201133366 六、發明說明： 【發明所屬之技術領域】 本發明涉及網路技術領域，特別涉及一種提高網站資 料傳輸速度的方法、裝置及系統。 【先前技術】 目前，隨著互聯網服務種類的日益豐富，網站伺服器 向用戶端傳輸的資料量越來越大，例如：各種圖文資料、 語音資料、視頻資料等。如此大量的網站資料在同一時間 傳輸給用戶端時，將會導致網路資料傳輸速度的急劇下降 ，甚至造成整個網站的癱瘓。以網路廣告爲例，網路廣告 可以迅速向用戶群傳達商家資訊，激發用戶的購買欲，因 此，在用戶瀏覽某一網站時，通常該網站伺服器會向用戶 用戶端傳輸一些網路廣告資料，若有大量的用戶在同一時 間瀏覽該網站，該網站伺服器將會在同一時刻向用戶端傳 輸大量的廣告資料，導致網路資料傳輸速度的減慢，甚至 造成網站伺服器癱瘓。爲了降低網路廣告資料的傳輸對網 路傳輸速度造成的影響，現有技術下，常常通過減少向用 戶端傳輸的廣告資料量來達到提高網站資料傳輸速度的效 果，然而，盲目的減少廣告資料量無疑會降低廣告的投放 效果。如何能在保證廣告投放效果的基礎上，而又能提高 網站廣告的傳輸資料成爲亟待解決的一個重要問題。 【發明內容】 -5- 201133366 本發明實施例提供一種提高網站資料傳輸速度的方法 '裝置及系統，用以在保證廣告投放效果的基礎上，減少 廣告投放時傳輸的資料量。 本發明實施例提供的具體技術方案如下： —種提高網站資料傳輸速度的方法，包括： 根據用戶瀏覽網站時的操作行爲獲得相應的特徵屬性 集合’再根據所述特徵屬性集合在預設的規則庫中獲取與 該特徵屬性集合相匹配的至少一條規則； 根據獲得的至少一條規則篩選出與該規則約束的場景 相對應的至少一條廣告，並向所述用戶投放該至少一條廣 告； 監測用戶針對所述至少一條廣告的投放產生的操作行 爲’並將收集的相關參數轉化爲相應的規則對所述規則庫 進行更新。 一種用於提高網站資料傳輸速度的裝置，包括： 獲取單元’用於根據用戶瀏覽網站時的操作行爲獲得 相應的特徵屬性集合，再根據所述特徵屬性集合在預設的 規則庫中獲取與該特徵屬性集合相匹配的至少一條規則； 第一處理單元’用於根據獲得的至少一條規則篩選出 與該規則約束的場景相對應的至少一條廣告，並向所述用 戶投放該至少一條廣告； 第二處理單元，用於監測用戶針對所述至少一條廣告 的投放產生的操作行爲’並將收集的相關參數轉化爲相應 的規則對所述規則庫進行更新。 -6 · 201133366 一種用於提高網站資料傳輸速度的系統，包括： 規則庫，用於保存用以搜索廣告的各種規則； 廣告投放管理裝置，用於根據用戶瀏覽網站時的操作 行爲獲得相應的特徵屬性集合，以及根據所述特徵屬性集 合在預設的規則庫中獲取與該特徵屬性集合相匹配的至少 一條規則，再根據獲得的至少一條規則篩選出與該規則約 束的場景相對應的至少一條廣告，並向所述用戶投放該至 少一條廣告，以及監測用戶針對所述至少一條廣告的投放 產生的操作行爲，並將收集的相關參數轉化爲相應的規則 對所述規則庫進行更新。 本發明實施例中，爲了對好的投放經驗進行積累，引 入了規則庫的槪念，它針對廣告投放後帶來的諸多效果， 依據投放關聯的諸多因素進行分類，並對每一類別的投放 效果中較好的部分進行統計歸納，總結出每類投放中較優 的一些投放匹配規則，規則庫的建立和進化均直接依據於 廣告投放效果，廣告投放效果的有所變化，將會通過規則 庫即時地反應在其保存的用於指導廣告選擇的各類規則上 ，使得廣告的選擇完全依賴於其投放效果，也使得規則庫 的更新進化即時地基於廣告投放效果來實現，令各種規則 的優化得以自動化和即時化，具有實現代價小，週期短和 優化速度快等優點。這樣，便無需盲目地減少廣告投放量 ，而是根據用戶的實際需求有目的有針對性的投放相應的 廣告，而減少不必要廣告的投放量，從而在保證廣告投放 效果的基礎上，減少了網站廣告在投放時傳輸的資料量， 201133366 提高了系統的資料傳輸速度，進而提升了網站的服務品質 【實施方式】 爲了，本申請實施例中，採用基於廣告效果的規則庫 來支援廣告的投放策略的選擇，以提高網站資料的傳輸速 度。其具體爲：用於管理廣告投放的裝置根據用戶瀏覽網 站時的操作行爲獲得相應的特徵屬性集合（如，用戶當時 瀏覽網頁的場景——包括瀏覽的時段，瀏覽的網頁ID和 廣告位元ID以及用戶標識ID等等），並根據所述特徵屬 性集合在預設的規則庫中獲取與該特徵屬性集合相匹配的 至少一條規則，再根據獲得的至少一條規則篩選出與該規 則約束的場景相對應的至少一條廣告，並向所述用戶投放 該至少一條廣告，以及監測用戶針對所述至少一條廣告的 投放產生的操作行爲，並將收集的相關參數轉化爲相應的 規則對所述規則庫進行更新。其中，所述特徵屬性集合用 於描述用戶瀏覽時間的特殊性、瀏覽的網頁和廣告的特性 、用戶的長期興趣偏好及用戶瀏覽網站時最近的操作行爲 意圖偏好等等。這樣，便無需肓目地減少廣告投放量，而 是根據用戶的實際需求有目的有針對性的投放相應的廣告 ’而減少不必要廣告的投放量，從而在保證廣告投放效果 的基礎上’減少了網站廣告在投放時傳輸的資料量，提高 了系統的資料傳輸速度，進而提升了網站的服務品質。 所謂廣告效果’即是指廣告被投放之後，用於衡量其 • 8 - 201133366 受用戶歡迎程度的指標，包含多種預設參數，例如，用戶 點擊率，到'達目標頁面以後的瀏覽量，註冊量，收藏量， 購買量等諸多指標。 規則庫：是指從過去廣告投放後帶來的效果中，依據 投放關聯的諸多因素分類，對每一類別的投放效果較好的 投放進行統計歸納，總結出來每類廣告投放較優的一些投 放匹配規則的總集合，針對該規則庫，需要即時地不斷進 行遺傳進化的經驗積累，並利用積累的經驗指導曰後的廣 告投放。 下面結合附圖對本申請優選的實施方式進行詳細說明 〇 參閱圖1所示，本申請實施例中，用於管理廣告投放 以提高網站資料傳輸速度的系統包括規則庫1 〇和廣告投 放管理裝置1 1，其中 規則庫1 〇，用於保存用以搜索廣告的各種規則，是 所有廣告投放策略實施經驗的積累，並始終進行著即時進 化更新。規則庫1 0中各種規則的積累使得實施效果好的 廣告投放策略得以保存，從而爲後續的操作提供了寶貴的 經驗。本實施例中，在制定投放廣告的廣告投放策略時， 綜合考慮了影響廣告投放效果的所有類型的因數，一次性 的進行廣告投放策略的選取，保證了廣告投放策略的全局 最優。例如：在爲一例廣告選擇廣告投放策略時，根據當 時的廣告位元、投放場景、瀏覽用戶的興趣和近期瀏覽行 爲等特徵資料來設置其廣告投放策略中的各種參數，如， -9- 201133366 投放時間，投放次數等等》 廣告投放管理裝置11’用於根據用戶瀏覽網站時的 操作行爲獲得相應的特徵屬性集合，以及根據所述特徵屬 性集合在預設的規則庫中獲取與該特徵屬性集合相匹配的 至少一條規則’再根據獲得的至少一條規則篩選出與該規 則約束的場景相對應的至少一條廣告，並向所述用戶投放 該至少一條廣告，以及監測用戶針對所述至少一條廣告的 投放產生的操作行爲’並將收集的相關參數轉化爲相應的 規則對所述規則庫進行更新。 本申請實施例中，在選擇廣告投放策略時，可以查找 歷史上相同或是相似投放實例所採用的廣告投放策略作爲 參考資料’再對此類投放實例的廣告效果對應的投放規則 按照投放效果分數從大到小的順序進行排序，找出最優效 果的幾則廣告投放策略及相應的廣告特徵參數，並對這些 廣告特徵參數進行適當機率的組合變異或是擴展變異，然 後依據變異後的廣告特徵參數選出符合條件的備選廣告， 並依據投放效果的綜合評分對各備選廣告進行機率競選操 作，篩選出最終被投放的廣告，接著，對投放的廣告進行 即時的跟蹤，監測其投放效果，最後依據投放效果當前選 擇的廣告投放策略作出調整和更新，積累好的投放模式， 摒棄差的投放模式，從而使廣告投放策略得以優化。這樣 ，即減少了網路廣告在網路中傳輸的資料量，又可以收到 很好的廣告投放效果。 參閱圖2所示，本申請實施例中，廣告投放管理裝置 -10- 201133366 11包括獲取單元110、第一·處理單元111和第二處理單元 1 1 2，其中， 獲取單元1 1 〇 ’用於根據用戶瀏覽網站時的操作行爲 獲得相應的特徵屬性集合’再根據所述特徵屬性集合在預 設的規則庫中獲取與該特徵屬性集合相匹配的至少一條規 則。 第一處理單元1 1 1 ’用於根據獲得的至少一條規則篩 選出的與該規則約束的場景相對應的至少一條廣告，並向 所述用戶投放該至少一條廣告； 第二處理單元1 1 2，用於監測用戶針對所述至少一條 廣告的投放產生的操作行爲’並將收集的相關參數轉化爲 相應的規則對所述規則庫進行更新》 本申請實施例中，在上述規則庫1 0中，—條規則由 下列幾種資料向量構成，包括： A、 廣告位元特徵向量（記爲忍），包含的分量有： 廣告位元對應的網站頻道（記爲Fj )、廣告位元類目（記 爲)、廣告位元所在網頁的類目（β )、廣告位元所在 網頁的關鍵字（Ffl4 )。上述各參量的關係可以表示爲： Fa-(F：,F：',Fa\F：) 〇 B、 廣告位元投放場景特徵向量（記爲巧），包含的 分量有·投放時段（記爲F〗）、日期類型（記爲F62)、季 節（記爲F/ )、時事標記（記爲Ffc4 )，其中，時事標記用 於標注最近是否有大事’所謂大事的類型包含但不限於： 地震、政治、經濟、高考、醫療等等。上述各參量的關係 -11 - 201133366 可以表示爲·· Fb = (F〖，Fb2，Fb3，Fb4)。 本發明實施例中，將向量尺連接向量G生成的新向 量^ = (GA)，稱爲廣告位元向量，該向量描述了廣告投 放時不依賴於用戶的整體投放影響因素。 C、 用戶自然屬性和歷史長期的興趣行爲特徵向量（ 記爲忍），包含的分量有：用戶性別（記爲Fj )、用戶年 齢段（記爲)、用戶興趣（記爲F/，即用戶日常的上網 規律，分節假日，時段）、用戶購物興趣（記爲F/，即用 戶曰常瀏覽和購買的貨品類目）、用戶喜歡的關鍵字（記 爲Θ )、用戶品牌傾向（記爲F/ )、用戶消費檔次（記爲 ，即用戶瀏覽和購買的貨品的價格段）、用戶商家傾向 (記爲尸/)、用戶地域（記爲F/ )和用戶信用度（記爲 )。上述各參量的關係可以表示爲：尺«,G2，···,#，） D、 用戶近期即時的瀏覽和購物特徵向量（記爲巧） ，包含的分量有：短期和當前點擊廣告類目（記爲β )、 短期和當前瀏覽貨品類目（記爲β )、短期和當前購買 貨品類目（記爲巧3 )、短期和當前點擊廣告位類目（記爲 巧4 )、短期和當前瀏覽網頁類目（記爲Θ )。上述各參量 的關係可以表示爲：巧=(θ，β,···，β)。 本發明實施例中，將向量G連接向量圮生成的新向量 4 =d巧）稱爲用戶特徵向量，代表用戶自身長短期特徵 屬性，也稱爲用戶特徵屬性向量。 E、 廣告位元廣告投放策略特徵向量（記爲，包 含的分量有廣告投放策略（記爲Θ )和相應的配置參數 -12- 201133366 (記爲Θ)。其中，廣告投放策略，是廣告在展現時使用 的投放方式，如，採用關鍵字-內容匹配演算法投放、採 用用戶-行爲匹配演算法投放或者按照廣告效果投放；而 與廣告投放策略相對應的配置參數，可以包含用戶ID和 廣告關鍵字等等。上述各參量的關係可以表示爲： Fe=(Fe\Fe2) F、 被投放的廣告特徵向量（記爲巧），包含的分量 有：廣告產品類型（記爲F；)、廣告類目（記爲F/ )、廣 告展現形式（記爲F/，即圖文，文字鏈，或者flash )、 廣告內容自定義參數（記爲i=>4，即用於點擊搜索的關鍵字 等）、廣告的競價關鍵字（記爲F/)、廣告的競價價格（ 記爲)、廣告主的信譽度（記爲Fr7 )、廣告貨品的品牌 (記爲F/ )、廣告貨品的價格段（記爲)、廣告商家類 型（S3爲)、廣告商家地域（記爲F)1 )。上述各參量 的關係可以表示爲： 本申請實施例中’將向量弋上，/;，/^，€，巧連接生成新 向里厂-(Fa, H ，Fe，Fy)，該向量就是用於制定厣告投放策 略的規則庫的具體描述。 G、 廣告效果歸一化指標向量（記爲 ，包含的分 量有：點擊率（記爲S )、點擊收入（記爲F/)、引入流 量（記爲G )、收藏數（記爲F/ )、成交金額（記爲) 、傭金金額（記爲尸g6 )、成交率（記爲 <)和註冊率（ 記爲g ) 通過向量& ’我們就可以計算用於描述廣告投放效 -13- 201133366 果的分數S，S的計算公式如下： 8 S = J^w.x]v〇rm(F·), 8 其中，Σ w，·=1 ’ 被稱爲權重係數； i=1 iVorm(/^) = l〇〇x(e/Fp，是歸—函數，將 轉化爲 〇_ 1〇〇 之間的數値。 所以S的範圍是0 · 1 〇 〇 ’權重係數μ由管理人員根據 經驗値預先設置的’例如’確認點擊率尸是用於衡量廣告 投放效果的最重要因素，那麼可以預設％ = 1，則 8 •S = lxM>rm('丨）+ Σ〇χ版= ，又例如，確認 Fg 的所 /=2 有分量都一樣重要’那麼，可以預設w,. = 1/8 = 0.125。總之 W- 在越趨近數値1 ’則表示&對應的分量在衡量廣告投放 效果中的權重越大。 本申請實施例中，將向量 新向量心,，將向量4稱爲廣告投放效 果統計指標向量。 基於上述參數設置，下面以一個具體的應用場景爲例 進行詳細說明。假設初始選擇投放的廣告有三個，分別稱 爲廣告A、廣告Β和廣告C »在這三則廣告投放一段時間 後，系統在某用戶登錄網站時，需要根據這三則廣告的投 放效果來確定選擇哪一個廣告向登錄用戶進行投放。 本實施例中，假設規則庫中的預設規則和用戶訪問情 -14- 201133366 景如下： 三則廣告A，B，C : 廣告A :廣告產品MP3，廣告產品價格<1〇〇〇元，店 主信用200分，廣告採用圖片展現，選擇關鍵字精確匹配 投放，競價價格0.3元。 廣告B :廣告產品觸控手機，廣告產品價格>2000元 ’店主信用500分，廣告採用flash展現，選擇關鍵字模 糊匹配投放，競價價格0.8元。 廣告C :廣告產品玩偶，廣告產品價格< 100元，店 主信用30分，廣告採用圖片展現，選擇關鍵字模糊匹配 投放，競價價格1元。 上述各廣告由管理人員在網路側發佈，預先存儲在資 料庫中，由廣告搜索引擎獲得， 而對應上述三條廣告，在規則庫中預設了以下6條規 則： 1、 Rl=(男性用戶，對數碼感興趣，收入中上，最 近購買觸控手機，經常訪問新聞類廣告，點擊的廣告是 MP3，廣告產品價格<2000元’廣告投放時段爲週末，投 放該廣告的廣告主信用大於20分，廣告採用fiash展現， 廣告採用關鍵字精確方式投放，0.2元<點擊競價收入均價 <0.4 元）。 2、 R2=(男性用戶’對運動裝備感興趣，收入未知 ，最近購買旱冰鞋’經常訪問博客類廣告位元，點擊的廣 告是觸控手機，廣告品價格>2000元’廣告投放時段是週 -15- 201133366 末上午’投放該廣告的廣告主信用大於3 00分，廣告採用 flash展現，廣告採用關鍵字模糊方式投放，〇.3元 <點擊 競價收入均價< 1元） 3、 R3=(男性用戶，對運動裝備感興趣，無收入（ 學生），最近購買香水，經常訪問動漫類廣告位元，點擊 的廣告是玩偶，廣告品價格< 1 00元，廣告投放時段是工 作曰晚上，投放該廣告的廣告主信用大於20分，廣告採 用圖片展現，廣告採用關鍵字模糊方式投放，0.3元 <點擊 競價收入均價< 1 · 3元）。 