Dictionary.cpp
上传用户:sanxfzhen
上传日期:2014-12-28
资源大小:2324k
文件大小:33k
- //////////////////////////////////////////////////////////////////////
- //ICTCLAS简介:计算所汉语词法分析系统ICTCLAS(Institute of Computing Technology, Chinese Lexical Analysis System),
- // 功能有:中文分词;词性标注;未登录词识别。
- // 分词正确率高达97.58%(973专家评测结果),
- // 未登录词识别召回率均高于90%,其中中国人名的识别召回率接近98%;
- // 处理速度为31.5Kbytes/s。
- //著作权: Copyright?2002-2005中科院计算所 职务著作权人:张华平 刘群
- //遵循协议:自然语言处理开放资源许可证1.0
- //Email: zhanghp@software.ict.ac.cn
- //Homepage:www.nlp.org.cn;mtgroup.ict.ac.cn
- /****************************************************************************
- *
- * Copyright (c) 2000, 2001
- * Machine Group
- * Software Research Lab.
- * Institute of Computing Tech.
- * Chinese Academy of Sciences
- * All rights reserved.
- *
- * This file is the confidential and proprietary property of
- * Institute of Computing Tech. and the posession or use of this file requires
- * a written license from the author.
- * Filename: Dictionary.cpp
- * Abstract:
- * implementation of the CDictionary class.
- * Author: Kevin Zhang
- * (zhanghp@software.ict.ac.cn)
- * Date: 2002-1-8
- *
- * Notes:
- *
- ****************************************************************************/
- #include "stdafx.h"
- #include "Dictionary.h"
- #include "Utility.h"
- #include <string.h>
- #include <stdlib.h>
- #include <malloc.h>
- #include <stdio.h>
- //////////////////////////////////////////////////////////////////////
- // Construction/Destruction
- //////////////////////////////////////////////////////////////////////
- CDictionary::CDictionary()
- {
- //initilization
- memset(m_IndexTable,0,sizeof(m_IndexTable));
- m_pModifyTable=NULL;
- m_bReleased=true;
- }
- CDictionary::~CDictionary()
- {
- ReleaseDict();
- }
- /*********************************************************************
- *
- * Func Name : Load
- *
- * Description: Load the dictionary from the file .dct
- *
- *
- * Parameters : sFilename: the file name
- *
- * Returns : success or fail
- * Author : Kevin Zhang
- * History :
- * 1.create 2002-1-9
- *********************************************************************/
- bool CDictionary::Load(char *sFilename,bool bReset)
- {
- FILE *fp;
- int i,j,nBuffer[3];
- if((fp=fopen(sFilename,"rb"))==NULL)
- return false;//fail while opening the file
- ReleaseDict();
- m_bReleased=false;
- for(i=0;i<CC_NUM;i++)
- {
- fread(&(m_IndexTable[i].nCount),sizeof(int),1,fp);
- if(m_IndexTable[i].nCount>0)
- m_IndexTable[i].pWordItemHead=new WORD_ITEM[m_IndexTable[i].nCount];
- else
- {
- m_IndexTable[i].pWordItemHead=0;
- continue;
- }
- j=0;
- while(j<m_IndexTable[i].nCount)
- {
- fread(nBuffer,sizeof(int),3,fp);
- m_IndexTable[i].pWordItemHead[j].sWord=new char[nBuffer[1]+1];
- if(nBuffer[1])//String length is more than 0
- {
- fread(m_IndexTable[i].pWordItemHead[j].sWord,sizeof(char),nBuffer[1],fp);
- }
- m_IndexTable[i].pWordItemHead[j].sWord[nBuffer[1]]=0;
- if(bReset)//Reset the frequency
- m_IndexTable[i].pWordItemHead[j].nFrequency=0;
- else
- m_IndexTable[i].pWordItemHead[j].nFrequency=nBuffer[0];
- m_IndexTable[i].pWordItemHead[j].nWordLen=nBuffer[1];
- m_IndexTable[i].pWordItemHead[j].nHandle=nBuffer[2];
- j+=1;//Get next item in the original table.
- }
- }
- fclose(fp);
- return true;
- }
- /*********************************************************************
- *
- * Func Name : Save
- *
- * Description: Save the dictionary as the file .dct
- *
- *
- * Parameters : sFilename: the file name
- *
- * Returns : success or fail
- * Author : Kevin Zhang
- * History :
- * 1.create 2002-1-9
- *********************************************************************/
- bool CDictionary::Save(char *sFilename)
- {
- FILE *fp;
- int i,j,nCount,nBuffer[3];
- PWORD_CHAIN pCur;
- if((fp=fopen(sFilename,"wb"))==NULL)
- return false;//fail while opening the file
- for(i=0;i<CC_NUM;i++)
- {
- pCur=NULL;
- if(m_pModifyTable)
- {//Modification made
- nCount=m_IndexTable[i].nCount+m_pModifyTable[i].nCount-m_pModifyTable[i].nDelete;
- fwrite(&nCount,sizeof(int),1,fp);
- pCur=m_pModifyTable[i].pWordItemHead;
- j=0;
- while(pCur!=NULL&&j<m_IndexTable[i].nCount)
- {//Output to the file after comparision
- if(strcmp(pCur->data.sWord,m_IndexTable[i].pWordItemHead[j].sWord)<0||(strcmp(pCur->data.sWord,m_IndexTable[i].pWordItemHead[j].sWord)==0&&pCur->data.nHandle<m_IndexTable[i].pWordItemHead[j].nHandle))
- {//Output the modified data to the file
- nBuffer[0]=pCur->data.nFrequency;
- nBuffer[1]=pCur->data.nWordLen;
- nBuffer[2]=pCur->data.nHandle;
- fwrite(nBuffer,sizeof(int),3,fp);
- if(nBuffer[1])//String length is more than 0
- fwrite(pCur->data.sWord,sizeof(char),nBuffer[1],fp);
- pCur=pCur->next;//Get next item in the modify table.
