貝殼基於事理圖譜的應用與實踐

{"type":"doc","content":[{"type":"blockquote","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"大家好，我是貝殼找房的孫拔羣，今天給大家帶來的是基於事理圖譜的智能培訓，我將從以下五個方面來跟大家進行分享。"}]},{"type":"bulletedlist","content":[{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"貝殼找房實體圖譜進展"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"貝殼找房事理圖譜進展"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"知識圖譜在智能培訓的應用"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"知識運營閉環的建立"}]}]},{"type":"listitem","attrs":{"listStyle":null},"content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"貝殼找房行業圖譜未來思考"}]}]}]}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"貝殼找房實體圖譜進展"}]},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/57\/57662585468393e3f3722152e0b36811.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"首先，我們團隊希望通過知識引入、知識組織、知識應用和知識運營這四個步驟成爲貝殼的知識中臺，進而成爲房產行業知識標準的制定者，通過知識的供給爲業務賦能。2018年我們面臨這樣一個問題：鏈家向貝殼找房這樣一個平臺化演進，我們有很多數據上的問題回答不了。所以我們希望通過大量的知識引入，能夠讓我們更加深入理解行業的全貌。所以我們設定了這樣一個目標：通過引入的更新覆蓋主流的房地產業務，包含一些重點實體信息，並對引入的數據進行知識分層，經過清洗融合挖掘之後，提供有價值的業務數據，形成我們的情報系統。"}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"1. 知識引入"}]},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/13\/1318cab02058cbe8948f3a1a8ceeb9cf.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我們需要對所有的數據進行三個級別的知識融合：第一級別是獲取跟外界交換過來的基礎數據；第二級別是進行一些簡單的清洗融合；第三級別是根據業務對數據進行非常深層次的挖掘。隨着級別的加深，數據的量級會越來越小，但是每個單條的數據價值會更高，這樣就可以通過這些數據形成一個情報系統。我們將這樣的情報系統分爲三個部分：提效、增量和決策，分別應對了我們三種主要的業務需求場景。兩年以來，我們的數據倉庫累計支持了大概10條線，24箇中心和50多個部門。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我們獲取了大量知識，會面臨各種各樣的不同訴求：不同來源不同類型的數據怎麼統一管理；數據和知識價值的深層次挖掘怎麼實現；知識來自方方面面，怎樣從中發現和獲取種意外之喜。面對這些訴求，我們選擇知識圖譜作爲基礎數據業務應用的橋樑，希望通過知識圖譜，能夠將所有的知識進行一個合理的組織。"}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"2. 知識組織"}]},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/cd\/cd8ff70b5f267ff325cf2cd79afad870.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"經過兩年的建設，我們將公司所有的知識分爲五個方向：第一是實體基礎知識，比如說房源小區類的知識；第二是作業規範知識，比如說二手和新房的帶看等線上作業規範；第三是公司規則知識，比如鏈家經典的ACN經紀人協作網絡，基本上顛覆了經紀人合作模式；第四是政策解讀知識，比如哪裏要修地鐵，哪裏的小學變更入學政策，這些都算政策解讀知識；第五是以上所有知識的關聯性。形成了一個包含60多個實體類型，總計規模在400多億，接近500億量級的知識圖譜。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/d6\/d681984e876be45a2306a59b7986500f.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/a0\/a0f864b8afa2b80a9cb96debabfef3df.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"但是知識圖譜的質量是不容易評估的，通過不斷的去優化、去集中、去聚焦所有的知識含量，把這個規模變小了，但體驗反而更好。