4、 R4=(女性用戶，對運動裝備感興趣，收入高端 ，最近購買香水，經常訪問新聞類廣告位元，點擊的廣告 是觸控手機，廣告產品價格>5 000元，廣告投放時段是工 作曰上午，投放該廣告的廣告主信用大於5 00分，廣告採 用圖片展現，廣告採用關鍵字精確方式投放，0.3元 <點擊 競價收入均價< 1 · 3元） 5、 R5=(女性用戶，對玩偶感興趣，收入中，最近 購買MP 3，經常訪問博客類廣告位元，點擊的廣告是玩偶 ，廣告產品價格< 1 〇〇元，投放時段是週末晚上，投放該 廣告的廣告主信用大於30分，廣告採用圖片展現，廣告 採用關鍵字精確方式投放，〇·5元 <點擊競價收入均價<0.8 元）。 6、 R6=(女性用戶，對裝飾物感興趣，收入中上， 最近購買MP3，經常訪問動漫類廣告位元，點擊的廣告是 觸控手機，廣告產品價格>2〇〇〇元，投放時段是週末上午 -16- 201133366 ，投放該廣告的廣告主信用大於3 00分，廣告採用圖片展 現，廣告採用關鍵字模糊方式投放，〇.5元<點擊競價收入 均價< 0.8元） 基於上述規則，假設用戶訪問情景如下： 情景1 :(用戶U1，在週末上午，經常訪問新聞類 廣告位元） 情景2 :(用戶U2，在工作日晚上，經常訪問博客 類廣告位元） 情景3 :(用戶U3，在工作日上午，經常訪問新聞 類廣告位元） 廣告投放管理裝置11根據上述三種場景，收集用戶 的訪問資訊，並將該訪問資訊存儲在網站曰誌中，並在對 網站日誌加以分析後，提取出各用戶的特徵屬性集合。 那麼，可以獲得這三個用戶的特徵屬性集合，分別爲 用戶U,的特徵屬性是（男性，對數碼感興趣，收入 中上，最近購買觸控手機 用戶U2的特徵屬性是（女性，對玩偶裝備感興趣， 收入中，最近購買MP3 ) 用戶U3的特徵屬性是（女性，對運動裝備感興趣， 收入高，最近購買觸控手機）。 那麼，參閱圖3所示，本申請實施例中，廣告投放管 理裝置1 1基於廣告投放效果對廣告投放進行管理的詳細 流程如下： -17- 201133366 步驟300:確定某用戶登錄網站系統後，根據該用戶 瀏覽網站時的操作行爲獲得相應的特徵屬性集合，並根據 該特徵屬性集合在預設的規則庫中選擇相匹配的規則，該 規則用於選擇符合所述用戶的特徵屬性的備選廣告。 例如，對於用戶U,的訪問（男性，對數碼感興趣， 收入中上，最近購買觸控手機，訪問時段在週末上午，經 常訪問新聞類廣告位元），通過函數β similarity (¢/, A)可以計 算規則庫1 0中所有規則和u,的相似度數値，然後對相似 度數値進行倒排序，根據設定的閾値，取排名在τ ο p X的 規則，這些規則就是在規則庫中找到的與用戶U 1的特徵 屬性相同或者相似的規則。 HSimiiavUy ^) = Π Σ sim{Norm(xi), Norm{yJ.)) > 其中，X ’ y0F，/7 = ^/；；，/；，^，^，')，i 取値[a，f]，尸〇 〜巧爲規則庫中預設的用於描述各類廣告屬性的集合，G 〜巧用於組建/=;，j爲6中包含的分量。 當然，上述F = ，^巧，^巧)僅爲一種舉例，實際應用 中，若基於實際應用環境，增加了更多定義的向量集合， 如F = ，尸2,......A)，厂是其中的六種，則上述公 式"《祕_(以=]^石伽(伽/^/)身/7»1(兄0)同樣適用，其中，又， yDF > F = {FvF2,……，Fn) ，i取値[1，η ] ’ F〇〜尸n爲規則庫中 預設的用於描述各類廣告屬性的集合，巧〜尺用於組建/=； ，j爲巧中包含的分量。 • 18- 201133366 經過採用檢索函數打伽办吻，針對用戶U1可以選中 規則R1 :(男性用戶’對數碼感興趣，收入中上，最近 購買觸屏手機’經常訪問新聞類廣告位，點擊的廣告是 MP3 ’廣告產品價格<2000元，投放時段是週末，投放該 廣告的廣告主信用大於10分，廣告採用fUsh展現，廣告 採用關鍵字精確方式投放’ 0.2元<點擊競價收入均價<0.4 元）。 實際應用中’最終選擇的規則可以是一條，也可以是 多條，本申請實施例中’假設選中的與登錄用戶的用戶特 徵集合相匹配的規則爲R4，R5 , R6。 步驟3 1 0 :根據選中的規則篩選出相應的備選廣告。 例如’假設選中的與用戶特徵集合相匹配的規則爲 R4，R5，R6，那麼接著，將用戶ID和從選中的規則中提 取的關鍵字作爲參數，傳遞給廣告搜索引擎，由廣告搜索 引擎根據獲得的參數搜索出相應的備選廣告。本實施例中 ，假設廣告搜索引擎根據選中的規則R4、R5和R6，篩選 出相應的備選廣告分別爲廣告A、廣告B和廣告C。 步驟3 20 :對獲得的備選廣告進行機率競選。 本申請實施例中，採用以下方式對備選廣告進行機率 競選： 將根據規則R4、R5和R6篩選出的備選廣告表示爲 : <，其中，i是對應的規則下標’ j表示具體獲得的備 選廣告數量’本實施例中’ i的取値是4’ 5’ 6。那麼， 所有篩選出的備選廣告展開如下： -19 · 201133366 .7 4 -7 5 7 6 d^^ 14 15 16 Jy4y4 /_V - 4 5 6 ΛΛΛ 機率競選步驟如下： 根據計算的機率競選評分的數値，對選擇的規 則Ri進行倒排序，採用的函數是 ^result (x,y) = e 乂 Hsimilarity 0,少），其中，β爲預設的效果 膨脹因數，初始設爲1，管理人員可以根據β參數選取的 測試效果來優化，S爲 y對應的規則的效果分數， ，Gw =(Fa，K，巧，G)，X表示用戶特定訪問對 應的廣告位元向量匕和用戶特徵向量匕的連接向量，歸 屬於Gw，y表示選定的規則R中的廣告位元分量Ffl6和用 戶特徵分量&的連接向量，也歸屬於Fa6a/。 對排序完的Ri進行τ〇Ρχ (排名前X的結果）選取， 再針對選取的ΤορΧ，確定相應的備選廣告，此處，假設 χ = 2 ’那麼最終確定選取的規則就是R4、R5，而相應的 備選廣告即是廣告Α和廣告Β，表示爲，將此廣告 集合簡記爲Ad。 最後，針對集合Ad再次進行隨機抽樣，抽樣數目爲 Y (依據系統的參數設定，假設Y = 1 )，那麼，最終得到 的機率競選結果可以是廣告A，也可以是廣告B。 步驟3 3 0 :對最終選中的廣告進行投放展現。 步驟3 40 :監測用戶針對最終投放的廣告的操作行爲 -20- 201133366 ，並根據收集的廣告投放效果資料對規則庫1 0進 〇 在上述步驟340中，在對最終選中的廣告進行 現後，進而在步驟3 50中對投放產生的日誌進行即 收集。日誌記錄的主要內容包含但不限於：用戶id 時間，點擊的廣告位元，瀏覽的廣告位元，收藏的 〇 在與投放時間相隔一段時間後，對上述廣告的 果進行計算，具體爲計算廣告的投放效果資料（包 分數S和支援度N )，再根據計算出的廣告投放效 ’對規則庫1 0中保存的規則進行更新，本申請實 ，對規則庫1 0進行更新時包括兩種操作：1、根據 放效果資料提取出對應的新規則添加至規則庫1 〇弓 根據廣告投放效果資料對規則庫1 0中現有的規則 化。 所謂提取即是指將大量出現（即機率大於某個 的廣告效果統計指標向量扣轉化爲規則。 例如，用戶U在某個時間段Τ，訪問了特定的 頁W’網頁上的廣告位元Ρ，該廣告位元ρ上展現 告A ’用戶看見廣告a後，點擊了廣告a的鏈結 廣告Λ進行推廣的產品詳情頁p，接著購買了該詳 上的產品I，並收藏了產品J。用戶的這一系列的 爲’會被系統記錄爲·· （ U，Τ，W，Ρ，A，I，J ) 前述的集合C和集合D; 行更新 投放展 時記錄 ，訪問 產品等 投放效 括效果 果資料 施例中 廣告投 3 ，2、 進行優 閾値） 一個網 的是廣 ，進入 情頁P 操作行 ，詳見 -21 - 201133366 接者，把記錄下的用戶u的一系列操作行爲通過分 析整理，對應保存爲該用戶的特徵屬性集合。包括：τ轉 化成對應的投放的時段Ti、投放的季節Ts、是否有重大 節日Tf等等；201133366 VI. Description of the Invention: [Technical Field] The present invention relates to the field of network technologies, and in particular, to a method, device and system for improving website data transmission speed. [Prior Art] At present, with the increasing variety of Internet services, the amount of data transmitted by the web server to the client is increasing, for example, various graphic materials, voice data, video materials, and the like. When such a large amount of website data is transmitted to the client at the same time, it will lead to a sharp drop in the speed of network data transmission, and even cause the embarrassment of the entire website. In the case of online advertising, online advertising can quickly convey business information to the user community and stimulate the user's desire to purchase. Therefore, when a user browses a website, the website server usually transmits some network advertisements to the user terminal. According to the information, if a large number of users browse the website at the same time, the website server will transmit a large amount of advertisement data to the client at the same time, which causes the network data transmission speed to slow down, and even causes the website server to be paralyzed. In order to reduce the impact of the transmission of online advertising data on the transmission speed of the network, in the prior art, the effect of increasing the speed of data transmission of the website is often achieved by reducing the amount of advertising data transmitted to the user terminal, however, the amount of advertising data is blindly reduced. Undoubtedly will reduce the effectiveness of advertising. How to improve the transmission of website advertisements on the basis of ensuring the effectiveness of advertisements has become an important issue to be solved urgently. SUMMARY OF THE INVENTION -5-201133366 Embodiments of the present invention provide a method and apparatus for improving the speed of data transmission of a website, which are used to reduce the amount of data transmitted during advertisement delivery on the basis of ensuring the effect of advertisement delivery. The specific technical solutions provided by the embodiments of the present invention are as follows: A method for improving the speed of data transmission of a website, comprising: obtaining a corresponding feature attribute set according to an operation behavior when a user browses a website, and then according to the feature attribute set in a preset rule Obtaining at least one rule matching the feature attribute set in the library; filtering at least one advertisement corresponding to the scene restricted by the rule according to the obtained at least one rule, and delivering the at least one advertisement to the user; The operation behavior generated by the delivery of the at least one advertisement 'translates the collected related parameters into corresponding rules to update the rule base. An apparatus for improving the speed of data transmission of a website, comprising: an