- }
- else if(m_IndexTable[i].pWordItemHead[j].nFrequency==-1)
- {//The item has been removed,so skip it
- j+=1;
- }
- else if(strcmp(pCur->data.sWord,m_IndexTable[i].pWordItemHead[j].sWord)>0||(strcmp(pCur->data.sWord,m_IndexTable[i].pWordItemHead[j].sWord)==0&&pCur->data.nHandle>m_IndexTable[i].pWordItemHead[j].nHandle))
- {//Output the index table data to the file
- nBuffer[0]=m_IndexTable[i].pWordItemHead[j].nFrequency;
- nBuffer[1]=m_IndexTable[i].pWordItemHead[j].nWordLen;
- nBuffer[2]=m_IndexTable[i].pWordItemHead[j].nHandle;
- fwrite(nBuffer,sizeof(int),3,fp);
- if(nBuffer[1])//String length is more than 0
- fwrite(m_IndexTable[i].pWordItemHead[j].sWord,sizeof(char),nBuffer[1],fp);
- j+=1;//Get next item in the original table.
- }
- }
- if(j<m_IndexTable[i].nCount)
- {
- while(j<m_IndexTable[i].nCount)
- {
- if(m_IndexTable[i].pWordItemHead[j].nFrequency!=-1)
- {//Has been deleted
- nBuffer[0]=m_IndexTable[i].pWordItemHead[j].nFrequency;
- nBuffer[1]=m_IndexTable[i].pWordItemHead[j].nWordLen;
- nBuffer[2]=m_IndexTable[i].pWordItemHead[j].nHandle;
- fwrite(nBuffer,sizeof(int),3,fp);
- if(nBuffer[1])//String length is more than 0
- fwrite(m_IndexTable[i].pWordItemHead[j].sWord,sizeof(char),nBuffer[1],fp);
- }
- j+=1;//Get next item in the original table.
- }
- }
- else////No Modification
- while(pCur!=NULL)//Add the rest data to the file.
- {
- nBuffer[0]=pCur->data.nFrequency;
- nBuffer[1]=pCur->data.nWordLen;
- nBuffer[2]=pCur->data.nHandle;
- fwrite(nBuffer,sizeof(int),3,fp);
- if(nBuffer[1])//String length is more than 0
- fwrite(pCur->data.sWord,sizeof(char),nBuffer[1],fp);
- pCur=pCur->next;//Get next item in the modify table.
- }
- }
- else
- {
- fwrite(&m_IndexTable[i].nCount,sizeof(int),1,fp);
- //write to the file
- j=0;
- while(j<m_IndexTable[i].nCount)
- {
- nBuffer[0]=m_IndexTable[i].pWordItemHead[j].nFrequency;
- nBuffer[1]=m_IndexTable[i].pWordItemHead[j].nWordLen;
- nBuffer[2]=m_IndexTable[i].pWordItemHead[j].nHandle;
- fwrite(nBuffer,sizeof(int),3,fp);
- if(nBuffer[1])//String length is more than 0
- fwrite(m_IndexTable[i].pWordItemHead[j].sWord,sizeof(char),nBuffer[1],fp);
- j+=1;//Get next item in the original table.
- }
- }
- }
- fclose(fp);
- return true;
- }
- /*********************************************************************
- *
- * Func Name : AddItem
- *
- * Description: Add a word item to the dictionary
- *
- *
- * Parameters : sWord: the word
- * nHandle:the handle number
- * nFrequency: the frequency
- * Returns : success or fail
- * Author : Kevin Zhang
- * History :
- * 1.create 2002-1-9
- *********************************************************************/
- bool CDictionary::AddItem(char *sWord, int nHandle,int nFrequency)
- {
- char sWordAdd[WORD_MAXLENGTH-2];
- int nPos,nFoundPos;
- PWORD_CHAIN pRet,pTemp,pNext;
- int i=0;
- if(!PreProcessing(sWord, &nPos,sWordAdd,true))
- return false;
- if(FindInOriginalTable(nPos,sWordAdd,nHandle,&nFoundPos))
- {//The word exists in the original table, so add the frequency
- //Operation in the index table and its items
- if(m_IndexTable[nPos].pWordItemHead[nFoundPos].nFrequency==-1)
- {//The word item has been removed
- m_IndexTable[nPos].pWordItemHead[nFoundPos].nFrequency=nFrequency;
- if(!m_pModifyTable)//Not prepare the buffer
- {
- m_pModifyTable=new MODIFY_TABLE[CC_NUM];
- memset(m_pModifyTable,0,CC_NUM*sizeof(MODIFY_TABLE));
- }
- m_pModifyTable[nPos].nDelete-=1;
- }
- else
- m_IndexTable[nPos].pWordItemHead[nFoundPos].nFrequency+=nFrequency;
- return true;
- }
- //The items not exists in the index table.
