• 微博
  • 微信微信二维码

广东省人民政府首页  >  要闻动态  >  广东要闻

雪球知股乎系列三:情绪因子选股正当时

来源: 南方日报网络版     时间: 2018年10月24日 14:09:18
【字体:

我的钱被新宝5黑了信誉平台官方,招龙虎和代理注册实力招商主管Q扣99277339,永久注册网址www.yicai269.com时时彩龙虎和,意彩彩票代理,意彩官网,意彩平台注册,意彩乐官网,意彩娱乐官方网站雪球知股乎系列三:情绪因子选股正当时

WTI高位"陨落" OPEC酝酿减产计划或延长9个月 Deltastock:欧元、日元、英镑交易策略 中国共产党第十九次全国代表大会在京开幕 6大货币对、美元指数及黄金阻力/支持位 美国房市数据虽疲软 但债市为美元指数保驾护航 CFTC的LabCFTC发布虚拟货币入门手册引文 詹姆斯这态度够坚决! 夺冠 我是认真的! EIA原油库存大降不低汽油库存增长 产量创纪录降幅也无能为力 杜兰特最后一投 折射勇士尴尬 求稳却忘了初心 哪些迹象表明NBA正处于鼎盛时期?收视率爆表,勇士统治联盟 蓝军后防大将或转会巴萨,他居然是在皇马出道的! 有一种人,一辈子都不会老 独家专访(上)|金软景:10岁立志当职业球员,今天冲着最好的结果来上海! 巴萨欲购蓝军铁卫,与阿拉巴组成左路双保险 周挺:3年前特别想帮阿尔滨保级

(原)(标)(题)(:雪)(球)(知)(股)(乎)(系)(列)(三)(:情)(绪)(因)(子)(选)(股)(正)(当)(时)(
报)(告)(要)(点)(
作)(为)(“猎)(金)(系)(列)(十)(六)(”,本)(报)(告)(研)(究)(雪)(球)(网)(的)(第)(三)(个)(数)(据)(源)(:非)(格)(式)(化)(的)(用)(户)(发)(帖)(文)(本)(。通)(过)(机)(器)(学)(习)(的)(方)(式)(对)(帖)(子)(进)(行)(情)(绪)(判)(断)(,构)(建)(情)(绪)(选)(股)(因)(子)(。
研)(究)(时)(窗)(为)(2014年)(1月)(-2017年)(7月)(,共)(计)(得)(到)(30,658,560篇)(帖)(子)(。对)(初)(始)(的)(帖)(子)(经)(过)(一)(系)(列)(处)(理)(得)(到)(降)(噪)(之)(后)(的)(信)(息)(。选)(择)(27,654篇)(帖)(子)(作)(为)(样)(本)(内)(数)(据)(,通)(过)(人)(工)(的)(方)(式)(对)(帖)(子)(打)(标)(签)(,确)(定)(情)(绪)((正)(向)(、负)(向)(、中)(性)(三)(种)(情)(绪)(),研)(究)(多)(种)(机)(器)(学)(习)(模)(型)(的)(效)(果)(,选)(择)(SVM作)(为)(最)(终)(的)(训)(练)(模)(型)(。模)(型)(样)(本)(内)(精)(准)(度)(达)(到)(70.16%,在)(正)(向)(情)(绪)(集)(合)(、负)(向)(情)(绪)(集)(合)(的)(召)(回)(率)(分)(别)(达)(到)(76.9%、68.2%。进)(一)(步)(判)(断)(样)(本)(外)(情)(绪)(,得)(到)(原)(始)(的)(情)(绪)(因)(子)(数)(据)(源)(。
构)(建)(正)(向)(情)(绪)(比)(、负)(向)(情)(绪)(比)(、综)(合)(情)(绪)(比)(因)(子)((负)(向)(情)(绪)(比)(即)(周)(期)(范)(围)(内)(负)(面)(帖)(子)(数)(/总)(帖)(子)(数)()。测)(试)(结)(果)(显)(示)(,负)(向)(情)(绪)(比)(因)(子)(具)(有)(较)(好)(的)(选)(股)(能)(力)(。整)(体)(来)(讲)(,负)(向)(情)(绪)(比)(越)(低)(,未)(来)(收)(益)(会)(越)(好)(,尤)(其)(是)(在)(500股)(票)(池)(内)(部)(选)(股)(能)(力)(突)(出)(。
将)(负)(向)(情)(绪)(比)(因)(子)(与)(反)(转)(因)(子)(结)(合)(,在)(中)(证)(500股)(票)(池)(内)(部)(构)(建)(双)(因)(子)(增)(强)(策)(略)(。双)(因)(子)(策)(略)(年)(化)(超)(额)(收)(益)(率)(、夏)(普)(率)(、最)(大)(回)(撤)(分)(别)(达)(到)(11.6%、1.87、6.5%,显)(著)(优)(于)(反)(转)(因)(子)(的)(相)(应)(表)(现)(。这)(意)(味)(着)(当)(期)(下)(跌)(较)(多)(但)(市)(场)(负)(面)(情)(绪)(没)(那)(么)(严)(重)(的)(股)(票)(将)(会)(有)(更)(高)(的)(超)(额)(收)(益)(。
