use of com.hankcs.hanlp.corpus.dependency.CoNll.CoNLLWord in project HanLP by hankcs.
the class NeuralNetworkDependencyParser method parse.
@Override
public CoNLLSentence parse(List<Term> termList) {
List<String> posTagList = PosTagUtil.to863(termList);
List<String> wordList = new ArrayList<String>(termList.size());
for (Term term : termList) {
wordList.add(term.word);
}
List<Integer> heads = new ArrayList<Integer>(termList.size());
List<String> deprels = new ArrayList<String>(termList.size());
parser_dll.parse(wordList, posTagList, heads, deprels);
CoNLLWord[] wordArray = new CoNLLWord[termList.size()];
for (int i = 0; i < wordArray.length; ++i) {
wordArray[i] = new CoNLLWord(i + 1, wordList.get(i), posTagList.get(i), termList.get(i).nature.toString());
wordArray[i].DEPREL = deprels.get(i);
}
for (int i = 0; i < wordArray.length; ++i) {
int index = heads.get(i) - 1;
if (index < 0) {
wordArray[i].HEAD = CoNLLWord.ROOT;
continue;
}
wordArray[i].HEAD = wordArray[index];
}
return new CoNLLSentence(wordArray);
}
use of com.hankcs.hanlp.corpus.dependency.CoNll.CoNLLWord in project HanLP by hankcs.
the class DemoDependencyParser method main.
public static void main(String[] args) {
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。");
System.out.println(sentence);
// 可以方便地遍历它
for (CoNLLWord word : sentence) {
System.out.printf("%s --(%s)--> %s\n", word.LEMMA, word.DEPREL, word.HEAD.LEMMA);
}
// 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];
System.out.printf("%s --(%s)--> %s\n", word.LEMMA, word.DEPREL, word.HEAD.LEMMA);
}
// 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)
System.out.println(head.LEMMA);
else
System.out.printf("%s --(%s)--> ", head.LEMMA, head.DEPREL);
}
}
use of com.hankcs.hanlp.corpus.dependency.CoNll.CoNLLWord in project HanLP by hankcs.
the class TestDependencyCorpus method testMakeCRF.
/**
* 导出CRF训练语料
*
* @throws Exception
*/
public void testMakeCRF() throws Exception {
BufferedWriter bw = new BufferedWriter(new OutputStreamWriter(new FileOutputStream("D:\\Tools\\CRF++-0.58\\example\\dependency\\dev.txt")));
LinkedList<CoNLLSentence> coNLLSentences = CoNLLLoader.loadSentenceList("D:\\Doc\\语料库\\依存分析训练数据\\THU\\dev.conll.fixed.txt");
for (CoNLLSentence coNLLSentence : coNLLSentences) {
for (CoNLLWord coNLLWord : coNLLSentence.word) {
bw.write(coNLLWord.NAME);
bw.write('\t');
bw.write(coNLLWord.CPOSTAG);
bw.write('\t');
bw.write(coNLLWord.POSTAG);
bw.write('\t');
int d = coNLLWord.HEAD.ID - coNLLWord.ID;
int posDistance = 1;
if (// 在后面
d > 0) {
for (int i = 1; i < d; ++i) {
if (coNLLSentence.word[coNLLWord.ID - 1 + i].CPOSTAG.equals(coNLLWord.HEAD.CPOSTAG)) {
++posDistance;
}
}
} else {
for (// 在前面
int i = 1; // 在前面
i < -d; // 在前面
++i) {
if (coNLLSentence.word[coNLLWord.ID - 1 - i].CPOSTAG.equals(coNLLWord.HEAD.CPOSTAG)) {
++posDistance;
}
}
}
bw.write((d > 0 ? "+" : "-") + posDistance + "_" + coNLLWord.HEAD.CPOSTAG);
bw.newLine();
}
bw.newLine();
}
bw.close();
}
use of com.hankcs.hanlp.corpus.dependency.CoNll.CoNLLWord in project HanLP by hankcs.
the class TestDependencyCorpus method testPosTag.
/**
* 细粒度转粗粒度
*
* @throws Exception
*/
public void testPosTag() throws Exception {
DictionaryMaker dictionaryMaker = new DictionaryMaker();
LinkedList<CoNLLSentence> coNLLSentences = CoNLLLoader.loadSentenceList("D:\\Doc\\语料库\\依存分析训练数据\\THU\\dev.conll.fixed.txt");
for (CoNLLSentence coNLLSentence : coNLLSentences) {
for (CoNLLWord coNLLWord : coNLLSentence.word) {
dictionaryMaker.add(new Item(coNLLWord.POSTAG, coNLLWord.CPOSTAG));
}
}
System.out.println(dictionaryMaker.entrySet());
}
use of com.hankcs.hanlp.corpus.dependency.CoNll.CoNLLWord in project HanLP by hankcs.
the class MinimumSpanningTreeParser method parse.
@Override
public CoNLLSentence parse(List<Term> termList) {
if (termList == null || termList.size() == 0)
return null;
termList.add(0, new Term("##核心##", Nature.begin));
Node[] nodeArray = new Node[termList.size()];
Iterator<Term> iterator = termList.iterator();
for (int i = 0; i < nodeArray.length; ++i) {
nodeArray[i] = new Node(iterator.next(), i);
}
Edge[][] edges = new Edge[nodeArray.length][nodeArray.length];
for (int i = 0; i < edges.length; ++i) {
for (int j = 0; j < edges[i].length; ++j) {
if (i != j) {
edges[j][i] = makeEdge(nodeArray, i, j);
}
}
}
// 最小生成树Prim算法
int max_v = nodeArray.length * (nodeArray.length - 1);
float[] mincost = new float[max_v];
Arrays.fill(mincost, Float.MAX_VALUE / 3);
boolean[] used = new boolean[max_v];
Arrays.fill(used, false);
used[0] = true;
PriorityQueue<State> que = new PriorityQueue<State>();
// 找虚根的唯一孩子
float minCostToRoot = Float.MAX_VALUE;
Edge firstEdge = null;
Edge[] edgeResult = new Edge[termList.size() - 1];
for (Edge edge : edges[0]) {
if (edge == null)
continue;
if (minCostToRoot > edge.cost) {
firstEdge = edge;
minCostToRoot = edge.cost;
}
}
if (firstEdge == null)
return null;
que.add(new State(minCostToRoot, firstEdge.from, firstEdge));
while (!que.isEmpty()) {
State p = que.poll();
int v = p.id;
if (used[v] || p.cost > mincost[v])
continue;
used[v] = true;
if (p.edge != null) {
// System.out.println(p.edge.from + " " + p.edge.to + p.edge.label);
edgeResult[p.edge.from - 1] = p.edge;
}
for (Edge e : edges[v]) {
if (e == null)
continue;
if (mincost[e.from] > e.cost) {
mincost[e.from] = e.cost;
que.add(new State(mincost[e.from], e.from, e));
}
}
}
CoNLLWord[] wordArray = new CoNLLWord[termList.size() - 1];
for (int i = 0; i < wordArray.length; ++i) {
wordArray[i] = new CoNLLWord(i + 1, nodeArray[i + 1].word, nodeArray[i + 1].label);
wordArray[i].DEPREL = edgeResult[i].label;
}
for (int i = 0; i < edgeResult.length; ++i) {
int index = edgeResult[i].to - 1;
if (index < 0) {
wordArray[i].HEAD = CoNLLWord.ROOT;
continue;
}
wordArray[i].HEAD = wordArray[index];
}
return new CoNLLSentence(wordArray);
}
Aggregations