- //As following, we have to find the item whether exists in the modify data region
- //If exists, change the frequency .or else add a item
- if(!m_pModifyTable)//Not prepare the buffer
- {
- m_pModifyTable=new MODIFY_TABLE[CC_NUM];
- memset(m_pModifyTable,0,CC_NUM*sizeof(MODIFY_TABLE));
- }
- if(FindInModifyTable(nPos,sWordAdd,nHandle,&pRet))
- {
- if(pRet!=NULL)
- pRet=pRet->next;
- else
- pRet=m_pModifyTable[nPos].pWordItemHead;
- pRet->data.nFrequency+=nFrequency;
- return true;
- }
- //find the proper position to add the word to the modify data table and link
- pTemp=new WORD_CHAIN;//Allocate the word chain node
- if(pTemp==NULL)//Allocate memory failure
- return false;
- memset(pTemp,0,sizeof(WORD_CHAIN));//init it with 0
- pTemp->data.nHandle=nHandle;//store the handle
- pTemp->data.nWordLen=strlen(sWordAdd);
- pTemp->data.sWord=new char[1+pTemp->data.nWordLen];
- strcpy(pTemp->data.sWord,sWordAdd);
- pTemp->data.nFrequency=nFrequency;
- pTemp->next=NULL;
- if(pRet!=NULL)
- {
- pNext=pRet->next;//Get the next item before the current item
- pRet->next=pTemp;//link the node to the chain
- }
- else
- {
- pNext=m_pModifyTable[nPos].pWordItemHead;
- m_pModifyTable[nPos].pWordItemHead=pTemp;//Set the pAdd as the head node
- }
- pTemp->next=pNext;//Very important!!!! or else it will lose some node
- //Modify in 2001-10-29
- m_pModifyTable[nPos].nCount++;//the number increase by one
- return true;
- }
- bool CDictionary::DelItem(char *sWord,int nHandle)
- {
- char sWordDel[WORD_MAXLENGTH-2];
- int nPos,nFoundPos,nTemp;
- PWORD_CHAIN pPre,pTemp,pCur;
- if(!PreProcessing(sWord, &nPos,sWordDel))
- return false;
- if(FindInOriginalTable(nPos,sWordDel,nHandle,&nFoundPos))
- {
- if(!m_pModifyTable)//Not prepare the buffer
- {
- m_pModifyTable=new MODIFY_TABLE[CC_NUM];
- memset(m_pModifyTable,0,CC_NUM*sizeof(MODIFY_TABLE));
- }
- m_IndexTable[nPos].pWordItemHead[nFoundPos].nFrequency=-1;
- m_pModifyTable[nPos].nDelete+=1;
- if(nHandle==-1)//Remove all items which word is sWordDel,ignoring the handle
- {
- /* nTemp=nFoundPos-1;//Check its previous position
- while(nTemp>0&&strcmp(m_IndexTable[nPos].pWordItemHead[nFoundPos].sWord,sWordDel)==0)
- {
- m_IndexTable[nPos].pWordItemHead[nTemp].nFrequency=-1;
- m_pModifyTable[nPos].nDelete+=1;
- nTemp-=1;
- }
- */ nTemp=nFoundPos+1;//Check its previous position
- while(nTemp<m_IndexTable[nPos].nCount&&strcmp(m_IndexTable[nPos].pWordItemHead[nFoundPos].sWord,sWordDel)==0)
- {
- m_IndexTable[nPos].pWordItemHead[nTemp].nFrequency=-1;
- m_pModifyTable[nPos].nDelete+=1;
- nTemp+=1;
- }
- }
- return true;
- }
- //Operation in the modify table and its items
- if(FindInModifyTable(nPos,sWordDel,nHandle,&pPre))
- {
- pCur=m_pModifyTable[nPos].pWordItemHead;
- if(pPre!=NULL)
- pCur=pPre->next;
- while(pCur!=NULL && _stricmp(pCur->data.sWord,sWordDel)==0&&(pCur->data.nHandle==nHandle||nHandle<0))
- {
- pTemp=pCur;
- if(pPre!=NULL)//pCur is the first item
- pPre->next=pCur->next;
- else
- m_pModifyTable[nPos].pWordItemHead=pCur->next;
- pCur=pCur->next;
- delete pTemp->data.sWord;//Delete the word
- delete pTemp;
- }
- return true;
- }
- return false;
- }
- bool CDictionary::DelModified()
- {
- PWORD_CHAIN pTemp,pCur;
- if(!m_pModifyTable)
- return true;
- for(int i=0;i<CC_NUM;i++)
- {
- pCur=m_pModifyTable[i].pWordItemHead;
- while(pCur!=NULL)
- {
- pTemp=pCur;
- pCur=pCur->next;
- delete pTemp->data.sWord;
- delete pTemp;
- }
- }
- delete [] m_pModifyTable;
- m_pModifyTable=NULL;
- return true;
- }
- bool CDictionary::IsExist(char *sWord, int nHandle)
- {
- char sWordFind[WORD_MAXLENGTH-2];
- int nPos;
- if(!PreProcessing(sWord, &nPos,sWordFind))
- return false;
- return(FindInOriginalTable(nPos,sWordFind,nHandle)||FindInModifyTable(nPos,sWordFind,nHandle));
- }
- bool CDictionary::GetHandle(char *sWord,int *pnCount,int *pnHandle,int *pnFrequency)
- {