obtaining unit configured to obtain a corresponding feature attribute set according to an operation behavior when a user browses a website, and obtain the same in the preset rule base according to the feature attribute set; At least one rule matched by the set of feature attributes; the first processing unit is configured to: filter at least one advertisement corresponding to the scene restricted by the rule according to the obtained at least one rule, and deliver the at least one advertisement to the user; The second processing unit is configured to monitor an operation behavior generated by the user for the delivery of the at least one advertisement, and convert the collected related parameters into corresponding rules to update the rule base. -6 · 201133366 A system for improving the speed of data transmission on a website, comprising: a rule base for storing various rules for searching for advertisements; and an advertisement management device for obtaining corresponding features according to an operation behavior when a user browses a website Attribute set, and acquiring at least one rule matching the feature attribute set in a preset rule base according to the feature attribute set, and filtering at least one item corresponding to the scene restricted by the rule according to the obtained at least one rule Advertising, and delivering the at least one advertisement to the user, and monitoring an operation behavior generated by the user for the delivery of the at least one advertisement, and converting the collected related parameters into corresponding rules to update the rule base. In the embodiment of the present invention, in order to accumulate a good delivery experience, a concept library is introduced, which categorizes the effects brought by the advertisement, according to various factors of the delivery association, and serves each category. The better part of the effect is statistically summarized, and some of the better delivery matching rules in each type of delivery are summarized. The establishment and evolution of the rule base are directly based on the effect of the advertisement delivery, and the effect of the advertisement delivery changes. The library responds instantly to the various rules it uses to guide the selection of advertisements, so that the choice of advertisements is completely dependent on its delivery effect, and the update evolution of the rule base is realized based on the effect of advertisement delivery in real time, so that various rules are Optimization is automated and instant, with the advantages of low cost, short cycle and fast optimization. In this way, there is no need to blindly reduce the amount of advertising, but the targeted advertising is targeted according to the actual needs of the user, and the amount of unnecessary advertisements is reduced, thereby reducing the effectiveness of the advertising. The amount of data transmitted by the website advertisement at the time of delivery, 201133366 improves the data transmission speed of the system, thereby improving the service quality of the website. [Embodiment] In the embodiment of the present application, the rule base based on the advertisement effect is used to support the advertisement. The choice of strategy to improve the speed of the transmission of website data. Specifically, the device for managing advertisement delivery obtains a corresponding feature attribute set according to an operation behavior when the user browses the website (for example, a scene in which the user browses the webpage at the time), including a browsing time period, a browsed webpage ID, and an advertisement bit ID. And the user identifier ID and the like, and acquiring at least one rule matching the feature attribute set in the preset rule base according to the feature attribute set, and filtering the scene bound by the rule according to the obtained at least one rule. Corresponding at least one advertisement, and delivering the at least one advertisement to the user, and monitoring an operation behavior generated by the user for the delivery of the at least one advertisement, and converting the collected related parameters into corresponding rules for the rule base Update. The feature attribute set is used to describe the particularity of the user's browsing time, the characteristics of the browsed webpage and advertisement, the long-term interest preference of the user, and the recent operational behavior intent preference of the user when browsing the website. In this way, there is no need to significantly reduce the amount of advertising, but to target the targeted advertising according to the actual needs of the user', and reduce the amount of unnecessary advertising, thereby reducing the effectiveness of the advertising. The amount of data transmitted by the website advertisements at the time of delivery increases the speed of data transmission of the system, thereby improving the service quality of the website. The so-called "advertising effect" refers to the metrics used to measure the popularity of the ads after the ads are delivered, including a variety of preset parameters, such as user click-through rate, to the number of page views after the target page, registration Volume, collection volume, purchase volume and many other indicators. Rule base: refers to the results from past advertising, based on the classification of many factors related to the delivery, statistical summary of the delivery of each category, and summed up some of the better delivery of each type of advertising The total set of matching rules, for which the rule base needs to be continuously accumulated in the process of genetic evolution, and the accumulated experience is used to guide the subsequent advertising. The preferred embodiment of the present application is described in detail below with reference to the accompanying drawings. Referring to FIG. 1 , in the embodiment of the present application, a system for managing advertisement delivery to improve website data transmission speed includes a rule base 1 and an advertisement delivery management apparatus 1 1, the rule base 1 〇, used to save various rules for searching for advertisements, is the accumulation of experience in the implementation of all advertising strategy, and always undergoes an evolutionary update. The accumulation of various rules in the rule base 10 allows the