我們建立了一個實體建模基礎工具及一些其他的配套工具，能夠極快的加速所有的知識庫建設過程。2020年7月份，中國電子技術標準化研究院邀請貝殼一起進行知識圖譜國標的建設。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/fc\/fc1f796aa437c6bde15d95271f4d7ee3.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"貝殼找房事理圖譜進展"}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"1. 需求"}]},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/d3\/d3385e4c9ea82d293b5d52d94d404664.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我們的事理圖譜從2019年開始做，但是對於一些深交互的問題，通過規則是不能很好的處理。舉個例子：某小區附近幼兒園的屬性和入學條件，在第一層次上的回答是：小區附近向東走500米，有個開心幼兒園，走路五分鐘即可到達，這是實體圖譜直接可以解答的。但是用戶對於每一套房的每一個居住屬性，他的訴求都非常的深入，他想要的結果不僅僅是這些，比如說再加深一層：幼兒園是什麼屬性？是一級一類公立幼兒園。我們需要對公立幼兒園代表的含義進行解讀，然後又引申出普惠幼兒園和非普惠幼兒園有什麼樣的優缺點，進而明確客戶買這套房之後，憑藉怎樣的條件能夠入學，這直接影響了客戶能否進行購買。客戶購買之後，讀了幼兒園之後和直升小學的規則，這個問題越來越深，已經不能單純依靠基於這套房或者基於個人的一些知識就能夠完成，所以在這個層面上，一方面對實體圖譜的質量要求越來越高，另一方面需要有事理解析的能力。針對於這種場景，我們通過重點業務去優化若干個實體子圖，再去新建幾個事理子圖，最後通過運營機制去保障數據質量。"}]},{"type":"heading","attrs":{"align":null,"level":3},"content":[{"type":"text","text":"2. 方法"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"對於如何實現，我們的方法設定是：通過一套構建工具去復現schema和data的集合，然後放在業務中去應用，呈現出一些知識算法，利用數據質量的評測體系，可以不斷的去形成數據的質量提升閉環。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/c7\/c777b0183fc640bf267dcd98d4f046c2.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"舉個例子（如下圖所示），在事理圖譜中，對於訴求，很多時候是人工拆解，所以我們設立了這樣一個工具，它能夠實現通過左邊純文本的方式轉換成右邊有點像決策樹或者事理子圖的方式。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/15\/15283183e8c518cae9febea60deb255c.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/41\/410a91ceb2d936c2b001e25adecd9d76.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"剛纔介紹了學區的例子，現在來介紹在稅費計算上的應用。在購房的時候一般接觸的是契稅是多少、有沒有增值稅等，實際上國家對於房產交易的稅費，有很多細節，只是一般人接觸不到。但經紀人在爲客戶服務時，對估算稅費的準確率是有嚴格要求的，如果估算錯，差額震盪超過百分之幾，就要全額賠付客戶。所以對於我們來說挑戰比較大，針對這一訴求，我們進行了專項的事理圖譜的推進。我們在建設事理圖譜時，不只建設事理圖譜，還將事理圖譜和實體圖譜相結合，並且不斷地下鑽。舉一個例子（如下圖），這裏面涉及到4個稅種，個人所得稅、契稅、增值稅和增值稅附加，每個稅種在開始計算的時候，通過第一層級的子圖推演，對於每一個綠色標籤，可以繼續向下推演它的規則是什麼樣，這個人符合什麼樣的一個前進路徑。比如對於個人所得稅契稅，可以繼續向下推演，可以生成一個事理子圖，生成事理子圖之後，我們可以將所有規則去掛靠到每一套房源，然後去映射成幾種用戶類型或者幾種業主類型，這樣就可以將事理圖譜的結果直接映射成最容易使用的3元組釋放到KBQA或者是其他的場景裏面去使用，可以極大的方便作業過程，也能夠提升準確率。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/9e\/9ed88a2f1b3f75c8c92e8c1c3c57d721.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/a6\/a62721025cb6f3c568c496adfb4ae377.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"這個工作我們持續了4個月左右，學區事理子圖現在已經可以覆蓋北京市140多個學區的規則、90%的社區和80%的房源；稅費事理子圖現在已經覆蓋全國17個城市的稅費規則、100%的房源，經專家測算準確率基本上達到100%。