- char sWordGet[WORD_MAXLENGTH-2];
- int nPos,nFoundPos,nTemp;
- PWORD_CHAIN pPre,pCur;
- *pnCount=0;
- if(!PreProcessing(sWord, &nPos,sWordGet))
- return false;
- if(FindInOriginalTable(nPos,sWordGet,-1,&nFoundPos))
- {
- pnHandle[*pnCount]=m_IndexTable[nPos].pWordItemHead[nFoundPos].nHandle;
- pnFrequency[*pnCount]=m_IndexTable[nPos].pWordItemHead[nFoundPos].nFrequency;
- *pnCount+=1;
- /* nTemp=nFoundPos-1;//Check its previous position
- while(nTemp>0&&strcmp(m_IndexTable[nPos].pWordItemHead[nTemp].sWord,sWordGet)==0)
- {
- pnHandle[*pnCount]=m_IndexTable[nPos].pWordItemHead[nTemp].nHandle;
- pnFrequency[*pnCount]=m_IndexTable[nPos].pWordItemHead[nTemp].nFrequency;
- *pnCount+=1;
- nTemp-=1;
- }
- */ nTemp=nFoundPos+1;//Check its previous position
- while(nTemp<m_IndexTable[nPos].nCount&&strcmp(m_IndexTable[nPos].pWordItemHead[nTemp].sWord,sWordGet)==0)
- {
- pnHandle[*pnCount]=m_IndexTable[nPos].pWordItemHead[nTemp].nHandle;
- pnFrequency[*pnCount]=m_IndexTable[nPos].pWordItemHead[nTemp].nFrequency;
- *pnCount+=1;
- nTemp+=1;
- }
- return true;
- }
- //Operation in the index table and its items
- if(FindInModifyTable(nPos,sWordGet,-1,&pPre))
- {
- pCur=m_pModifyTable[nPos].pWordItemHead;
- if(pPre!=NULL)
- pCur=pPre->next;
- while(pCur!=NULL && _stricmp(pCur->data.sWord,sWordGet)==0)
- {
- pnHandle[*pnCount]=pCur->data.nHandle;
- pnFrequency[*pnCount]=pCur->data.nFrequency;
- *pnCount+=1;
- pCur=pCur->next;
- }
- return true;
- }
- return false;
- }
- /*********************************************************************
- *
- * Func Name : FindInOriginalTable
- *
- * Description: judge the word and handle exist in the inner table and its items
- *
- *
- * Parameters : nInnerCode: the inner code of the first CHines char
- * sWord: the word
- * nHandle:the handle number
- * *nPosRet:the position which node is matched
- *
- * Returns : success or fail
- * Author : Kevin Zhang
- * History :
- * 1.create 2002-1-9
- *********************************************************************/
- bool CDictionary::FindInOriginalTable(int nInnerCode,char *sWord,int nHandle,int *nPosRet)
- {
- PWORD_ITEM pItems=m_IndexTable[nInnerCode].pWordItemHead;
- int nStart=0,nEnd=m_IndexTable[nInnerCode].nCount-1,nMid=(nStart+nEnd)/2,nCount=0,nCmpValue;
- while(nStart<=nEnd)//Binary search
- {
- nCmpValue=strcmp(pItems[nMid].sWord,sWord);
- if(nCmpValue==0&&(pItems[nMid].nHandle==nHandle||nHandle==-1))
- {
- if(nPosRet)
- {
- if(nHandle==-1)//Not very strict match
- {//Add in 2002-1-28
- nMid-=1;
- while(nMid>=0&&strcmp(pItems[nMid].sWord,sWord)==0)
- //Get the first item which match the current word
- nMid--;
- if(nMid<0||strcmp(pItems[nMid].sWord,sWord)!=0)
- nMid++;
- }
- *nPosRet=nMid;
- return true;
- }
- if(nPosRet)
- *nPosRet=nMid;
- return true;//find it
- }
- else if(nCmpValue<0||(nCmpValue==0&&pItems[nMid].nHandle<nHandle&&nHandle!=-1))
- {
- nStart=nMid+1;
- }
- else if(nCmpValue>0||(nCmpValue==0&&pItems[nMid].nHandle>nHandle&&nHandle!=-1))
- {
- nEnd=nMid-1;
- }
- nMid=(nStart+nEnd)/2;
- }
- if(nPosRet)
- {
- //Get the previous position
- *nPosRet=nMid-1;
- }
- return false;
- }
- /*********************************************************************
- *
- * Func Name : FindInModifyTable
- *
- * Description: judge the word and handle exist in the modified table and its items
- *
- *
- * Parameters : nInnerCode: the inner code of the first CHines char
- * sWord: the word
- * nHandle:the handle number
- * *pFindRet: the node found
- *
- * Returns : success or fail
- * Author : Kevin Zhang
- * History :
- * 1.create 2002-1-9
- *********************************************************************/
- bool CDictionary::FindInModifyTable(int nInnerCode,char *sWord,int nHandle,PWORD_CHAIN *pFindRet)
- {
- PWORD_CHAIN pCur,pPre;
- if(m_pModifyTable==NULL)//empty
- return false;
- pCur=m_pModifyTable[nInnerCode].pWordItemHead;