另)(外)(从)(情)(绪)(角)(度)(出)(发)(也)(做)(了)(一)(些)(其)(他)(方)(面)(的)(选)(股)(因)(子)(的)(尝)(试)(,如)(根)(据)(综)(合)(情)(绪)(比)(因)(子)(构)(建)(的)(波)(动)(率)(因)(子)(在)(全)(市)(场)(具)(有)(不)(错)(的)(选)(股)(能)(力)(,月)(度)(因)(子)(值)(越)(大)(,未)(来)(收)(益)(越)(高)(。
1
背)(景)(回)(顾)(
1.1将)(不)(可)(思)(议)(进)(行)(到)(底)(
将)(各)(种)(奇)(奇)(怪)(怪)(的)(单)(位)(罗)(列)(一)(下)(,我)(们)(会)(发)(现)(有)(个)(很)(有)(意)(思)(的)(对)(象)(:不)(可)(思)(议)(单)(位)(。这)(个)(单)(位)(是)(描)(述)(人)(类)(觉)(得)(无)(法)(想)(象)(的)(事)(情)(如)(果)(真)(的)(发)(生)(大)(概)(需)(要)(的)(时)(间)(间)(隔)(。假)(设)(以)(下)(几)(个)(场)(景)(:
场)(景)(1:我)(穿)(越)(时)(空)(,来)(到)(春)(秋)(,对)(话)(老)(子)(:“老)(人)(家)(,我)(将)(您)(的)(道)(德)(经)(抄)(录)(在)(纸)(上)(以)(便)(万)(世)(传)(颂)(好)(不)(好)(?”。老)(人)(家)(沉)(默)(不)(语)(……他)(觉)(得)(不)(可)(思)(议)(,何)(谓)(纸)(?想)(法)(根)(本)(不)(可)(能)(实)(现)(,何)(谈)(万)(世)(传)(颂)(。
场)(景)(2:我)(不)(死)(心)(,赶)(到)(盛)(唐)(,请)(教)(玄)(奘)(:“圣)(僧)(,我)(将)(佛)(法)(抄)(录)(在)(纸)(上)(以)(便)(万)(世)(传)(颂)(好)(不)(好)(?”。玄)(奘)(微)(微)(一)(笑)(:“出)(家)(人)(不)(打)(诳)(语)(。”但)(他)(心)(里)(的)(小)(九)(九)(在)(说)(:没)(个)(几)(百)(年)(这)(不)(可)(能)(实)(现)(。
这)(里)(的)(“万)(世)(”、“几)(百)(年)(”就)(是)(不)(同)(时)(代)(的)(人)(在)(衡)(量)(同)(一)(件)(事)(情)(的)(不)(可)(思)(议)(单)(位)(。“不)(可)(思)(议)(”指)(标)(的)(缩)(短)(见)(证)(了)(人)(类)(的)(进)(步)(和)(科)(技)(的)(力)(量)(。目)(前)(最)(具)(有)(争)(议)(性)(的)(就)(是)(人)(工)(智)(能)(。人)(工)(智)(能)(已)(经)(深)(刻)(的)(改)(变)(了)(我)(们)(的)(生)(活)(方)(式)(、思)(维)(习)(惯)(。而)(量)(化)(投)(资)(领)(域)(是)(计)(算)(机)(最)(“钱)(”沿)(的)(地)(方)(,该)(如)(何)(应)(对)(这)(种)(冲)(击)(?
随)(着)(三)(张)(报)(表)(的)(研)(究)(效)(用)(曲)(线)(在)(逐)(步)(降)(低)(,大)(数)(据)(以)(及)(机)(器)(学)(习)(在)(量)(化)(投)(资)(领)(域)(的)(应)(用)(愈)(演)(愈)(烈)(。我)(们)(在)(《雪)(球)(知)(股)(乎)(系)(列)(之)(一)(:和)(关)(注)(度)(因)(子)(有)(个)(约)(定)(》、《雪)(球)(知)(股)(乎)(系)(列)(二)(:百)(万)(投)(资)(组)(合)(里)(的)(宝)(贝)(》详)(细)(的)(介)(绍)(了)(大)(数)(据)(、机)(器)(学)(习)(、人)(工)(智)(能)(的)(关)(系)(和)(发)(展)(。下)(面)(我)(们)(简)(单)(回)(顾)(一)(下)(前)(面)(两)(篇)(报)(告)(的)(一)(些)(关)(键)(性)(内)(容)(,方)(便)(后)(续)(的)(研)(究)(。
1.2雪)(球)(网)(前)(情)(回)(顾)(
雪)(球)(网)(有)(很)(多)(功)(能)(界)(面)(,我)(们)(通)(过)(爬)(虫)(的)(方)(式)(获)(取)(雪)(球)(网)(上)(用)(户)(UGC(用)(户)(自)(主)(生)(成)()的)(行)(为)(内)(容)(。雪)(球)(网)(用)(户)(行)(为)(共)(计)(有)(三)(种)(:关)(注)(行)(为)(、发)(帖)(行)(为)(、投)(资)(组)(合)(调)(仓)(行)(为)(。另)(外)(,发)(帖)(行)(为)(是)(一)(种)(非)(格)(式)(化)(文)(本)(,关)(注)(行)(为)(是)(一)(种)(格)(式)(化)(文)(本)(,包)(括)(买)(入)(、卖)(出)(、关)(注)(三)(种)(行)(为)(,其)(中)(关)(注)(行)(为)(占)(比)(95.5%。
数)(据)(的)(时)(间)(窗)(口)(为)(2010年)(12月)(16日)(至)(2017年)(7月)(21日)(,在)(此)(期)(间)(共)(获)(得)(4,254,659个)(用)(户)(、51,320,982条)(基)(本)(行)(为)(。
