/// <summary>
/// 依存句法分析（神经网络句法模型需要-Xms1g -Xmx1g -Xmn512m）
/// 内部采用NeuralNetworkDependencyParser实现，用户可以直接调用NeuralNetworkDependencyParser.compute(sentence)
/// 也可以调用基于MaxEnt的依存句法分析器MaxEntDependencyParser.compute(sentence)
/// 《基于神经网络的高性能依存句法分析器》
///https://www.hankcs.com/nlp/parsing/neural-network-based-dependency-parser.html
///《最大熵依存句法分析器的实现》
///https://www.hankcs.com/nlp/parsing/to-achieve-the-maximum-entropy-of-the-dependency-parser.html
///《基于CRF序列标注的中文依存句法分析器的Java实现》
///https://www.hankcs.com/nlp/parsing/crf-sequence-annotation-chinese-dependency-parser-implementation-based-on-java.html
/// </summary>
public void DependencyParser()
{
CoNLLSentence sentence = HanLP.parseDependency("徐先生还具体帮助他确定了把画雄鹰、松鼠和麻雀作为主攻目标。");
Console.WriteLine(sentence);
// 可以方便地遍历它
foreach (CoNLLWord word in sentence)
{
Console.WriteLine($"{word.LEMMA} --({word.DEPREL})--> {word.HEAD.LEMMA}\n");
}
// 也可以直接拿到数组，任意顺序或逆序遍历
CoNLLWord[] wordArray = sentence.getWordArray();
for (int i = wordArray.Length - 1; i >= 0; i--)
{
CoNLLWord word = wordArray[i];
Console.WriteLine($"{word.LEMMA} --({word.DEPREL})-->{word.HEAD.LEMMA}\n");
}
// 还可以直接遍历子树，从某棵子树的某个节点一路遍历到虚根
CoNLLWord head = wordArray[12];
while ((head = head.HEAD) != null)
{
if (head == CoNLLWord.ROOT)
{
Console.WriteLine(head.LEMMA);
}
else
{
Console.WriteLine($"{head.LEMMA} --({head.DEPREL})--> ");
}
}
}
public override CoNLLSentence Parse(List <Term> terms)
{
var table = new Table();
table.v = new string[terms.Count][];
for (int i = 0; i < terms.Count; i++)
{
var term = terms[i];
var line = new string[4];
table.v[i] = line;
line[0] = term.word;
line[2] = DependencyUtil.compilePOS(term.nature);
line[1] = line[2].Substring(0, 1);
}
_crfModel.Tag(table);
var words = new CoNLLWord[table.Size];
for (int i = 0; i < words.Length; i++)
{
words[i] = new CoNLLWord(i + 1, table.v[i][0], table.v[i][2], table.v[i][1]);
}
for (int i = 0; i < table.Size; i++)
{
var line = table.v[i];
var dtag = new DTag(line[3]);
if (dtag.pos.EndsWith("ROOT"))
{
words[i].HEAD = CoNLLWord.ROOT;
}
else
{
var index = ConvertOffset2Index(dtag, table, i);
if (index == -1)
{
words[i].HEAD = CoNLLWord.NULL;
}
else
{
words[i].HEAD = words[index];
}
}
}
for (int i = 0; i < words.Length; i++)
{
words[i].DEPREL = BigramDependencyModel.Get(words[i].NAME, words[i].POSTAG, words[i].HEAD.NAME, words[i].HEAD.POSTAG);
}
return(new CoNLLSentence(words));
}