- pPre=NULL;
- while(pCur!=NULL&&(_stricmp(pCur->data.sWord,sWord)<0||(_stricmp(pCur->data.sWord,sWord)==0&&pCur->data.nHandle<nHandle)))
- //sort the link chain as alphabet
- {
- pPre=pCur;
- pCur=pCur->next;
- }
- if(pFindRet)
- *pFindRet=pPre;
- if(pCur!=NULL && _stricmp(pCur->data.sWord,sWord)==0&&(pCur->data.nHandle==nHandle||nHandle<0))
- {//The node exists, delete the node and return
- return true;
- }
- return false;
- }
- /*********************************************************************
- *
- * Func Name : GetWordType
- *
- * Description: Get the type of word
- *
- *
- * Parameters : sWord: the word
- * Returns : the type
- * Author : Kevin Zhang
- * History :
- * 1.create 2002-1-9
- *********************************************************************/
- int CDictionary::GetWordType(char *sWord)
- {
- int nType=charType((unsigned char *)sWord),nLen=strlen(sWord);
- if(nLen>0&&nType==CT_CHINESE&&IsAllChinese((unsigned char *)sWord))
- return WT_CHINESE;//Chinese word
- else if(nLen>0&&nType==CT_DELIMITER)
- return WT_DELIMITER;//Delimiter
- else
- return WT_OTHER;//other invalid
- }
- /*********************************************************************
- *
- * Func Name : PreProcessing
- *
- * Description: Get the type of word
- *
- *
- * Parameters : sWord: the word
- * Returns : the type
- * Author : Kevin Zhang
- * History :
- * 1.create 2002-1-9
- *********************************************************************/
- bool CDictionary::PreProcessing(char *sWord, int *nId, char *sWordRet,bool bAdd)
- {
- //Position for the delimeters
- int nType=charType((unsigned char *)sWord),nLen=strlen(sWord);
- int nEnd=nLen-1,nBegin=0;
- if(nLen==0)
- return false;
- while(nEnd>=0&&sWord[nEnd]==' ')
- nEnd-=1;
- while(nBegin<=nEnd&&sWord[nBegin]==' ')
- nBegin+=1;
- if(nBegin>nEnd)
- return false;
- if(nEnd!=nLen-1||nBegin!=0)
- {
- strncpy(sWord,sWord+nBegin,nEnd-nBegin+1);
- sWord[nEnd-nBegin+1]=0;
- }
- /*
- if((bAdd||strlen(sWord)>4)&&IsAllChineseNum(sWord))
- { //Only convert the Chinese Num to 3755 while
- //Get the inner code of the first Chinese Char
- strcpy(sWord,"五十八");
- }
- */
- if(nType==CT_CHINESE)//&&IsAllChinese((unsigned char *)sWord)
- {//Chinese word
- *nId=CC_ID(sWord[0],sWord[1]);
- //Get the inner code of the first Chinese Char
- strcpy(sWordRet,&sWord[2]);//store the word,not store the first Chinese Char
- return true;
- }
- /* if(nType==CT_NUM&&IsAllNum((unsigned char *)sWord))
- {
- *nId=3756;
- //Get the inner code of the first Chinese Char
- sWordRet[0]=0;//store the word,not store the first Chinese Char
- return true;
- }
- */ if(nType==CT_DELIMITER)
- {//Delimiter
- *nId=3755;
- //Get the inner code of the first Chinese Char
- strcpy(sWordRet,sWord);//store the word,not store the first Chinese Char
- return true;
- }
- /*
- if(nType==CT_LETTER&&IsAllLetter((unsigned char *)sWord))
- {
- *nId=3757;
- //Get the inner code of the first Chinese Char
- sWordRet[0]=0;//store the word,not store the first Chinese Char
- return true;
- }
- if(nType==CT_SINGLE&&IsAllSingleByte((unsigned char *)sWord))
- {
- *nId=3758;
- //Get the inner code of the first Chinese Char
- sWordRet[0]=0;//store the word,not store the first Chinese Char
- return true;
- }
- if(nType==CT_INDEX&&IsAllIndex((unsigned char *)sWord))
- {
- *nId=3759;
- //Get the inner code of the first Chinese Char
- sWordRet[0]=0;//store the word,not store the first Chinese Char
- return true;
- }
- */
- return false;//other invalid
- }
- /*********************************************************************
- *
- * Func Name : MergePOS
- *
- * Description: Merge all the POS into nHandle,
- * just get the word in the dictionary and set its Handle as nHandle
- *
- *
- * Parameters : nHandle: the only handle which will be attached to the word
- * Returns : the type
- * Author : Kevin Zhang
- * History :
