToDouble(word);
if (vt != null)
{
list.Add(vt);
}
}
});
});
return(list);
}
/// <summary>
/// 执行annotate操作，分析此段话的时间区间
/// 归纳事件的时间序列
/// </summary>
/// <param name="rawText">待处理文本</param>
/// <param name="beginDateTime">起始日期, 默认使用系统当前时间，建议设置为事件发生日期时间</param>
public ScenarioTimeManager InductiveEventTimeSeries(string rawText, string beginDateTime = null)
{
if (rawText == null || rawText.Length == 0)
{
return(null);
}
//annotate text
edu.stanford.nlp.pipeline.StanfordCoreNLPClient pipeline = new edu.stanford.nlp.pipeline.StanfordCoreNLPClient(_props, NLPConfiguration.CoreNLPAddress, Convert.ToInt32(NLPConfiguration.CoreNLPPort));
edu.stanford.nlp.pipeline.Annotation document = new edu.stanford.nlp.pipeline.Annotation(rawText);
//date format and set for reference
string formateDate = beginDateTime != null?Convert.ToDateTime(beginDateTime).ToString("yyyy-MM-dd") : DateTime.Now.ToString("yyyy-MM-dd");
document.set(docDateAnnotationClass, formateDate);
//annotate timex
pipeline.annotate(document);
java.util.AbstractList sentences = document.get(sentencesAnnotationClass) as java.util.AbstractList;
if (sentences == null)
{
return(null);
}
//create scenario manager (timeline, scenario, info, etc)
ScenarioTimeManager stManager = new ScenarioTimeManager(Convert.ToDateTime(formateDate), rawText);
//1. 分析时间序列
foreach (edu.stanford.nlp.util.CoreMap sentence in sentences)
{
//}{debug 展示句子内容
string text = (string)sentence.get(textAnnotationClass);
//edu.stanford.nlp.util, edu.stanford.nlp.coref.data.Mention
var mentions = sentence.get(mentionsAnnotationClass) as java.util.AbstractList;
//从mentions entites里找到EntityTypeAnnotation
foreach (edu.stanford.nlp.util.CoreMap anno in mentions)
{
string entityType = (string)anno.get(entityTypeAnnotation);
//reference : https://nlp.stanford.edu/pubs/lrec2012-sutime.pdf
if (entityType == "DATE") //date without time
{
//hashmap
java.util.HashMap probHash = anno.get(namedEntityTagProbsAnnotation) as java.util.HashMap;
//extract information
double prob = Convert.ToDouble(probHash.get(entityType).ToString()); //java.lang.Double -> string -> double
if (prob < CONFIDENCE)
{
continue;
}
int offset = (anno.get(tokenBeginAnnotation) as java.lang.Integer).intValue(); //begin offset
string normalizedTimex = (string)anno.get(normalizedNamedEntityTagAnnotationClass);
if (!normalizedTimex.Contains("T")) //不包含TIME的DATE可能是综述性时间，需要加入记录
{
DateTime dtime = normalizedTimex.ToDateTime();
stManager.AddTimeStamp(dtime, offset); //置信度标注
}
}
else if (entityType == "TIME") // a time point indicating a particular instance on a time scale
{
//hashmap
java.util.HashMap probHash = anno.get(namedEntityTagProbsAnnotation) as java.util.HashMap;
//extract information
double prob = Convert.ToDouble(probHash.get(entityType).ToString()); //java.lang.Double -> string -> double
if (prob < CONFIDENCE)
{
continue;
}
int offset = (anno.get(tokenBeginAnnotation) as java.lang.Integer).intValue(); //begin offset
string normalizedTimex = (string)anno.get(normalizedNamedEntityTagAnnotationClass);
DateTime dtime = normalizedTimex.TimeExpression().ToDateTime();
stManager.AddTimeStamp(dtime, offset); //置信度标注
}
#region Duration
//else if (entityType == "DURATION") // the amount of intervening time between the two end-points of a time interval
//{
// java.util.HashMap probHash = anno.get(namedEntityTagProbsAnnotation) as java.util.HashMap;
// double prob = Convert.ToDouble(probHash.get(entityType).ToString()); //java.lang.Double -> string -> double
// int offset = (anno.get(tokenBeginAnnotation) as java.lang.Integer).intValue(); //begin offset
// string normalizedTimex = (string)anno.get(normalizedNamedEntityTagAnnotationClass);
// //set defaultduration information
// int days = 0, hours =0 ,minutes =0 ,seconds = 0, milliseconds = 0;
// //Exact
// if (!normalizedTimex.Contains("/")) //duration ranges are not part of TIMEX3 standard
// if(!normalizedTimex.Contains("X")) //Inexact time
// {
// if (normalizedTimex.Contains("D"))
// days = Convert.ToInt32(normalizedTimex.Replace('P', ' ').Replace('D', ' ').Trim());
// else if (normalizedTimex.Contains("H"))
// hours = Convert.ToInt32(normalizedTimex.Replace('P', ' ').Replace('T',' ').Replace('H', ' ').Trim());
// }
// //create timespan
// TimeSpan span = new TimeSpan(days, hours, minutes, seconds, milliseconds);
// if (prob > CONFIDENCE) stManager.AddTimeStamp(span, offset); //置信度标注
//}
#endregion
}
stManager.AddSentence(sentence);
}
return(stManager);
}