implementation of a good advertising strategy to be preserved, thus providing valuable experience for subsequent operations. In this embodiment, when formulating an advertisement delivery strategy for placing an advertisement, all types of factors affecting the effect of the advertisement delivery are comprehensively considered, and the selection of the advertisement delivery strategy is performed at one time, thereby ensuring the global optimization of the advertisement delivery strategy. For example, when selecting an ad serving strategy for an ad, the various parameters in the ad serving strategy are set according to the feature data of the current advertising bit, the delivery scenario, the browsing user's interest, and the recent browsing behavior, for example, -9-201133366 The delivery time, the number of times of delivery, and the like, the advertisement trafficking device 11' is configured to obtain a corresponding feature attribute set according to the operation behavior when the user browses the website, and obtain the feature attribute in the preset rule base according to the feature attribute set. The set matching at least one rule 'filters at least one advertisement corresponding to the scene restricted by the rule according to the obtained at least one rule, and delivers the at least one advertisement to the user, and monitors the user for the at least one advertisement The operational behavior generated by the delivery' and the related parameters collected are converted into corresponding rules to update the rule base. In the embodiment of the present application, when the advertisement delivery policy is selected, the advertisement delivery strategy adopted by the same or similar delivery instance may be searched as the reference material, and the delivery rule corresponding to the advertisement effect of the delivery instance is determined according to the delivery performance score. Sort from the largest to the smallest, find out the optimal advertising effect and corresponding advertising feature parameters, and make appropriate combinations of these advertising feature parameters to mutate or expand the variability, and then based on the mutated advertising The characteristic parameter selects the qualified advertisements that meet the conditions, and performs a campaign campaign for each candidate advertisement according to the comprehensive score of the delivery effect, selects the final advertisement to be delivered, and then immediately tracks the delivered advertisement to monitor the delivery effect. Finally, the ad delivery strategy currently selected according to the delivery effect is adjusted and updated, the good delivery mode is accumulated, and the poor delivery mode is abandoned, so that the advertisement delivery strategy is optimized. In this way, the amount of data transmitted by the network advertisement on the network is reduced, and the advertisement effect is good. Referring to FIG. 2, in the embodiment of the present application, the advertisement delivery management device-10-201133366 11 includes an acquisition unit 110, a first processing unit 111, and a second processing unit 112, wherein the acquisition unit 1 1 Obtaining a corresponding feature attribute set according to the operation behavior when the user browses the website, and then acquiring at least one rule matching the feature attribute set in the preset rule base according to the feature attribute set. The first processing unit 1 1 1 ' is configured to filter at least one advertisement corresponding to the rule-constrained scene according to the obtained at least one rule, and deliver the at least one advertisement to the user; the second processing unit 1 1 2 For monitoring the operation behavior generated by the user for the delivery of the at least one advertisement, and converting the collected related parameters into corresponding rules to update the rule base. In the embodiment of the present application, in the rule base 10 The rule is composed of the following data vectors, including: A. The advertising bit feature vector (recorded as forbearance), the components included are: the website channel corresponding to the advertising bit (denoted as Fj), the advertising bit category (Remarked as), the category (β) of the webpage where the ad slot is located, and the keyword (Ffl4) of the webpage where the ad slot is located. The relationship between the above parameters can be expressed as: Fa-(F:,F:',Fa\F:) 〇B, the advertising bit is used to deliver the scene feature vector (indicated as a coincidence), and the included components have a delivery period (denoted as F〗), date type (denoted as F62), season (denoted as F/), current event mark (denoted as Ffc4), where the current event mark is used to mark whether there is a major event recently. The type of so-called big event includes but is not limited to: earthquake , politics, economics, college entrance examination, medical care, etc. The relationship between the above parameters -11 - 201133366 can be expressed as · Fb = (F 〖, Fb2, Fb3, Fb4). In the embodiment of the present invention, the new vector ^=(GA) generated by the vector rule connection vector G is called an advertisement bit vector, and the vector describes the influence factor of the overall placement of the advertisement when the advertisement is released. C. User natural attributes and historical long-term interest behavior feature vectors (remembered as tolerance), including components: user gender (denoted as Fj), user year segment (recorded as), user interest (denoted as F/, ie user Daily online rules, holidays, time slots), user shopping interest (denoted as F/, that is, the category of goods that users often browse and purchase), keywords that users like (marked as Θ), user brand tendency (marked as F / ), user consumption grade (denoted as the price segment of the goods browsed and purchased by the user), user merchant tendency (recorded as corpse /), user territory (denoted as F / ) and user credit (marked as). The relationship between the above parameters can be expressed as: ruler «, G2, ···, #,) D, the user's recent instant browsing and shopping feature vector (recorded as clever), including the short-term and current click advertising categories (denoted as β), short-term and current browsing categories (denoted as β), short-term and current categories of goods purchased (denoted as skill 3), short-term and current click-to-advertising categories (denoted as skill 4), short-term and Currently browsing the web page category (remarked as Θ). The relationship between the above parameters can be expressed as: 巧 = (θ, β, ···, β). In the embodiment of the present invention, the new vector 4 = d generated by the vector G connection vector 称为 is called a user feature vector, and represents a user's own long-term and short-term feature attribute, also called a user feature attribute vector. E. The advertising element advertisement delivery strategy feature vector (remembered that the component contains an advertisement delivery strategy (denoted as Θ) and the corresponding configuration parameter -12-201133366 (denoted as Θ). Among them, the advertisement delivery strategy is that the advertisement is The delivery method used in the presentation, for example, using a keyword-content matching algorithm, using a user-behavior matching algorithm to deliver or according to the performance of the advertisement; and the configuration parameters corresponding to the advertisement delivery strategy may include the user ID and the advertisement. Keywords, etc. The relationship of the above parameters can be expressed as: Fe=(Fe\Fe2) F, the advertisement feature vector to be served (indicated as coincidence), and the components included are: the type of advertising product (denoted as F;), Advertising category (denoted as F/), advertisement presentation form (denoted as F/, ie graphic, text chain, or flash), advertising content custom parameters (denoted as i=>4, ie used for click search Keyword, etc.), the auction's bidding keyword (denoted as F/), the auction's bid price (marked as), the advertiser's credit rating (denoted as Fr7), the brand of the advertising product (denoted as F/), the advertising item Price segment ), Advertising business type (S3 is), the ad business area (denoted F) 1). The relationship between the above parameters can be expressed as: In the embodiment of the present application, 'the vector is ,, /;, /^, €, and the new connection is generated to generate a new factory - (Fa, H, Fe, Fy), the vector is used A detailed description of the rule base for developing a trafficking strategy. G, the advertising effect normalization indicator vector (remembered, the included components are: click rate (denoted as S), click revenue (denoted as F /), incoming traffic (denoted as G), collection number (denoted as F / ), the transaction amount (denoted as), the commission amount (denoted as corpse g6), the transaction rate (denoted as <), and the registration rate (denoted as g) by vector & 'we can calculate to describe the effectiveness of advertising - 13- 201133366 The score of S, S is calculated as follows: 8 S = J^wx]v〇rm(F·), 8 where Σ w,·=1 ' is called weight coefficient; i=1 iVorm( /^) = l〇〇x(e/Fp, which is the return-function, which will be converted to the number between 〇_1〇〇. So the range of S is 0 · 1 〇〇' The weight coefficient μ is determined by the manager Experience 値 Pre-set 'for example' confirmation click rate is the most important factor to measure the effectiveness of advertising, then you can preset % = 1, then 8 • S = lxM > rm ('丨) + Σ〇χ version = For example, to confirm that the /=2 component of Fg is equally important 'then, you can preset w,. = 1/8 = 0.125. In short, W- is closer to 値1 ', indicating & In the embodiment of the present application, the vector new vector heart is used, and the vector 4 is referred to as the advertising effect statistical indicator vector. Based on the above parameter settings, the following is a specific application scenario. The example is explained in detail. It is assumed that there are three advertisements initially selected for advertisement, namely advertisement A, advertisement advertisement and advertisement C. After the three advertisements are served for a period of time, the system needs to be based on the three advertisements when a user logs in to the website. The effect of the delivery to determine which advertisement to select to serve to the logged-in user. In this embodiment, it is assumed that the preset rules and user access conditions in the rule base are as follows: three advertisements A, B, C: advertisement A :Advertising product MP3, advertising product price <1 〇〇〇 yuan, store owner credit 200 points, advertising using image display, select keyword exact match delivery, bid price 0.3 yuan. Advertising B: advertising products touch mobile phone, advertising product price > 2000 yuan 'store owner credit 500 points, the advertisement uses flash display, select the keyword fuzzy match delivery, the bid price is 0.8 yuan. C: advertising product doll, advertising product price < 100 yuan, the owner credit 30 points, the advertisement is displayed in the picture, the keyword is selected to be fuzzy matching, and the bid price is 1 yuan. The above advertisements are posted by the management on the network side, pre-stored in In the database, it is obtained by the advertisement search engine, and corresponding to the above three advertisements, the following six rules are preset in the rule base: 1. Rl=(male user, interested in digital, income in the middle, recently purchased touch phone) , often visit news ads, click on the advertisement is MP3, the price of the advertising product < 2000 yuan' advertising time is the weekend, the advertiser credit for placing the advertisement is greater than 20 points, the advertisement is displayed by fiash, and the advertisement is delivered by keyword precise method. , 0.2 yuan < click on the auction price average price < 0.4 yuan). 2, R2 = (male user 'interested in sports equipment, income unknown, recently purchased roller skates' often visit blog-type advertising bits, clicked ads are touch phones, advertising prices > 2000 yuan' advertising time is周-15- 201133366 At the end of the morning, the advertiser's credit for placing the advertisement is greater than 30,000 points, the advertisement is displayed in flash, and the advertisement is placed in a keyword fuzzy manner, 〇.3 yuan <click auction price average price < 1 yuan) 3 , R3 = (male users, interested in sports equipment, no income (student), recently purchased perfume, often visit anime advertising space, click on the ad is a doll, advertising price < 1 00, advertising time is At night, the advertiser's credit for placing the advertisement is greater than 20 points, the advertisement is displayed by the image, and the advertisement is placed in a keyword fuzzy manner, 0.3 yuan < click auction price average price < 1 · 3 yuan). 