- * 1.create 2002-1-21
- *********************************************************************/
- bool CDictionary::MergePOS(int nHandle)
- {
- int i,j,nCompare;
- char sWordPrev[WORD_MAXLENGTH];
- PWORD_CHAIN pPre,pCur,pTemp;
- if(!m_pModifyTable)//Not prepare the buffer
- {
- m_pModifyTable=new MODIFY_TABLE[CC_NUM];
- memset(m_pModifyTable,0,CC_NUM*sizeof(MODIFY_TABLE));
- }
- for( i=0;i<CC_NUM;i++)//Operation in the index table
- {//delete the memory of word item array in the dictionary
- sWordPrev[0]=0;//Set empty
- for(j=0;j<m_IndexTable[i].nCount;j++)
- {
- nCompare=_stricmp(sWordPrev,m_IndexTable[i].pWordItemHead[j].sWord);
- if((j==0||nCompare<0)&&m_IndexTable[i].pWordItemHead[j].nFrequency!=-1)
- {//Need to modify its handle
- m_IndexTable[i].pWordItemHead[j].nHandle=nHandle;//Change its handle
- strcpy(sWordPrev,m_IndexTable[i].pWordItemHead[j].sWord);//Refresh previous Word
- }
- else if(nCompare==0&&m_IndexTable[i].pWordItemHead[j].nFrequency!=-1)
- {//Need to delete when not delete and same as previous word
- m_IndexTable[i].pWordItemHead[j].nFrequency=-1;//Set delete flag
- m_pModifyTable[i].nDelete+=1;//Add the number of being deleted
- }
- }
- }
- for( i=0;i<CC_NUM;i++)//Operation in the modify table
- {
- pPre=NULL;
- pCur=m_pModifyTable[i].pWordItemHead;
- sWordPrev[0]=0;//Set empty
- while(pCur!=NULL)
- {
- if(_stricmp(pCur->data.sWord,sWordPrev)>0)
- {//The new word
- pCur->data.nHandle=nHandle;//Chang its handle
- strcpy(sWordPrev,pCur->data.sWord);//Set new previous word
- pPre=pCur;//New previous pointer
- pCur=pCur->next;
- }
- else
- {//The same word as previous,delete it.
- pTemp=pCur;
- if(pPre!=NULL)//pCur is the first item
- pPre->next=pCur->next;
- else
- m_pModifyTable[i].pWordItemHead=pCur->next;
- pCur=pCur->next;
- delete pTemp->data.sWord;//Delete the word
- delete pTemp;//Delete the item
- }
- }
- }
- return true;
- }
- /*********************************************************************
- *
- * Func Name : GetMaxMatch
- *
- * Description: Get the max match to the word
- *
- *
- * Parameters : nHandle: the only handle which will be attached to the word
- * Returns : success or fail
- * Author : Kevin Zhang
- * History :
- * 1.create 2002-1-21
- *********************************************************************/
- bool CDictionary::GetMaxMatch(char *sWord, char *sWordRet,int *npHandleRet)
- {
- char sWordGet[WORD_MAXLENGTH-2],sFirstChar[3];
- int nPos,nFoundPos,nTemp;
- PWORD_CHAIN pCur;
- *npHandleRet=-1;
- if(!PreProcessing(sWord, &nPos,sWordGet))
- return false;
- sWordRet[0]=0;
- strncpy(sFirstChar,sWord,strlen(sWord)-strlen(sWordGet));//Get the first char
- sFirstChar[strlen(sWord)-strlen(sWordGet)]=0;//Set the end flag
- FindInOriginalTable(nPos,sWordGet,-1,&nFoundPos);
- nTemp=nFoundPos;//Check its previous position
- if(nFoundPos==-1)
- nTemp=0;
- while(nTemp<m_IndexTable[nPos].nCount&&CC_Find(m_IndexTable[nPos].pWordItemHead[nTemp].sWord,sWordGet)!=m_IndexTable[nPos].pWordItemHead[nTemp].sWord)
- {//Get the next
- nTemp+=1;
- }
- if(nTemp<m_IndexTable[nPos].nCount&&CC_Find(m_IndexTable[nPos].pWordItemHead[nTemp].sWord,sWordGet)==m_IndexTable[nPos].pWordItemHead[nTemp].sWord)
- {
- strcpy(sWordRet,sFirstChar);
- strcat(sWordRet,m_IndexTable[nPos].pWordItemHead[nTemp].sWord);
- *npHandleRet=m_IndexTable[nPos].pWordItemHead[nTemp].nHandle;
- return true;
- }//Cannot get the item and retrieve the modified data if exists
- //Operation in the index table and its items
- if(m_pModifyTable&&m_pModifyTable[nPos].pWordItemHead)//Exists
- pCur=m_pModifyTable[nPos].pWordItemHead;
- else
- pCur=NULL;
- while(pCur!=NULL&&strcmp(pCur->data.sWord,sWordGet)<=0&&CC_Find(pCur->data.sWord,sWordGet)!=pCur->data.sWord)//
- {
- pCur=pCur->next;
- }
- if(pCur!=NULL&&CC_Find(pCur->data.sWord,sWordGet)!=pCur->data.sWord)
- {//Get it
- strcpy(sWordRet,sFirstChar);
