Case 2:15-cv-02049 Document 1 Filed 12/03/15 Page 1 of 64 PageID #: 1 IN THE UNITED STATES DISTRICT COURT FOR THE EASTERN DISTRICT OF TEXAS MARSHALL DIVISION FELLOWSHIP FILTERING TECHNOLOGIES, LLC, Plaintiff, Civil Action No._________ v. JURY TRIAL DEMANDED ALIBABA.COM, INC.; ALIBABA SINGAPORE E-COMMERCE PRIVATE LTD.; ALIBABA GROUP HOLDING LTD.; ALIBABA.COM HONG KONG LTD.; ALIBABA.COM LTD.; ALIBABA.COM INVESTMENT HOLDING LTD.; ALIBABA.COM INVESTMENT LTD.; ALIBABA (CHINA) TECHNOLOGY CO., LTD.; TAOBAO HOLDING LTD.; TAOBAO CHINA HOLDING LTD.; AND TAOBAO (CHINA) SOFTWARE CO. Defendants. COMPLAINT FOR PATENT INFRINGEMENT Plaintiff Fellowship Filtering Technologies, LLC (“Fellowship Filtering” or “Plaintiff”), by and through its attorneys, brings this action and makes the following allegations of patent infringement relating to U.S. Patent No. 5,884,282 (“the ‘282 patent”). Defendants Alibaba.com, Inc., Alibaba Singapore E-commerce Private Ltd., Alibaba Group Holding Ltd., Alibaba.com Hong Kong Ltd., Alibaba.com Ltd.,Alibaba.com Investment Holding Ltd., Alibaba.com Investment Ltd., Alibaba (China) Technology Co., Ltd., Taobao Holding Ltd., Taobao China Holding Ltd., and Taobao (China) Software Co., Ltd. (collectively, “Alibaba” or “Defendant”) infringes Fellowship Filtering’s ‘282 patent in violation of the patent laws of the United States of America, 35 U.S.C. § 1 et seq. INTRODUCTION 1. In an effort to expand its product base and profit from the sale of infringing computer-based data analytics technologies, Alibaba has undertaken to copy the technologies and Case 2:15-cv-02049 Document 1 Filed 12/03/15 Page 2 of 64 PageID #: 2 inventions of Gary Robinson, the inventor or the ‘282 patent and a co-owner of Fellowship Filtering. 2. Alibaba runs an online global marketplace through its websites Alibaba.com, AliExpress.com, and Taobao.com (collectively, the “Alibaba Websites”) that implement infringing technologies to facilitate the sale of products. Alibaba positions its data analytics systems as providing customers and suppliers with revolutionary mechanisms for gaining insights into customer behavior that facilitate the sale of products. Alibaba’s data analytics systems incorporate the inventions disclosed in Mr. Robinson’s ‘282 patent. “[W]e offer you personalized recommendations based on your browsing and search history, e.g. newly displayed products and other buyers’ preferred products relating to your search terms.”1 European and U.S. Patents assigned to Alibaba have cited Mr. Robinson’s work as relevant prior art.2 3. Quan Yuan, Technical Director of TaoBao Recommendation has described Alibaba’s use of predictive algorithms as integral to the success of Alibaba’s products. The below slide from a 2014 presentation by Mr. Yuan shows how Alibaba provides recommendations using collaborative filtering techniques developed by Mr. Robinson. 1 Alibaba Product Recommendations, ALIBABA WEBSITE (last visited December 1, 2015), available at: http://www.alibaba.com/recommended-products.html; see also Alibaba Group Holding Limited Form 20-F, ALIBABA GROUP SECURITIES AND EXCHANGE FILING at 89 (June 2015) (“Alibaba Group’s mobile products can help grow revenue with location-based services and targeted product recommendations to individual consumers.”). 2 See U.S. Patent No. 8,234,291, WO Patent App. 200846338, and EP Patent App. 2075720 (all assigned to Alibaba Group Holding Limited and citing Mr. Robinson’s paper regarding effective filtering entitled “A Statistical Approach to the Spam Problem” as prior art.). 