4, R4 = (female users, interested in sports equipment, high income, recently purchased perfume, often access news advertising, meta-click advertising, advertising product price > 5,000 yuan, advertising time is In the morning, the advertiser’s credit for the advertisement is greater than 50,000 points, the advertisement is displayed in the image, and the advertisement is delivered in the exact way of the keyword, 0.3 yuan < click on the auction price average price < 1 · 3 yuan) 5, R5 = ( Female users, interested in dolls, income, recently purchased MP 3, often visit blog-like advertising slots, clicked ads are dolls, advertising product price < 1 〇〇元, delivery time is weekend night, the advertisement is placed The advertiser credit is greater than 30 points, the advertisement is displayed in the image, and the advertisement is delivered in the exact way of the keyword, 〇·5 yuan < click auction price average price < 0.8 yuan). 6, R6 = (female users, interested in decorations, income in the middle, recently bought MP3, often visit anime advertising bit, click on the ad is touch phone, advertising product price > 2 yuan, placed The time is the weekend -16-201133366, the advertiser's credit for placing the advertisement is greater than 300 points, the advertisement is displayed by the image, and the advertisement is placed in a keyword fuzzy manner, 〇.5 yuan < click auction price average price < 0.8 yuan) Based on the above rules, assume that the user access scenario is as follows: Scenario 1: (User U1, frequently accessing news advertising slots on weekend mornings) Scenario 2: (User U2, frequently accessing blog class advertising slots on weekday evenings) Scenario 3: (user U3, frequently accessing the news advertisement bit in the morning of the working day) The advertisement distribution management device 11 collects the user's access information according to the above three scenarios, and stores the access information in the website, and is in the right After the website log is analyzed, the feature attribute set of each user is extracted. Then, the feature attribute set of the three users can be obtained, respectively, and the feature attribute of the user U is (male, interested in digital, income in the middle, and the feature attribute of the recently purchased touch mobile phone user U2 is (female, pair of dolls). Interested in equipment, income, recently purchased MP3) User U3's characteristic attributes are (female, interested in sports equipment, high income, recently purchased touch phone). Then, referring to FIG. 3, in the embodiment of the present application, The detailed process of the advertisement delivery management device 1 1 for managing the advertisement delivery based on the advertisement delivery effect is as follows: -17- 201133366 Step 300: After determining that a user logs into the website system, obtain a corresponding feature attribute set according to the operation behavior of the user when browsing the website. And selecting, according to the feature attribute set, a matching rule in a preset rule base, the rule is used to select an alternative advertisement that meets the feature attribute of the user. For example, access to the user U (male, digital) Interested, middle income, recently purchased touch phones, visits during weekends, frequent access to news ads Bit), by the function β similarity (¢/, A), the similarity degree 値 of all the rules and u, in the rule base 10 can be calculated, and then the similarity number 値 is sorted, and according to the set threshold 値, the ranking is τ ο The rules of p X, which are the rules found in the rule base that are identical or similar to the feature attributes of user U 1. HSimiiavUy ^) = Π Σ sim{Norm(xi), Norm{yJ.)) > X ' y0F, /7 = ^/;;, /;,^,^,'),i takes 値[a,f], corpse~ is the default used in the rule base to describe various advertising attributes. The set, G ~ is used to form /=;, j is the component contained in 6. Of course, the above F = , ^巧, ^巧) is only an example, in practical applications, if based on the actual application environment, more defined vector sets are added, such as F = , corpse 2, ... A ), the factory is one of the six, then the above formula " "secret _ (to =] ^ stone gamma (gamma / ^ /) body / 7» 1 (brother 0) also applies, which, in addition, yDF > F = {FvF2, ..., Fn) , i take 値 [1, η ] ' F〇 ~ corpse n is a preset used in the rule base to describe various types of advertising attributes, and the ruler is used to form /=; , j is the component contained in the game. • 18- 201133366 After using the search function to play the kiss, the rule R1 can be selected for the user U1: (male user 'interested in digital, income in the middle, recently purchased touch screen mobile phone' often access news advertising space, click The advertisement is MP3 'advertising product price < 2000 yuan, the delivery period is weekend, the advertiser credit for placing the advertisement is greater than 10 points, the advertisement is displayed by fUsh, and the advertisement is delivered with keyword precise method '0.2 yuan < click auction price average price <0.4 yuan). In the actual application, the rule of the final selection may be one, or may be multiple. In the embodiment of the present application, the rule that matches the selected user feature set of the login user is assumed to be R4, R5, and R6. Step 3 1 0: Filter the corresponding candidate advertisement according to the selected rule. For example, 'assume that the selected rule matching the user feature set is R4, R5, R6, then the user ID and the keyword extracted from the selected rule are passed as parameters to the advertisement search engine, and the advertisement search The engine searches for the corresponding alternate ad based on the obtained parameters. In this embodiment, it is assumed that the advertisement search engine selects the corresponding candidate advertisements as advertisement A, advertisement B, and advertisement C according to the selected rules R4, R5, and R6, respectively. Step 3 20: Probabilistic campaign for the candidate advertisements obtained. In the embodiment of the present application, the candidate advertisements are