- strcat(sWordRet,pCur->data.sWord);
- *npHandleRet=pCur->data.nHandle;
- return true;
- }
- return false;
- }
- /*********************************************************************
- *
- * Func Name : GetPOSValue
- *
- * Description: Get the POS value according the POS string
- *
- *
- * Parameters :
- * Returns : the value
- * Author : Kevin Zhang
- * History :
- * 1.create 2002-1-29
- *********************************************************************/
- int CDictionary::GetPOSValue(char *sPOS)
- {
- int nPOS;
- char *sPlusPos,sTemp[4];
- if(strlen(sPOS)<3)
- {
- nPOS=sPOS[0]*256+sPOS[1];
- }
- else
- {
- sPlusPos=strchr(sPOS,'+');
- strncpy(sTemp,sPOS,sPlusPos-sPOS);
- sTemp[sPlusPos-sPOS]=0;
- nPOS=100*GetPOSValue(sTemp);
- strncpy(sTemp,sPlusPos+1,4);
- nPOS+=atoi(sTemp);
- }
- return nPOS;
- }
- /*********************************************************************
- *
- * Func Name : GetPOSString
- *
- * Description: Get the POS string according the POS value
- *
- *
- * Parameters :
- * Returns : success or fail
- * Author : Kevin Zhang
- * History :
- * 1.create 2002-1-29
- *********************************************************************/
- bool CDictionary::GetPOSString(int nPOS, char *sPOSRet)
- {
- if(nPOS>'a'*25600)
- {
- if((nPOS/100)%256!=0)
- sprintf(sPOSRet,"%c%c+%d",nPOS/25600,(nPOS/100)%256,nPOS%100);
- else
- sprintf(sPOSRet,"%c+%d",nPOS/25600,nPOS%100);
- }
- else
- {
- if(nPOS>256)
- sprintf(sPOSRet,"%c%c",nPOS/256,nPOS%256);
- else
- sprintf(sPOSRet,"%c",nPOS%256);
- }
- return true;
- }
- int CDictionary::GetFrequency(char *sWord, int nHandle)
- {
- char sWordFind[WORD_MAXLENGTH-2];
- int nPos,nIndex;
- PWORD_CHAIN pFound;
- if(!PreProcessing(sWord, &nPos,sWordFind))
- return 0;
- if(FindInOriginalTable(nPos,sWordFind,nHandle,&nIndex))
- {
- return m_IndexTable[nPos].pWordItemHead[nIndex].nFrequency;
- }
- if(FindInModifyTable(nPos,sWordFind,nHandle,&pFound))
- {
- return pFound->data.nFrequency;
- }
- return 0;
- }
- bool CDictionary::Output(char *sFilename)
- {
- FILE *fp;
- int i,j;
- PWORD_CHAIN pCur;
- char sPrevWord[WORD_MAXLENGTH]="", sCurWord[WORD_MAXLENGTH],sPOS[10];
- if((fp=fopen(sFilename,"wb"))==NULL)
- return false;//fail while opening the file
- if(m_pModifyTable)
- {//Modification made, not to output when modify table exists.
- return false;
- }
- for(i=0;i<CC_NUM;i++)
- {
- pCur=NULL;
- j=0;
- while(j<m_IndexTable[i].nCount)
- {
- GetPOSString(m_IndexTable[i].pWordItemHead[j].nHandle,sPOS);
- //Get the POS string
- sprintf(sCurWord,"%c%c%s",CC_CHAR1(i),CC_CHAR2(i),m_IndexTable[i].pWordItemHead[j].sWord);
- if(strcmp(sPrevWord,sCurWord)!=0)
- fprintf(fp,"n%s %s",sCurWord,sPOS);
- else
- fprintf(fp," %s",sPOS);
- strcpy(sPrevWord,sCurWord);
- j+=1;//Get next item in the original table.
- }
- }
- fclose(fp);
- return true;
- }
- bool CDictionary::OutputChars(char *sFilename)
- {
- FILE *fp;
- int i,j;
- char sPrevWord[WORD_MAXLENGTH]="", sCurWord[WORD_MAXLENGTH];
- if((fp=fopen(sFilename,"wb"))==NULL)
- return false;//fail while opening the file
- if(m_pModifyTable)
- {//Modification made, not to output when modify table exists.
- return false;
- }
- for(i=0;i<CC_NUM;i++)
- {
- j=0;
- while(j<m_IndexTable[i].nCount)
- {
- sprintf(sCurWord,"%c%c%s",CC_CHAR1(i),CC_CHAR2(i),m_IndexTable[i].pWordItemHead[j].sWord);
- if(strcmp(sPrevWord,sCurWord)!=0&&m_IndexTable[i].pWordItemHead[j].nFrequency>50)//
- fprintf(fp,"%s",sCurWord);
- strcpy(sPrevWord,sCurWord);
- j+=1;//Get next item in the original table.
- }
- }
- fclose(fp);
- return true;
- }
- bool CDictionary::Merge(CDictionary dict2, int nRatio)
- //Merge dict2 into current dictionary and the frequency ratio from dict2 and current dict is nRatio
- {
- int i,j,k,nCmpValue;
- char sWord[WORD_MAXLENGTH];
- if(m_pModifyTable||dict2.m_pModifyTable)
- {//Modification made, not to output when modify table exists.