我)(们)(根)(据)(关)(注)(行)(为)(构)(建)(了)(关)(注)(度)(因)(子)(,该)(因)(子)(统)(计)(了)(时)(间)(周)(期)(范)(围)(内)(某)(个)(股)(票)(被)(关)(注)(的)(次)(数)(,结)(论)(有)(四)(:
1、关)(注)(度)(因)(子)(具)(有)(反)(转)(效)(应)(,当)(期)(关)(注)(度)(低)(的)(股)(票)(未)(来)((月)(度)(/周)(度)()会)(有)(溢)(价)(;
2、雪)(球)(网)(关)(注)(度)(因)(子)(与)(其)(他)(互)(联)(网)(文)(本)(关)(注)(度)(因)(子)((主)(流)(财)(经)(媒)(体)(网)(站)(、股)(吧)()或)(者)(类)(关)(注)(度)(因)(子)((同)(花)(顺)(点)(击)(率)(等)()的)(秩)(相)(关)(性)(在)(0.7-0.8左)(右)(,具)(有)(高)(同)(质)(性)(;
3、雪)(球)(网)(关)(注)(度)(因)(子)(换)(手)(率)(较)(高)(、具)(有)(一)(定)(的)(小)(市)(值)(效)(应)(;
4、关)(注)(度)(因)(子)(对)(于)(价)(值)(因)(子)(有)(一)(定)(的)(增)(强)(作)(用)(,估)(值)(低)(关)(注)(度)(低)(的)(股)(票)(未)(来)(具)(有)(更)(大)(的)(想)(象)(空)(间)(。
根)(据)(投)(资)(组)(合)(的)(调)(仓)(行)(为)(构)(建)(价)(值)(变)(动)(因)(子)(,该)(因)(子)(将)(百)(万)(投)(资)(组)(合)(合)(并)(成)(一)(个)(组)(合)(,构)(建)(月)(度)(市)(值)(变)(动)(率)(因)(子)(,进)(一)(步)(剥)(离)(反)(转)(因)(子)(,该)(因)(子)(具)(有)(反)(转)(效)(应)(:当)(期)(调)(入)(越)(多)(的)(标)(的)(未)(来)(下)(跌)(的)(越)(多)(,调)(出)(越)(多)(的)(标)(的)(未)(来)(上)(涨)(的)(越)(多)(。(具)(体)(参)(见)(往)(期)(报)(告)()。
而)(本)(篇)(是)(我)(们)(基)(于)(雪)(球)(网)(用)(户)(发)(帖)(行)(为)(推)(出)(的)(第)(三)(篇)(报)(告)(。研)(究)(基)(本)(框)(架)(如)(下)(:
1、对)(用)(户)(的)(发)(帖)(进)(行)(初)(始)(筛)(选)(,删)(除)(无)(效)(的)(帖)(子)(以)(及)(文)(本)(过)(长)(的)(帖)(子)(;
2、选)(择)(27,654篇)(帖)(子)(用)(于)(人)(工)(打)(标)(签)(,然)(后)(通)(过)(机)(器)(学)(习)(的)(方)(式)(对)(历)(史)(全)(样)(本)(进)(行)(判)(断)(,确)(定)(每)(篇)(帖)(子)(的)(情)(绪)(;
3、构)(建)(情)(绪)(因)(子)(用)(于)(选)(股)(;
4、基)(于)(有)(效)(的)(情)(绪)(因)(子)(构)(建)(选)(股)(策)(略)(。
2
情)(绪)(分)(析)(综)(述)(
2.1情)(绪)(分)(析)(的)(基)(本)(流)(程)(
情)(绪)(分)(析)(通)(过)(运)(用)(NLP、文)(本)(挖)(掘)(、计)(算)(语)(言)(学)(等)(技)(术)(实)(现)(对)(各)(种)(非)(规)(范)(化)(的)(文)(本)(语)(言)(处)(理)(,达)(到)(相)(应)(的)(情)(绪)(判)(断)(的)(目)(的)(。目)(前)(主)(流)(的)(研)(究)(方)(法)(分)(成)(无)(监)(督)(的)(文)(本)(情)(感)(判)(断)(和)(有)(监)(督)(的)(情)(感)(判)(断)(。
无)(监)(督)(的)(情)(感)(判)(断)(主)(要)(运)(用)(情)(感)(词)(的)(相)(关)(信)(息)(进)(行)(文)(本)(情)(绪)(的)(判)(断)(。该)(方)(法)(面)(临)(的)(问)(题)(是)(:1、需)(要)(动)(态)(维)(护)(情)(绪)(库)(;2、忽)(略)(了)(文)(本)(的)(结)(构)(问)(题)(,导)(致)(部)(分)(关)(键)(信)(息)(的)(丢)(失)(。有)(监)(督)(学)(习)(的)(文)(本)(情)(感)(分)(析)(主)(要)(运)(用)(贝)(叶)(斯)(模)(型)(、支)(持)(向)(量)(机)((SVM,SupportVectorMachine)、神)(经)(网)(络)(等)(模)(型)(进)(行)(情)(感)(分)(类)(。不)(同)(的)(学)(者)(对)(机)(器)(学)(习)(在)(情)(绪)(分)(析)(方)(面)(的)(研)(究)(进)(行)(了)(不)(同)(的)(尝)(试)(和)(改)(进)(。