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/e4\/e484a7b471fc61980862d6b3f4b143c9.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"知識圖譜在智能培訓的應用"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"前面都是圖譜的構建過程，在做好這樣一個知識圖譜之後，我們希望它能夠在一些應用上落地，只有這樣纔可以去評估圖譜的質量，知道里面的數據價值，同時也能知道該如何去優化，使其變得更好。知識圖譜屬於基礎建設，一般都是螺旋迭代式發展，有時候需要獨立的去做這種平臺式的，有時候又需要強業務去牽引去提升自己的整體質量。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/0b\/0bba46af9ba5f373ef1eb907737a41e8.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在2020年下半年階段，就屬於需要業務去牽引強化內容的階段，所以我們做了很多業務上的應用，然後去發現數據質量效果。舉幾個例子（如下圖），這些都是線上去作業的經紀人常用的一些知識層面的訴求，在這裏面會思考一個問題，在做知識圖譜的時候，它每一個業務都需要知識，但是知識圖譜真正適用於什麼樣的需求，或者說哪一個業務真正的需要用知識圖譜去提升自己的能力，所以我們在不斷地尋求類似的應用訴求。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/d8\/d89d2e9d9ff816363e5a8ed1d59b4865.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/40\/40c96e95ff23c2dfa0ec6a522fd10cdd.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2020年初，我們接到智能培訓的業務需求。ToB、ToC服務類公司，對培訓的訴求非常高，過去都是純線下的培訓，我們統計過貝殼現有435門賦能培訓，其中線下培訓有374門，線上都集中在上面的那種規則類的培訓。線下的人對人培訓效率非常低，對專家的能力要求也非常高，所以說差異性非常高，導致線下培訓不太好做。在這個過程中，我們思考行業圖譜可以提供什麼，第一可以統一的提升消除各種因素造成的水平不均衡；第二可以提供標準的評價方式，提供一致性全維度的作業評價能力；第三可以規範化作業流程；第四可以積累很多的作業優秀物料，可以在未來的作業過程中直接使用。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/7a\/7a4271c82d640d8d171cb2ded0df5b3b.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"2020年2月份我們開始啓動基於行業圖譜智能培訓業務，智能培訓一共經歷了4個階段，第一個階段是產品化，通過行業知識圖譜的講盤知識和標準評價能力去產品化落地到講盤訓練場，講盤訓練場通過數據的反饋和提供的優質內容，爲行業知識圖譜做加強。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"舉個例子，如何去加強行業知識圖譜，知識圖譜落地到講盤訓練場應該都能理解如何去做，那麼如何反饋呢？在做重的ToB業務的時候，會發現定點的知識很難去系統性的收集，比如各種樓盤的知識，總會有缺少，有的地方特別稠密，有的地方特別稀疏。在培訓過程中要求經紀人要想參與培訓，就要貢獻這樣的資料，經過兩個星期的迭代，基本上收集了北京市7000多個小區的所有數據，而且數據質量非常高。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/86\/86aabd130db173d7db21cba4eda73526.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第二個階段是平臺化，我們發現不僅在二手講盤有培訓的訴求，在新房講盤、招聘講盤甚至業主面訪都有非常相似的訴求，這些訴求對於技術上的挑戰並不大，基本上都是一樣的，就是提供一個簡單的問答，加一個評價就足夠了，經紀人也非常喜歡這種培訓過程快，反饋效果也快的過程。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/7e\/7e461fa693078060e600c121e649f25a.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第三個階段是場景化，在培訓過程中，發現做得不夠深入，跟不上經紀人的進步速度，所以就挑選了其中一個場景做深入的場景化，選的是公司特色場景-VR場景，並做了兩步認證：基礎認證和專家認證，可以這麼認爲，前面的屬於基礎教育，VR屬於精英教育。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/7b\/7b861e8746edb3a2fd652762598ac7c8.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第四個是模塊化，做完場景化之後，發現很多能力都可以去複用給其他的業務，比如很經典的評價能力，除了在培訓裏面能用，在作業過程中可不可以用，在作業之後可不可以用？