2 Case 2:15-cv-02049 Document 1 Filed 12/03/15 Page 3 of 64 PageID #: 3 Qiang Yan and Quan Yuan (Tao Search and P13N Team), LARGE SCALE RECOMMENDATION IN E- COMMERCE (October 10, 2014), available at: http://www.slideshare.net/scmyyan/large-scale- recommendation-in-ecommerce-qiang-yan. 4. Patents and patent applications assigned to Alibaba Group Holding Limited have described the use of Mr. Robinson’s inventions as improving the functioning of computer systems and enabling the delivery of relevant computer based recommendations. If similar, the exemplary embodiments determine that the similarity score between the attribute value of the nominal attribute of the first product and the attribute value of the nominal attribute of the second product is relatively high. Otherwise, the similarity score is relatively low. As a result, the exemplary embodiments can determine a similarity score based on the semantic meaning that is implicitly included in attribute values and thereby improve the accuracy of computing a similarity score between values of a nominal attribute. U.S. Patent App. 13/381,822, Method and Apparatus of Determining A Linked List Of Candidate Products (filed October 18, 2011; published August 1, 2013) (emphasis added) (This patent application is assigned to Alibaba Group Holding Limited and cites as prior art two patents (of six) that reference the patent-in-suit as relevant prior art.). It is apparent that an effective recommended result is crucial since an aimless recommendation causes low acceptance of the recommended result and a waste of computing resource. U.S. Patent App. 14/028,279, Recommending Product Information (filed September 16, 2013; published March 20, 2014) (emphasis added) (patent application assigned to Alibaba Group Holding Limited). 3 Case 2:15-cv-02049 Document 1 Filed 12/03/15 Page 4 of 64 PageID #: 4 5. Mr. Robinson is a mathematician and inventor of computer-based recommendation engine technologies that enable the recommending of products and/or content based on novel algorithms that calculate the preferences based on the similarity and dissimilarity of users of a website. 6. Mr. Robinson studied mathematics at Bard College and New York University's Courant Institute of Mathematical Sciences. Mr. Robinson is the recipient of the National Science Foundation – SBIR award. 7. Mr. Robinson is a named inventor of numerous United States Patents. Mr. Robinson’s patents have been acquired by companies including Google, Inc. (“Google”).3 Patents referencing Mr. Robinson’s ‘282 patent have been purchased or assigned to companies including: International Business Machines Corporation (“IBM”), 4 Google,5 Amazon.com, Inc. (“Amazon”),6 and Intel Corporation (“Intel”).7 ROBINSON’S LANDMARK ELECTRONIC MAIL INVENTIONS 8. The Robinson Method, named after Gary Robinson, is a Bayesian statistical approach that uses a text-classifier, rule-based method for determining the relevancy of an email message. Numerous leading SPAM filtering technologies utilize the Robinson Method.8 3 See USPTO Assignment Abstract of Title Database Reel/Frame No. 021552/0256. 4 U.S. Patent Nos. 6,356,879; 6,931,397; 7,006,990; 7,080,064; 7,099,859; 7,389,285; 7,885,962; 8,700,448; and 8,825,681. 5 U.S. Patent Nos. 7,966,632; 8,290,964; and 8,762,394. 6 U.S. Patent Nos. 6,266,649; 7,113,917; 7,433,832; 7,478,054; 7,664,669; 7,778,890; 7,908,183; 7,921,042; 7,945,475; 8,001,003; 8,024,222; 8,108,255; 8,140391; and 8,180,689. 7 U.S. Patent Nos. 6,405,034, 7,590,415, and 7,797,343. 8 Ricardo Villamarín-Salomón & José Carlos Brustoloni, Bayesian Bot Detection Based on DNS Traffic Similarity, in SAC’09: ACM SYMPOSIUM ON APPLIED COMPUTING 2040—41 (2009); Masahiro Uemura & Toshihiro Tabata, Design and Evaluation of a Bayesian-filter-based Image Spam Filtering Method, in PROCEEDINGS OF THE 2008 INTERNATIONAL CONFERENCE ON INFORMATION SECURITY AND ASSURANCE 46-51 (2008) (“the Robinson Method”); MARCO ANTONIO BARRENO, Technical Report No. UCB/EECS-2008-63, EVALUATING THE SECURITY OF MACHINE LEARNING ALGORITHMS 45 (2008); Manabu Iwanaga et al., Evaluation of Anti-Spam Methods Combining Bayesian Filtering and Strong Challenge and Response, in PROCEEDINGS OF CNIS’03 (COMMUNICATION, NETWORK, AND INFORMATION SECURITY) 214—19 (2003); BLAINE NELSON, Technical Report No. UCB-EECS-2010-140, BEHAVIOR OF MACHINE LEARNING ALGORITHMS IN ADVERSARIAL ENVIRONMENTS 62-67 (2010); Gordon V. Cormack & Mona 4 Case 2:15-cv-02049 Document 1 Filed 12/03/15 Page 5 of 64 PageID #: 5 9. Mr. Robinson’s contributions to the field of electronic mail filtering are recognized as landmark technologies.9 Günther Hölbling, PERSONALIZED MEANS OF INTERACTING WITH MULTIMEDIA CONTENT 119 (2011). 