这)(里)(需)(要)(强)(调)(一)(点)(,无)(论)(是)(通)(过)(词)(库)(方)(式)(判)(断)(文)(本)(情)(绪)(,亦)(或)(通)(过)(机)(器)(学)(习)(的)(方)(式)(判)(断)(文)(本)(的)(情)(绪)(,都)(需)(要)(通)(过)(切)(词)(的)(方)(式)(对)(文)(本)(进)(行)(碎)(片)(化)(、向)(量)(化)(处)(理)(。
2.2雪)(球)(网)(情)(绪)(判)(断)(
我)(们)(将)(每)(一)(篇)(文)(本)(的)(情)(绪)(分)(为)(3类)(:乐)(观)(、悲)(观)(、中)(性)(,采)(用)(机)(器)(学)(习)(的)(方)(式)(对)(文)(本)(情)(绪)(进)(行)(判)(断)(。可)(选)(择)(的)(机)(器)(学)(习)(模)(型)(很)(多)(,通)(过)(对)(比)(分)(析)(,我)(们)(选)(择)(SVM作)(为)(最)(终)(的)(机)(器)(学)(习)(模)(型)((后)(面)(会)(有)(具)(体)(的)(讨)(论)()。
整)(个)(文)(本)(处)(理)(的)(流)(程)(如)(下)(:
雪)(球)(网)(数)(据)(抓)(取)(在)(此)(就)(不)(做)(赘)(述)(,具)(体)(参)(见)(往)(期)(报)(告)(。我)(们)(接)(下)(来)(重)(点)(关)(注)(其)(他)(关)(键)(步)(骤)(。
1、删)(除)(无)(效)(文)(本)(:此)(次)(研)(究)(对)(象)(为)(雪)(球)(网)(用)(户)(发)(的)(非)(格)(式)(化)(文)(本)(。但)(有)(些)(行)(为)(可)(能)(仅)(仅)(是)(雪)(球)(网)(的)(一)(种)(推)(广)(行)(为)(或)(者)(无)(意)(义)(的)(格)(式)(文)(本)(,鉴)(于)(这)(些)(文)(本)(对)(最)(终)(的)(情)(绪)(分)(析)(没)(有)(作)(用)(,所)(以)(删)(除)(该)(类)(文)(本)(。
2、删)(除)(过)(长)(文)(本)(:部)(分)(文)(本)(内)(容)(过)(长)(,一)(方)(面)(增)(加)(了)(情)(绪)(判)(断)(的)(难)(度)(:当)(文)(本)(过)(长)(时)(会)(对)(切)(词)(、情)(绪)(分)(析)(、包)(括)(最)(终)(的)(词)(向)(量)(维)(度)(产)(生)(较)(大)(的)(影)(响)(;另)(一)(方)(面)(极)(大)(的)(降)(低)(学)(习)(的)(效)(率)(。我)(们)(统)(计)(所)(有)(样)(本)(的)(字)(符)(长)(度)(,发)(现)(文)(本)(长)(度)(在)(700以)(内)(的)(样)(本)(数)(占)(比)(为)(97.5%。鉴)(于)(此)(,我)(们)(最)(终)(只)(保)(留)(文)(本)(长)(度)(=700的)(样)(本)(作)(为)(情)(绪)(分)(析)(对)(象)(。
3、文)(本)(清)(洗)(:原)(始)(的)(文)(本)(包)(含)(大)(量)(的)(噪)(音)(或)(者)(对)(情)(绪)(判)(断)(无)(效)(的)(信)(息)(,因)(此)(需)(要)(对)(初)(始)(的)(文)(本)(内)(容)(进)(行)(清)(洗)(,一)(方)(面)(提)(升)(文)(本)(质)(量)(、降)(低)(词)(向)(量)(空)(间)(,另)(一)(方)(面)(提)(升)(运)(算)(效)(率)(。针)(对)(于)(初)(始)(文)(本)(的)(清)(洗)(维)(度)(大)(致)(如)(下)(:
4、文)(本)(切)(分)(:文)(本)(清)(洗)(后)(,接)(下)(来)(需)(要)(将)(其)(分)(成)(样)(本)(内)(、样)(本)(外)(数)(据)(,同)(时)(针)(对)(于)(样)(本)(内)(的)(数)(据)(需)(要)(进)(一)(步)(切)(分)(成)(训)(练)(集)(和)(测)(试)(集)(。保)(证)(随)(机)(选)(择)(的)(情)(况)(下)(切)(分)(的)(原)(则)(是)(:确)(保)(时)(间)(维)(度)(覆)(盖)(度)(、A股)(覆)(盖)(度)(。我)(们)(抽)(取)(了)(27,654篇)(帖)(子)(作)(为)(样)(本)(内)(数)(据)(。进)(一)(步)(将)(样)(本)(内)(数)(据)(分)(成)(训)(练)(集)(和)(测)(试)(集)(,比)(例)(为)(70%:30%(训)(练)(集)(:测)(试)(集)()。这)(样)(就)(得)(到)(了)(19,332篇)(幅)(的)(训)(练)(集)(合)(、8,322篇)(幅)(的)(测)(试)(集)(。样)(本)(内)(正)(向)(情)(绪)(的)(文)(本)(占)(比)(48.07%、负)(向)(情)(绪)(的)(占)(比)(29.49%、中)(性)(情)(绪)(的)(文)(本)(占)(比)(22.45%。这)(个)(比)(例)(与)(目)(前)(市)(场)(的)(现)(象)(比)(较)(一)(致)(:国)(内)(市)(场)(普)(遍)(存)(在)(着)(报)(喜)(不)(报)(忧)(的)(现)(象)(,而)(散)(户)(投)(资)(者)(的)(情)(绪)(也)(存)(在)(着)(类)(似)(的)(比)(例)(分)(配)(情)(况)(。