campaigned in the following manner: The candidate advertisements selected according to the rules R4, R5, and R6 are expressed as: <, where i is the corresponding rule subscript 'j indicates specific The number of alternative advertisements obtained 'in this embodiment' is taken as 4' 5' 6. Then, all the selected alternative advertisements are as follows: -19 · 201133366 .7 4 -7 5 7 6 d^^ 14 15 16 Jy4y4 /_V - 4 5 6 ΛΛΛ The probability of the campaign is as follows: According to the calculated probability of the campaign Number 値, the selected rule Ri is reverse sorted, the function used is ^result (x, y) = e 乂Hsimilarity 0, less), where β is the preset effect expansion factor, initially set to 1, the manager It can be optimized according to the test effect selected by the β parameter, S is the effect score of the rule corresponding to y, Gw = (Fa, K, Q, G), and X represents the ad bit vector 匕 and the user feature vector corresponding to the user-specific access. The connection vector of 匕 belongs to Gw, and y represents the connection vector of the advertisement bit component Ffl6 and the user feature component & in the selected rule R, and is also attributed to Fa6a/. For the sorted Ri, τ〇Ρχ (the result of the top X) is selected, and then the corresponding candidate advertisement is determined for the selected ΤορΧ, where χ = 2 ', then the final rule is R4, R5. The corresponding alternative ad is the ad and the ad, expressed as a shorthand for this ad set as Ad. Finally, random sampling is performed on the set Ad again, and the number of samples is Y (according to the parameter setting of the system, assuming Y = 1), then the final probability election result may be advertisement A or advertisement B. Step 3 3 0 : Display the final selected ad. Step 3: monitoring the user's operation behavior for the final advertisement -20-201133366, and according to the collected advertisement delivery effect data, the rule base 10 is entered in the above step 340, after the final selected advertisement is performed. Then, in step 350, the logs generated by the delivery are collected. The main contents of the log record include, but are not limited to, user id time, clicked ad bit, browsed ad bit, and the 〇 of the bookmark is calculated after the time interval from the delivery time, specifically calculating the advertisement The delivery effect data (package score S and support degree N), and then the rules saved in the rule base 10 are updated according to the calculated advertisement delivery effect. In the present application, when the rule base 10 is updated, two types are included. Operation: 1. According to the release effect data, the corresponding new rule is added to the rule base. 1 The bow is based on the advertisement delivery effect data to the existing regularization in the rule base 10. The so-called extraction means that a large number of occurrences (ie, the probability that the probability is greater than a certain advertising performance statistic vector deduction into a rule. For example, the user U 访问 某个 Τ Τ 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问 访问The advertisement bit ρ displays the product detail page p after the user sees the advertisement a, clicks the link advertisement of the advertisement a, and then purchases the product I on the detail, and collects the product J. This series of 'will be recorded by the system as (· U, Τ, W, Ρ, A, I, J) the above-mentioned set C and set D; line update release time record, access products, etc. In the case of the effect data, the advertisement casts 3, 2, and the optimal threshold is 値). The net is wide, and enters the emotional page P operation line. For details, see -21 - 201133366. The user's series of operation actions are recorded. Analyze and collate, corresponding to the feature attribute collection saved as the user. Including: τ is converted into the corresponding time period Ti, the season Ts of the delivery, whether there is a major holiday Tf, etc.;
再把W和P通過廣告用戶關係管理（Customer relationship management, CRM)系統中的廣告位元資料 和已有廣告搜索引擎中的廣告位元文本資料，轉化成規則 庫10需要的廣告位元特徵資料集合，詳見前述的集合A 最後’通過廣告CRM系統中的廣告資料，以及通過 廣告客戶推廣產品系統，獲取A和I的詳細屬性，從而將 二者合倂後轉化爲被投放的廣告特徵資料（詳見前述的集 合F) 因而，用戶的一系列操作行爲（U，T，W，P，A，I ’ J )將被轉換爲前述的廣告效果統計指標向量。 然後根據公式 S = YjwixNorm(F^), /=1 計算廣告效果統計指標向量的效果分數snew和 的支持度Nnew，當Snew>設定的閩値，並且Nnew>設定 的閩値時，若不存在於規則庫〗〇中，則將作爲提 取出的新規則選取進入規則庫1 0，這樣便完成了新規則 的提取。 -22- 201133366 若已存在於規則庫1 0中，則將規則庫1 0中原有 保存的向量的效果分數記錄爲S(3id，支援度記錄爲 N〇ld，接著，依據下述公式計算合倂效果分數： 心卡〜+ (卜解磨 根據計算結果，若心>設定閾値，以及&>設定閾値 ’則將規則庫10中原有保存的規則€加的Send更新爲心 ’將N〇id更新爲#合；若*5合<設定閾値，或者；^<設定間値 ’則將對應的規則巧加從規則庫1 0中刪除，這樣，便完 成了對已有規則的優化。 上述支援度N的計算函數爲: | % | ，其中，設某—個時間段裏面，記 錄的向量集合記爲，X □ 。 另一方面，在上述實施例中，爲了在執行步驟300後 ’較佳地，還需要對選中的規則進行遺傳變異，以在規則 庫1 〇中添加新的規則。可以對選中的規則都進行遺傳變 異，也可以對選中的規則進行隨機抽樣，對抽取到的規則 進行遺傳變異。 本申請實施例中’可以採用的遺傳變異方式包含但不 限於：使用遺傳演算法對步驟3 00中選取的規則進行交叉 變異，其具體爲： 假設進行遺傳變異的規則爲R4= ’和115=(^/^，€，尸/，/^)’那麼， -23- 201133366 首先’將規則R4，R5進行編碼，可以採用自然編碼 方式。 接著’選規則R4和R5的變異點，爲了減少變異產 生大量無用後代，變異點可以選擇舄巧和/^間的位置’ 具體位置如下面的雙豎線所示： {Fa^Fb,Fc,Fd \\Fe,FfiFg) 那麼，便可以將^:^/^/^^&尸^根據變異點的 位置，拆分爲： (Fa，Fb，Fc，Fd、m(Fe，Ff，Fg) 然後，將拆分出的向量進行交叉連接： (6，Α，6Λ)·和（^^/，^^連接得到队，〜^/；)·，/^/^/1；) 這樣，便獲得的遺傳變異後的新規則 在上述實施例中，對已有規則進行遺傳變異，可以在 基於歷史效果選取top最優的規則的同時，給予廣告投放 策略適當機率的“變異”，這些變異保證了規則庫1 〇的“進 化”，可以發現和挖掘新的規則，有利於拓展廣告的投放 模式。 綜上所述，本發明實施例中，爲了對好的投放經驗進 -24- 201133366 行積累，引入了規則庫1 〇的槪念，它針對過去廣告投放 後帶來的諸多效果，依據投放關聯的諸多因素進行分類， 並對每一類別的投放效果中較好的部分進行統計歸納，總 結出每類投放中較優的一些投放匹配規則，並線上不斷進 行遺傳進化，積累出來指導日後的規則庫1 0的更新。這 樣，使得基於規則庫1 0的廣告投放變得簡單易行，能夠 較好地實現廣告投放的全局最優。另一方面，除了線上指 導投放，規則庫1 〇也是經驗的總結保留，線下可以指導 業務的發展和創新等。 規則庫10的建立和_化均直接依據於廣告投放效果 ，廣告投放效果的有所變化，將會通過規則庫1 0即時地 反應在其保存的用於指導廣告選擇的各類規則上，使得廣 告的選擇完全依賴於其投放效果，這樣，便形成了 ：廣告 投放一一投放效果跟蹤——規則更新——廣告再投放，這樣 一種大的投放迴圈，從而令目的和手段得到了統一。簡而 言之，規則庫1 〇的更新進化是即時的基於廣告效果來實 現的，可以令各種規則的優化得以自動化和即時化，具有 實現代價小，週期短和優化速度快等優點。這樣，便無需 盲目地減少廣告投放量，而是根據用戶的實際需求有目的 有針對性的投放相應的廣告，而減少不必要廣告的投放量 ，從而在保證廣告投放效果的基礎上，減少了網站廣告在 投放時傳輸的資料量，提高了系統的資料傳輸速度，進而 提升了網站的服務品質。 顯然，本領域的技術人員可以對本發明中的實施例進 -25- 201133366 行各種改動和變型而不脫離本發明的精神和範圍。這樣， 倘若本發明實施例中的這些修改和變型屬於本發明申請專 利範圍及其等同技術的範圍之內，則本發明中的實施例也 意圖包含這些改動和變型在內。 【圖式簡單說明】 圖1爲本申請實施例中廣告投放管理系統體系架構圖 圖2爲本申請實施例中廣告投放管理裝置功能結構圖 9 圖3爲本申請實施例中基於廣告投放效果對廣告投放 進行管理流程圖。 