- return false;
- }
- for(i=0;i<CC_NUM;i++)
- {
- j=0;
- k=0;
- while(j<m_IndexTable[i].nCount&&k<dict2.m_IndexTable[i].nCount)
- {
- nCmpValue=strcmp(m_IndexTable[i].pWordItemHead[j].sWord,dict2.m_IndexTable[i].pWordItemHead[k].sWord);
- if(nCmpValue==0)//Same Words and determine the different handle
- {
- if(m_IndexTable[i].pWordItemHead[j].nHandle<dict2.m_IndexTable[i].pWordItemHead[k].nHandle)
- nCmpValue=-1;
- else if(m_IndexTable[i].pWordItemHead[j].nHandle>dict2.m_IndexTable[i].pWordItemHead[k].nHandle)
- nCmpValue=1;
- }
- if(nCmpValue==0)
- {
- m_IndexTable[i].pWordItemHead[j].nFrequency=(nRatio*m_IndexTable[i].pWordItemHead[j].nFrequency+dict2.m_IndexTable[i].pWordItemHead[k].nFrequency)/(nRatio+1);
- j+=1;
- k+=1;
- }
- else if(nCmpValue<0)//Get next word in the current dictionary
- {
- m_IndexTable[i].pWordItemHead[j].nFrequency=(nRatio*m_IndexTable[i].pWordItemHead[j].nFrequency)/(nRatio+1);
- j+=1;
- }
- else//Get next word in the second dictionary
- {
- if(dict2.m_IndexTable[i].pWordItemHead[k].nFrequency>(nRatio+1)/10)
- {
- sprintf(sWord,"%c%c%s",CC_CHAR1(i),CC_CHAR2(i),dict2.m_IndexTable[i].pWordItemHead[k].sWord);
- AddItem(sWord,dict2.m_IndexTable[i].pWordItemHead[k].nHandle,dict2.m_IndexTable[i].pWordItemHead[k].nFrequency/(nRatio+1));
- }
- k+=1;
- }
- }
- while(j<m_IndexTable[i].nCount)//words in current dictionary are left
- {
- m_IndexTable[i].pWordItemHead[j].nFrequency=(nRatio*m_IndexTable[i].pWordItemHead[j].nFrequency)/(nRatio+1);
- j+=1;
- }
- while(k<dict2.m_IndexTable[i].nCount)//words in Dict2 are left
- {
- if(dict2.m_IndexTable[i].pWordItemHead[k].nFrequency>(nRatio+1)/10)
- {
- sprintf(sWord,"%c%c%s",CC_CHAR1(i),CC_CHAR2(i),dict2.m_IndexTable[i].pWordItemHead[k].sWord);
- AddItem(sWord,dict2.m_IndexTable[i].pWordItemHead[k].nHandle,dict2.m_IndexTable[i].pWordItemHead[k].nFrequency/(nRatio+1));
- }
- k+=1;
- }
- }
- return true;
- }
- //Delete word item which
- //(1)frequency is 0
- //(2)word is same as following but the POS value is parent set of the following
- //for example "江泽民/n/0" will deleted, because "江泽民/nr/0" is more detail and correct
- bool CDictionary::Optimum()
- {
- int nPrevPOS,i,j,nPrevFreq;
- char sPrevWord[WORD_MAXLENGTH],sCurWord[WORD_MAXLENGTH];
- for(i=0;i<CC_NUM;i++)
- {
- j=0;
- sPrevWord[0]=0;
- nPrevPOS=0;
- nPrevFreq=-1;
- while(j<m_IndexTable[i].nCount)
- {
- sprintf(sCurWord,"%c%c%s",CC_CHAR1(i),CC_CHAR2(i),m_IndexTable[i].pWordItemHead[j].sWord);
- if(nPrevPOS==30720||nPrevPOS==26368||nPrevPOS==29031||(strcmp(sPrevWord,sCurWord)==0&&nPrevFreq==0&&m_IndexTable[i].pWordItemHead[j].nHandle/256*256==nPrevPOS))
- {//Delete Previous word item
- //Delete word with POS 'x','g' 'qg'
- DelItem(sPrevWord,nPrevPOS);
- }
- strcpy(sPrevWord,sCurWord);
- nPrevPOS=m_IndexTable[i].pWordItemHead[j].nHandle;
- nPrevFreq=m_IndexTable[i].pWordItemHead[j].nFrequency;
- j+=1;//Get next item in the original table.
- }
- }
- return true;
- }
- void CDictionary::ReleaseDict()
- {
- if(m_bReleased) return;
- //Release the memory for new files
- int i,j;
- for( i=0;i<CC_NUM;i++)
- {//delete the memory of word item array in the dictionary
- for( j=0;j<m_IndexTable[i].nCount;j++)
- {
- if(m_IndexTable[i].pWordItemHead[j].sWord!=NULL)
- {
- delete m_IndexTable[i].pWordItemHead[j].sWord;
- m_IndexTable[i].pWordItemHead[j].sWord=NULL;
- }
- }
- if(m_IndexTable[i].pWordItemHead!=NULL)
- {
- delete [] m_IndexTable[i].pWordItemHead;
- m_IndexTable[i].pWordItemHead=NULL;
- }
- }
- DelModified();
- m_bReleased=true;
- }