5、Prep:接)(下)(来)(重)(点)(讲)(述)(Prep板)(块)(的)(流)(程)(。该)(部)(分)(主)(要)(讲)(述)(了)(如)(何)(切)(词)(、以)(及)(如)(何)(将)(切)(词)(以)(后)(的)(结)(果)(转)(化)(成)(模)(型)((贝)(叶)(斯)(、SVM等)()的)(输)(入)(。主)(要)(步)(骤)(包)(括)(切)(词)(、否)(定)(词)(、词)(性)(的)(应)(用)(、词)(向)(量)(空)(间)(生)(成)(等)(几)(个)(关)(键)(步)(骤)(。
5.1、切)(词)(:这)(一)(步)(的)(主)(要)(目)(的)(是)(对)(文)(本)(进)(行)(切)(词)(,碎)(片)(化)(文)(本)(后)(,作)(为)(进)(一)(步)(研)(究)(的)(对)(象)(。而)(切)(词)(的)(依)(据)(是)(词)(库)(。我)(们)(这)(里)(利)(用)(的)(是)(结)(巴)(分)(词)(系)(统)(。同)(时)(为)(了)(完)(善)(词)(库)(,我)(们)(通)(过)(自)(定)(义)(的)(方)(式)(导)(入)(搜)(狗)(金)(融)(词)(库)(,构)(成)(原)(始)(的)(词)(库)(表)(。
5.2、否)(定)(词)(:否)(定)(词)(的)(使)(用)(会)(导)(致)(语)(义)(反)(转)(。比)(如)(“不)(/太)(平)(”、“不)(/开)(心)(”、“并)(非)(/如)(此)(”、“从)(未)(/快)(乐)(”等)(。根)(据)(切)(词)(的)(结)(果)(会)(切)(出)(正)(向)(词)(汇)(,但)(实)(际)(上)(上)(面)(的)(表)(述)(本)(身)(传)(达)(的)(是)(一)(种)(否)(定)(的)(意)(义)(。需)(要)(对)(否)(定)(句)(进)(行)(特)(殊)(处)(理)(,如)(句)(子)(”Ididn’tlikethismovie”。部)(分)(学)(者)(提)(出)(“一)(旦)(出)(现)(否)(定)(词)(,那)(么)(在)(否)(定)(词)(之)(后)(、符)(号)(之)(前)(的)(所)(有)(切)(词)(结)(果)(都)(加)(上)(否)(定)(词)(,进)(而)(形)(成)(新)(的)(词)(汇)(”,根)(据)(此)(规)(则)(可)(以)(将)(句)(子)(“didn’tlikethismovie,butI”转)(换)(为)(“didn’tNOT_likeNOT_thisNOT_movie,butI”。
5.3、停)(用)(词)(:停)(用)(词)((stopwords)是)(指)(评)(论)(文)(本)(数)(量)(较)(多)(,但)(对)(文)(本)(的)(情)(绪)(倾)(向)(分)(析)(没)(有)(帮)(助)(的)(词)(语)(,例)(如)(代)(词)(“这)(”、“那)(”、“他)(”,副)(词)(“也)(”、“刚)(刚)(”,其)(他)(词)(汇)(如)(“而)(”、“及)(”、“与)(”、“或)(”等)(。根)(据)(停)(用)(词)(集)(合)(构)(建)(的)(停)(用)(词)(表)(删)(除)(停)(用)(词)(,能)(够)(减)(少)(数)(据)(噪)(音)(,提)(升)(文)(本)(情)(感)(分)(类)(的)(准)(确)(率)(。
部)(分)(停)(用)(词)(样)(本)(如)(下)(:[和)(,是)(,就)(,都)(,而)(,及)(,,著)(,或)(,一)(个)(,我)(们)(,你)(们)(,她)(们)(,他)(们)(]。这)(里)(需)(要)(注)(意)(的)(是)(需)(要)(将)(刚)(刚)(“文)(本)(清)(洗)(”步)(骤)(产)(生)(的)(一)(些)(词)(汇)(,如)(“STOCK_URL_”、“USER_URL_”等)(添)(加)(到)(停)(用)(词)(里)(面)(。
5.4、词)(性)(标)(注)(:通)(常)(而)(言)(,表)(达)(情)(感)(的)(词)(汇)(往)(往)(是)(形)(容)(词)(,此)(外)(动)(词)(、名)(词)(等)(也)(有)(类)(似)(的)(作)(用)(。但)(是)(数)(词)(、量)(词)(、介)(词)(等)(词)(汇)(可)(能)(并)(无)(太)(大)(的)(作)(用)(;另)(外)(如)(果)(不)(做)(词)(性)(的)(限)(制)(,最)(终)(词)(向)(量)(空)(间)(可)(能)(会)(比)(较)(大)(,也)(会)(影)(响)(运)(算)(效)(率)(。因)(此)(我)(们)(将)(仅)(仅)(考)(虑)(形)(容)(词)(、动)(词)(、名)(词)(等)(功)(能)(实)(词)(作)(为)(特)(征)(。