【主要元件符號說明】 I 〇 :規則庫 Π :廣告投放管理裝置 12 :廣告搜尋引擎 II 0 :獲取單元 1 1 1 :第一處理單元 Π 2 :第二處理單元 -26-Then, W and P are converted into the advertising bit feature data required by the rule base 10 through the advertising bit metadata in the customer relationship management (CRM) system and the advertising bit text data in the existing advertising search engine. Collection, see the above-mentioned collection A. Finally, 'through the advertising materials in the advertising CRM system, and through the advertiser's promotion product system, obtain the detailed attributes of A and I, and then merge the two into the delivered advertising characteristics. (See the set F above). Thus, a series of user behaviors (U, T, W, P, A, I ' J ) will be converted to the aforementioned advertising effect statistical indicator vector. Then, according to the formula S = YjwixNorm(F^), /=1, the effect score snew of the advertisement effect statistical indicator vector and the support degree Nnew are calculated. When Snew> is set, and Nnew> is set, if it does not exist, In the rule base ,, the new rule extracted as the rule is entered into the rule base 10, thus completing the extraction of the new rule. -22- 201133366 If it already exists in the rule base 10, the effect score of the original saved vector in the rule base 10 is recorded as S (3id, the support degree record is N〇ld, and then, according to the following formula倂 effect score: heart card ~ + (bundle according to the calculation result, if the heart > set threshold 値, and &> set threshold 値 ' will update the original saved rule in the rule base 10 plus the heart 'will N〇id is updated to #合; if *5合<Setting threshold 値, or;^<Setting 値' then the corresponding rule is deleted from the rule base 10, thus completing the existing rule The calculation function of the above support degree N is: | % | , wherein, in a certain time period, the set of records is recorded as X □. On the other hand, in the above embodiment, in order to execute the steps After 300 'better, it is also necessary to genetically mutate the selected rules to add new rules in the rule base 1. You can genetically mutate the selected rules, or you can randomly select the selected rules. Sampling, genetic variation of the extracted rules The genetic variation methods that can be used in the embodiments of the present application include, but are not limited to, cross-mutation of the rules selected in step 300 using a genetic algorithm, which is specifically: The rule for performing genetic variation is R4= 'and 115= (^/^, €, corpse/, /^) 'then, -23- 201133366 First of all, 'code the rules R4, R5, you can use the natural coding method. Then 'select the variation points of the rules R4 and R5, in order to reduce the variation Produce a large number of useless offspring, the variation point can choose the position between the well-behaved and /^' specific position as shown in the double vertical line below: {Fa^Fb, Fc, Fd \\Fe, FfiFg) Then, you can ^: ^/^/^^& corpse ^ is split according to the position of the mutated point: (Fa, Fb, Fc, Fd, m (Fe, Ff, Fg) Then, the split vectors are cross-connected: 6, Α, 6Λ)· and (^^/, ^^ are connected to get the team, ~^/;)·, /^/^/1;) In this way, the new rule after the genetic variation obtained is in the above embodiment. Genetic variation of existing rules can give appropriate probability of advertising strategy while selecting top optimal rules based on historical effects. The "variation", these variations ensure the "evolution" of the rule base, and can discover and mine new rules, which is conducive to expanding the delivery mode of the advertisement. In summary, in the embodiment of the present invention, in order to facilitate the delivery Experience into the accumulation of -24,33,336,366, introduced the sacred concept of the rule base, which is based on the many effects of past advertising, based on the many factors of the association, and The good part is statistically summarized, summarizing some of the best matching matching rules in each type of delivery, and continuously carrying out genetic evolution on the line, accumulating to guide the update of the rule base 10 in the future. In this way, the advertisement placement based on the rule base 10 is made simple and the global optimization of the advertisement delivery can be better achieved. On the other hand, in addition to online guidance, the rule base 1 is also a summary of experience, and offline can guide business development and innovation. The establishment and _ification of the rule base 10 are directly based on the effect of the advertisement delivery, and the effect of the advertisement delivery is changed, and will be instantly reflected by the rule base 10 in various rules for saving advertisement selection. The choice of advertising depends entirely on its delivery effect. In this way, it is formed: the advertisement delivery one-to-one delivery effect tracking - the rule update - the advertisement re-delivery, such a large delivery loop, so that the purpose and means are unified. In short, the update evolution of the rule base 1 是 is realized based on the immediate effect of advertising, which can automate and instantiate the optimization of various rules, and has the advantages of low cost, short cycle and fast optimization. In this way, there is no need to blindly reduce the amount of advertising, but the targeted advertising is targeted according to the actual needs of the user, and the amount of unnecessary advertisements is reduced, thereby reducing the effectiveness of the advertising. The amount of data transmitted by the website advertisements at the time of delivery increases the speed of data transmission of the system, thereby improving the service quality of the website. It is apparent that those skilled in the art can make various modifications and variations to the embodiments of the present invention without departing from the spirit and scope of the invention. Thus, the present embodiments of the invention are intended to cover such modifications and variations, and the modifications and variations of the invention are intended to be included within the scope of the invention. BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a schematic diagram of an architecture of an advertisement delivery management system according to an embodiment of the present application. FIG. 2 is a functional structure of an advertisement delivery management apparatus according to an embodiment of the present application. FIG. Ad delivery management flow chart. [Description of main component symbols] I 〇 : Rule base Π : Trafficking device 12 : Advertising search engine II 0 : Acquisition unit 1 1 1 : First processing unit Π 2 : Second processing unit -26-