文本检索中若干问题研究

《文本检索中若干问题研究》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、Y弘6205密级：保密期限砖幸却童夫肇博士研究生学位论文学号：她§2塾姓名：王菱塑专业：篮曼兰篮皇熊理导师：整星熬援学院；篮息王捏堂院二零零六年五月十三日统滴水算法那样盲目选择正下方作为滴水方向，而是考虑水滴的先前方向以及待选择的切分方向与水滴自身尺寸间的关系。以手写数字为例进行的实验表明，该算法能够有效克服传统滴水算法进行字符切分时由于连笔或笔画边缘毛刺可能带来的误差，提高了算法的切分正确率：关键词：信息检索文本图像滴水算法文本分类特征选择查询优化查询扩展相关反馈互信息RSEARCHoNSEVERALPRoBLEMSINTEXTRETRJE、後LInfo

2、rIllationRe仃ievaltecbnology(IR)aimsatrecognizinga11dacquiringinfornlation疗Dmmesetofinfo肌ation，andplaysa11imporfalltr01einourstudyandsciemificrcs砌．Especiallyintoday，theIntemetisappliedmoreandmorewidely’and也equantityofinfb衄ationincreasessharply．In硒mationRetrievaltectl芏lologyhasbecom

3、eanefficient印proachforpeopletodevelopandmalceuseofallsonsofinfbmationresourcese仃ectively’toacqu沁aIldabsorbinf-0mationfleetlya11droundlyThercsearchofthepresenttheSisinvolvesinrelatedtechnologiesoninfo姗ationretrievalsuchasdocumentpmcessin舀teXtclassificationaIldqueryoptimizationetc．T

4、hefoll叭vingareac：hievedreSultsin也isdissertation：1．FeatureselectionintextclassificationInthemesis，weintroducetlleconceptsofabsolutereliability’relativereliabilityandcompositivercliabilityandsetforththefeatureselectionalgorithmbasedonmumalinfonnationreliability．ThealgO^thmcombinesth

5、econIelativitvbetweenatermandtheclassandthedif矗奠_enceonthete眦amongalltheclasses，i．e．，therelia_bilityofthemaxiummutualinfonnationamongclasses．ExperimentsshowthatcomDaredtothebasicmutua】infonnationmnction，thealgorithmbasedonmutuaJinfo啪ationreliabilitycanimpmvetheprecision，recallandF

6、lmeasuresefrectively．Furthermorc，wealso印plynomalizationtosevaltraditional向nctionsormaI∞10calfbatureselectionbasedonthese矗Inctions．ExDerimentsshowthatnonnalizedf．canlreselectionandlocalf色atureselectioncanimDrovetheclassificationprecisionmoreorless．2．MuticlassclassificationItiscommo

7、ntosetathresholdforeachclassinordertosettletheproblemthatatcxtmaybelOngtodi腩rentclasses．WhenthesimilarityofthetextandoneclassisdbOvetbethresholdofthisclass．th即thetextisclassifiedtothisclass．1nthismcsis，、veresearchonthedetenninationofthresh01d，putfonvardthethresholddetemlinationalg

8、orithmbasedonthemaximizedevaluati