這是一個複用。再有就是在做培訓的時候，有一個天然訴求：要梳理整個業務的所有作業流程，而且這個過程是非常標準的，是完全符合公司專家對於業務的系統性理解。所以我們可以把整個內容加上能力全都發放給其他的業務產品。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/eb\/eb6cf6ec113369ee00421be0162a4939.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"經過4個模塊之後，大概有3個版本迭代上線，之後去採訪一些經紀人，他們對於講盤能力的提升，還是比較認可的。對於業務上的表現，我們長期監控了一些分公司的業績表現（如下圖），確實能夠看得出來，參與講壇智能培訓的人中成績好和成績不好的人，表現如灰色的柱狀圖，他們的累積業績差會越來越高，越來越大。其實這只是做一個證明，因爲我們都知道只要你學習，你就會進步，但是進步是一種定性的描述，但是定量描述沒有辦法做到，所以我們在嘗試建立一種定量的歸因方式。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/97\/97a7d601e34a412c30184e3c28502582.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"知識運營閉環的建立"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"前面講過，在做知識圖譜或者知識建設的時候，會存在一個問題，即很難去評價知識質量，通常會利用人工標註的方式去評估，但是對於沒有公允對標數據，該怎樣評價數據質量呢？因爲沒有找到權威性的人去告訴你什麼樣的數據是對的，還有就是在應用實踐中欠缺哪些點，這些可能也是需要通過在應用中得到我們欠缺的一面，對於如何獲取這些點，我們設立運營目標：第一是加強事理圖譜；第二是建立完備的評估體系，對於有標和無標數據進行質量評估，進而引導數據質量建設。我們有這樣一個機制，通過行業圖譜在業務上的應用，會發現質量評價的方式分爲這樣幾個部分：第一是空值檢測，第二差值檢測，第三是用戶引導，第四是主動探測，第五是抽樣標註，下面分別來解釋一下這幾個部分。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/35\/35fc5182b2b8e7c7573e9d61d784c254.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"空值檢測和差值檢測都是應對於高頻的B2C客服過程，因爲我們這個行業B2C的交互頻率非常高，類似於微信、企業微信或者釘釘這類IM通訊工具。針對用戶和經紀人之間的交互非常頻繁，我們做了一個插件，當用戶問一個問題之後，我們會給一個提示，說這個答案大概是什麼，然後你可以採納，也可以不採納。如果經紀人不採納，我們就要分析這個原因：有可能是因爲答案錯了，或者是答案不好，我們再去分析這樣的歸因，會找到這樣一個分佈：明確有多少的問題是在B2C過程就驗證了問題是不對的。至於爲什麼不去直接採用人工標註，因爲這些標註的人不如經紀人專業，另外它的量級也趕不上經紀人量級，我們直營的經紀人大概有40萬左右，這個量級已經非常大了，任何標註公司都是跟不上的。在這個過程中我們知道哪些數據是重點被提及的，如果再去做採集標註的話，應該是一些抽樣的方式，所以沒有辦法去直接在一線上得到價值最大化。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/f2\/f29d754c4d9f35d1526f6f28cd3afdfa.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"主動探測，我們每天大概去發佈這樣一些信息，直接觸達經紀人，問他們樓盤的信息到底是什麼，獲取一些結果，再根據加權的PV找到人工運營，幫我們補充這樣的數據。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"這裏詳細介紹一下檢測過程，PV線上加權已經可以覆蓋整體數據的76%，採納率不到50%。這樣我們就可以不斷地下鑽，找到最後的錯誤比大概在1.2%，對於這1.2%的錯誤比，我們可以預估在覈心數據裏面，數據準確率的天花板大概在98%左右。我們探測的時候發現，47.74%的量級大概是不到20萬，通過五折交叉驗證，觸達經紀人大概只需要每天0.9次，所以這個價值還是比較足夠的，通過這樣的鏈路去直接觸達經紀人去進行知識運營的探測。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/42\/424c288f3b7b17914932ec76830b5a4a.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/03\/03474bafcd3ff718bea9c3cbdc4f2672.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"貝殼找房行業圖譜未來思考"}]},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/d5\/d55f3296af38de8d3a8c86017aee1c66.