5.5、词)(向)(量)(:这)(一)(步)(主)(要)(说)(明)(的)(是)(模)(型)(的)(输)(入)(,即)(词)(向)(量)(生)(成)(问)(题)(。根)(据)(前)(面)(的)(文)(本)(切)(分)(、我)(们)(得)(到)(了)(每)(个)(文)(本)(的)(分)(词)(结)(果)((每)(一)(个)(文)(本)(是)(一)(个)(词)(向)(量)(空)(间)(,1*N维)(度)(,这)(里)(的)(N是)(每)(个)(文)(本)(切)(分)(后)(的)(词)(数)(目)();然)(后)(通)(过)(否)(定)(词)(处)(理)(将)(每)(句)(否)(定)(词)(位)(置)(以)(后)(的)(词)(语)(进)(行)(否)(定)(化)(处)(理)(;进)(一)(步)(丢)(弃)(没)(有)(意)(义)(的)(词)(汇)(,其)(中)(没)(有)(意)(义)(的)(词)(汇)(处)(理)(流)(程)(包)(括)(:停)(用)(词)(的)(处)(理)(以)(及)(词)(性)(的)(处)(理)(。这)(样)(我)(们)(就)(能)(得)(到)(每)(个)(文)(本)(处)(理)(完)(毕)(的)(词)(向)(量)((1*N’维)(,这)(里)(N’=N)。然)(后)(取)(所)(有)(文)(本)(切)(词)(后)(的)(词)(量)(并)(集)(,生)(成)(词)(向)(量)(全)(集)((1*M),进)(一)(步)(生)(成)(K*M(K个)(样)(本)()维)(度)(的)(向)(量)(空)(间)(。
下)(面)(以)(两)(个)(文)(本)(为)(例)(,将)(上)(面)(的)(过)(程)(进)(行)(简)(单)(的)(梳)(理)(。
对)(上)(面)(切)(词)(后)(的)(结)(果)(进)(行)(“Prep”阶)(段)(处)(理)(。1、切)(词)(,切)(词)(后)(结)(果)(参)(见)(上)(述)(表)(格)(;2、将)(例)(句)(1中)(的)(否)(定)(词)(“没)(”添)(加)(到)(后)(面)(的)(“离)(开)(”(否)(定)(词)(以)(后)(,符)(号)(之)(前)(的)(所)(有)(分)(词)(结)(果)()上)(形)(成)(新)(词)(汇)(“没)(离)(开)(”;3、将)(例)(句)(2中)(的)(“和)(”这)(个)(停)(用)(词)(去)(掉)(;4、然)(后)(取)(所)(有)(例)(句)(的)(词)(全)(集)([大)(部)(分)(、员)(工)(、没)(、没)(离)(开)(、女)(孩)(、非)(常)(、爱)(、吃)(、Ginger、甜)(品)(]。依)(照)(词)(向)(量)(全)(集)(对)(每)(个)(例)(句)(切)(词)(后)(的)(词)(是)(否)(出)(现)(在)(全)(集)(中)(做)(0/1判)(断)(处)(理)((0表)(示)(没)(有)(出)(现)(、1表)(示)(出)(现)(),处)(理)(完)(毕)(后)(结)(果)(如)(下)(:
然)(后)(根)(据)(每)(个)(文)(本)(词)(向)(量)(化)(后)(的)(结)(果)(,结)(合)(情)(绪)(的)(帖)(子)(得)(到)(模)(型)(的)(输)(入)(,进)(行)(训)(练)(。
6、SVM模)(型)(:前)(面)(在)(情)(绪)(判)(断)(的)(研)(究)(中)(提)(及)(,可)(以)(运)(用)(SVM、贝)(叶)(斯)(、神)(经)(网)(络)(等)(多)(种)(方)(法)(对)(文)(本)(情)(绪)(进)(行)(判)(断)(,在)(效)(率)(和)(效)(果)(方)(面)(性)(价)(比)(较)(高)(的)(是)(SVM模)(型)(。我)(们)(这)(里)(以)(SVM和)(贝)(叶)(斯)(模)(型)(的)(对)(比)(分)(析)(为)(例)(进)(行)(说)(明)(。
选)(择)(N个)(情)(绪)(文)(本)(作)(为)(测)(试)(对)(象)((训)(练)(集)(:测)(试)(集)(=70%:30%),分)(别)(测)(试)(N=6,000、8,000、10,000时)(,相)(应)(的)(模)(型)(表)(现)(。从)(测)(试)(结)(果)(来)(看)(:1、同)(样)(本)(下)(SVM要)(优)(于)(贝)(叶)(斯)(的)(表)(现)(;2、SVM模)(型)(对)(N的)(敏)(感)(性)(要)(优)(于)(贝)(叶)(斯)(模)(型)(,随)(着)(N的)(增)(大)(,SVM效)(果)(提)(升)(的)(更)(快)(;3、时)(间)(效)(率)(方)(面)(,SVM也)(显)(著)(的)(、稳)(定)(的)(优)(于)(贝)(叶)(斯)(表)(现)(。因)(此)(后)(面)(的)(分)(析)(中)(将)(运)(用)(SVM作)(为)(情)(绪)(判)(断)(的)(训)(练)(模)(型)(。