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"下面跟大家分享一下我們對於貝殼找房行業圖譜的未來思考。前面講到我們希望能夠成爲貝殼的知識中臺，進而成爲房產行業知識的標準制定者。在競品、實體、培訓教育評估問答的知識賦能過程中，可以成爲貝殼知識建設的第一責任體，進而成爲貝殼的核心競爭力。這個是我們對內的宣揚，但是對外的話，我們希望對房產行業這種深度非常深的服務行業去做一個知識標準的貢獻。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/50\/50f04b9782718bea0e196e2c9ed2db6c.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"對於我們重點業務的思考，大概從6個角度去分析培訓平臺："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第一是對於目標受衆的一個思考，我們嘗試了兩種類型，一個是基礎教育，一個是精英教育，對於我們這種流動性高的行業，基礎教育的ROI會更高。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第二是關於內核體驗的，問答式的優點就是快，只做QA。交互式就是AQA、QAQA這種類似於對話機器人的一個過程，體驗肯定是交互式的好，但是交互式的做起來比較麻煩，不過我們在未來還是會重點向交互式方向發展，因爲我們的場景更復雜，如果是純功能、純政策類的，可能用問答式更適合。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第三是在做業務發展的時候，選擇一個什麼樣的方向，我們在做前面講盤以及新房二手都是向平臺化發展，在做VR的時候，我們做的是場景化的發展，這個跟基礎教育和精英教育是比較類似的，它獨特的點是場景化對於SOP的確認路徑要求更高，對於推廣速度要求更高，所以在這個業務裏，我們可能會優先平臺化，其次場景化，大概7:3的配比，不同行業可能有不同的發展方式。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第四是過去我們培訓衍生出很多能力，但大都是向外去輸出，之後可能會多與其他的作業方面去進行多項聯動，這也是大勢所趨，畢竟我們不是做純教育的。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第五是如何去推廣培訓，一個是打口碑，就是不斷積累口碑，一個是有快速的應急反饋，但是這些都建立在一個條件之上，就是我們去做這件事情的時候，需要一些運營節奏的手段去配合。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"image","attrs":{"src":"https:\/\/static001.geekbang.org\/infoq\/be\/beef13edf2a6fe6524e91c7e6f1ff4f7.jpeg","alt":"圖片","title":null,"style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":null,"fromPaste":true,"pastePass":true}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"第六是之前我們只關注單點訴求，而對於一個人來說，他做一些技能的方方面面都要去考慮，這將是下一階段的一個轉變，在未來我們會深化培訓平臺，讓這樣一個鏈路成爲經紀人的成長通路。對於多向聯動，和小貝在培訓、作業輔助、事後診斷等方面應用到我們的業務中，最後觸達經紀人和管理者。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"嘉賓介紹："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","marks":[{"type":"strong"}],"text":"孫拔羣 "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"貝殼找房 | 高級技術經理"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"孫拔羣，畢業於哈爾濱工業大學，曾就職於騰訊、搜狗、微博等大型互聯網公司以及創業公司，2018年加入貝殼主持建設貝殼房產知識體系，通過數據引入、知識加工，建立了有貝殼特色的行業知識圖譜。同時，通過知識對業務賦能，支撐貝殼知識型業務，作爲公司主打智能化產品—小貝助手智能培訓方向負責人，專注於提升經紀人專業技能，打造培訓評價平臺。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"本文轉載自：DataFunTalk（ID：dataFunTalk）"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"原文鏈接："},{"type":"link","attrs":{"href":"https:\/\/mp.weixin.qq.com\/s\/ozMfVZd7eycUr3tZ8EV_Ig","title":"xxx","type":null},"content":[{"type":"text","text":"貝殼基於事理圖譜的應用與實踐"}]}]}]}