7、训)(练)(效)(果)(:通)(过)(以)(上)(的)(整)(体)(操)(作)(,我)(们)(得)(到)(了)(模)(型)(输)(入)(,运)(用)(SVM模)(型)(进)(行)(训)(练)(,得)(到)(模)(型)(的)(训)(练)(结)(果)(,我)(们)(从)(精)(准)(度)(和)(召)(回)(率)(两)(个)(维)(度)(考)(察)(效)(果)((根)(据)(前)(面)(的)(测)(试)(集)(合)(和)(训)(练)(集)(合)(的)(比)(例)(,我)(们)(在)(27,654个)(样)(本)(中)(共)(计)(得)(到)(8,322个)(测)(试)(文)(本)()。在)(测)(试)(集)(上)(,模)(型)(的)(精)(准)(度)(为)(70.16%。而)(从)(召)(回)(率)(的)(角)(度)(来)(看)(,正)(向)(、负)(向)(、中)(性)(文)(本)(的)(召)(回)(率)(分)(别)(为)(76.9%、68.2%、58.3%。
另)(外)(需)(要)(注)(意)(的)(是)(,正)(如)(前)(面)(在)(对)(比)(SVM和)(贝)(叶)(斯)(模)(型)(对)(测)(试)(样)(本)(数)(目)(的)(敏)(感)(性)(分)(析)(一)(样)(,我)(们)(分)(别)(测)(试)(了)(N=6,000、8,000、10,000、15,000、20,000、27,654个)(样)(本)(时)(,SVM模)(型)(的)(表)(现)(。随)(着)(N的)(增)(大)(,模)(型)(效)(果)(在)(不)(断)(提)(升)(,但)(考)(虑)(到)(打)(标)(签)(的)(人)(工)(成)(本)(,我)(们)(最)(终)(打)(了)(27,654个)(样)(本)(。
3
情)(绪)(因)(子)(构)(建)(
3.1因)(子)(逻)(辑)(
基)(于)(前)(面)(的)(整)(体)(研)(究)(,我)(们)(得)(到)(了)(情)(绪)(因)(子)(的)(初)(始)(数)(据)(源)(,进)(一)(步)(构)(建)(情)(绪)(因)(子)(。每)(篇)(帖)(子)(的)(情)(绪)(有)(三)(种)(可)(能)(:正)(向)(情)(绪)(、负)(向)(情)(绪)(、中)(性)(情)(绪)(。那)(么)(,我)(们)(分)(别)(定)(义)(了)(以)(下)(几)(种)(类)(别)(的)(情)(绪)(因)(子)(。
从)(占)(比)(的)(角)(度)(来)(讲)(,我)(们)(分)(别)(构)(建)(了)(正)(向)(情)(绪)(占)(比)(、负)(面)(情)(绪)(占)(比)(、综)(合)(情)(绪)(比)(三)(个)(因)(子)(。因)(子)(之)(间)(可)(能)(具)(有)(一)(定)(的)(相)(关)(性)(,但)(各)(因)(子)(反)(应)(的)(信)(息)(量)(不)(一)(样)(。
这)(里)(,我)(们)(研)(究)(了)(三)(个)(情)(绪)(因)(子)(的)(截)(面)(均)(值)(历)(史)(趋)(势)(图)(。以)(负)(面)(情)(绪)(比)(为)(例)(进)(行)(说)(明)((正)(面)(情)(绪)(比)(与)(其)(正)(好)(相)(反)(),14年)(至)(15年)(7月)(呈)(现)(单)(调)(递)(增)(的)(趋)(势)(,尤)(其)(是)(15年)(前)(半)(年)(,当)(时)(行)(情)(暴)(涨)(,但)(是)(从)(负)(面)(情)(绪)(比)(指)(标)(来)(看)(市)(场)(表)(现)(一)(定)(的)(恐)(慌)(情)(绪)(;15年)(8月)(至)(今)(呈)(现)(单)(调)(递)(减)(的)(趋)(势)(,今)(年)(以)(来)(有)(一)(定)(波)(动)(。
下)(面)(所)(有)(的)(测)(试)(都)(是)(在)(以)(下)(条)(件)(基)(础)(之)(上)(进)(行)(:
个)(股)(数)(据)(:全)(市)(场)(选)(股)(,剔)(除)(不)(满)(足)(交)(易)(条)(件)(的)(个)(股)(;
行)(业)(分)(类)(:中)(信)(一)(级)(行)(业)(;
数)(据)(区)(间)(:2014年)(1月)(1日)(至)(2017年)(7月)(21日)(;
3.2覆)(盖)(度)(分)(析)(
首)(先)(研)(究)(不)(同)(类)(别)(的)(月)(度)(情)(绪)(因)(子)(对)(A股)(的)(覆)(盖)(度)(。从)(结)(果)(来)(看)(,正)(向)(情)(绪)(比)(因)(子)(的)(覆)(盖)(度)(较)(高)(,覆)(盖)(度)(均)(值)(达)(到)(93.9%,尤)(其)(是)(2014年)(以)(后)(的)(覆)(盖)(度)(已)(经)(趋)(于)(全)(市)(场)(覆)(盖)(;而)(负)(面)(情)(绪)(覆)(盖)(度)(相)(对)(较)(低)(,但)(是)(月)(度)(因)(子)(的)(覆)(盖)(度)(均)(值)(已)(经)(达)(到)(83.8%。
3.3不)(同)(时)(间)(频)(度)(的)(对)(比)(分)(析)(
构)(建)(周)(度)(以)(及)(月)(度)(频)(度)(的)(上)(述)(因)(子)(,从)(IC以)(及)(分)(位)(数)(组)(合)(角)(度)(测)(试)(因)(子)(在)(全)(市)(场)(的)(有)(效)(性)(。
从)(检)(测)(结)(果)(来)(看)(:1、从)(频)(度)(角)(度)(分)(析)(,周)(度)(频)(度)(的)(因)(子)(几)(乎)(全)(部)(失)(效)(,没)(有)(选)(股)(的)(预)(测)(作)(用)(;2、从)(月)(度)(因)(子)(具)(体)(分)(析)(,负)(面)(情)(绪)(比)(因)(子)(有)(着)(较)(强)(的)(选)(股)(能)(力)(,负)(面)(情)(绪)(比)(越)(低)(意)(味)(着)(未)(来)(标)(的)(表)(现)(越)(好)(。
正)(面)(情)(绪)(因)(子)(失)(效)(而)(负)(面)(情)(绪)(有)(一)(定)(的)(预)(测)(作)(用)(这)(种)(现)(象)(也)(比)(较)(符)(合)(实)(际)(。国)(内)(股)(市)(一)(般)(是)(报)(喜)(不)(报)(忧)(,正)(面)(情)(绪)(存)(在)(着)(大)(量)(的)(噪)(音)(,很)(难)(从)(中)(找)(到)(有)(效)(的)(选)(股)(信)(息)(。而)(一)(旦)(股)(票)(形)(成)(负)(面)(情)(绪)(,那)(市)(场)(对)(股)(票)(的)(认)(可)(方)(面)(可)(能)(进)(入)(一)(个)(低)(谷)(,在)(悲)(观)(里)(面)(进)(行)(切)(分)(,悲)(观)(情)(绪)(较)(重)(的)(股)(票)(可)(能)(面)(临)(着)(更)(大)(的)(下)(行)(空)(间)(。
对)(悲)(观)(情)(绪)(比)(因)(子)(进)(行)(行)(业)(市)(值)(中)(性)(化)(,并)(进)(一)(步)(观)(察)(中)(性)(化)(之)(后)(的)(因)(子)(与)(价)(值)(、成)(长)(、反)(转)(、分)(析)(师)(情)(绪)(、换)(手)(率)(等)(因)(子)(的)(相)(关)(性)(。结)(果)(发)(现)(:中)(性)(化)(以)(后)(因)(子)(选)(股)(能)(力)(有)(所)(降)(低)(;另)(外)(,因)(子)(与)(传)(统)(的)(选)(股)(因)(子)(相)(关)(性)(较)(低)(。
虽)(然)(中)(性)(化)(以)(后)(的)(负)(面)(情)(绪)(比)(因)(子)(有)(效)(性)(有)(所)(降)(低)(,但)(在)(不)(同)(指)(数)(股)(票)(池)(上)(可)(能)(会)(有)(另)(一)(番)(惊)(喜)(,所)(以)(接)(下)(来)(测)(试)(中)(性)(化)(负)(面)(情)(绪)(比)(因)(子)(在)(中)(证)(500股)(票)(池)(内)(部)(的)(选)(股)(效)(果)(。
3.4悲)(观)(情)(绪)(比)(因)(子)(在)(中)(证)(500中)(有)(效)(性)(测)(试)(
验)(证)(中)(性)(化)(悲)(观)(情)(绪)(比)(因)(子)(在)(中)(证)(500股)(票)(池)(内)(部)(的)(表)(现)(。测)(试)(结)(果)(显)(示)(,在)(中)(证)(500股)(票)(池)(内)(部)(有)(较)(为)(突)(出)(的)(选)(股)(能)(力)(,IC达)(到)(0.022,IC_IR达)(到)(0.4。而)(观)(察)(分)(位)(数)(组)(合)(测)(试)(的)(结)(果)(发)(现)(悲)(观)(情)(绪)(比)(与)(未)(来)(收)(益)(率)(呈)(现)(负)(相)(关)(关)(系)(,悲)(观)(情)(绪)(比)(值)(越)(小)(,未)(来)(表)(现)(会)(越)(好)(。top组)(别)((悲)(观)(情)(绪)(比)(值)(最)(低)(的)()年)(化)(收)(益)(率)(为)(21.7%,夏)(普)(率)(为)(0.65。
后)(面)(如)(无)(特)(别)(强)(调)(,悲)(观)(情)(绪)(比)(因)(子)(均)(是)(行)(业)(市)(值)(中)(性)(化)(以)(后)(的)(结)(果)(。
)


相关文章

版权所有:南方新闻网 粤ICP备05070829 网站标识码4400000131
主办:南方新闻网 协办:广东省经济和信息化委员会 承办:南方新闻网
建议使用1024×768分辨率 IE7.0以上版本浏览器