10. Mr. Robinson has published academic articles on statistical approaches to identifying content. A 2003 article in Linux Journal described these mathematical approaches for identifying unsolicited bulk email. Mr. Robinson’s approach was notable because it assigned scores to both “spam” and “ham” and used an algorithm to guess intelligently whether an incoming email was spam. This approach was incorporated in products such as SpamAssassin, which used a Bayesian statistical approach using a text-classifier rule to distinguish “spam” and “ham” messages.10 11. Mr. Robinson’s inventions relating to filtering technologies have been widely adopted by spam filters including Spam Assassin11 (PC Magazine’s Editor’s Choice for spam Mojdeh, Autonomous Personal Filtering Improves Global Spam Filter Performance, in PROCEEDINGS OF THE 6TH CONFERENCE ON EMAIL AND ANTI-SPAM 2 (2009). 9 See also U.S. Patent No. 8,234,291, WO Patent App. 200846338, and EP Patent App. 2075720 (all assigned to Alibaba Group Holding Limited. 10 Gary Robinson, A Statistical Approach to the Spam Problem, LINUX JOURNAL 107 (2003). 11 SpamAssassin Pro, in PC MAGAZINE February 25, 2003 at 82 (awarding SpamAssassin Pro its editors’ choice award); The SpamAssassin Project: Train SpamAssassin's Bayesian Classifier, http://spamassassin.apache.org/full/3.2.x/doc/sa-learn.html (“Gary Robinson's f(x) and combining algorithms, as used in SpamAssassin”); Credits - The Perl Programming Language - Algorithms, http://cpansearch.perl.org/src/JMASON/Mail-SpamAssassin-3.2.5/CREDITS (“The Bayesian-style text classifier used by SpamAssassin's BAYES rules is based on an approach outlined by Gary Robinson. Thanks, Gary!”). 5 Case 2:15-cv-02049 Document 1 Filed 12/03/15 Page 6 of 64 PageID #: 6 filtering), SpamSieve12 (MacWorld’s Software of the Year), and SpamBayes13 (PC Worlds Editor’s Choice for spam filtering). ROBINSON’S DEVELOPMENT OF CONTENT FILTERING SYSTEMS 12. Prior to developing groundbreaking electronic mail filtering technologies, Mr. Robinson used his insights to develop the automated content filtering technologies that are used today by Alibaba and many of the world’s largest corporations without attribution or compensation. 13. In the late 1980’s, Mr. Robinson developed a system for collecting preference information and providing recommendations. His company, 212-ROMANCE, was an automated, voice-based dating service that used a passive data collection process to determine likely romantic matches.14 Mr. Robinson’s contributions to the field of content filtering were pioneering. Matthew French, Romantic Beginnings Have Worldwide Effect, BOSTON BUS. J., May 20, 2002. 12 David Progue, From the Deck of David Progue: The Follow-Up Edition, N.Y. TIMES, April 5, 2006, http://www.nytimes.com/2006/04/05/technology/06POGUE-EMAIL.html (“Spam Sieve is just incredibly, amazingly accurate; my in box is clean, baby, clean!”). 13 Tom Spring, Spam Slayer: 2003 Spam Awards, PCWORLD MAGAZINE, December 15, 2003, at 36 (“What makes the program unique is that SpamBayes doesn't use predetermined spam definitions. Rather, it constantly evolves by scanning your in-box to build custom definitions.”); MARCO ANTONIO BARRENO, Technical Report No. UCB/EECS-2008-63, EVALUATING THE SECURITY OF MACHINE LEARNING ALGORITHMS 45 (2008) (“SpamBayes classifies using token scores based on a simple model of spam status proposes by Robinson . . . . SpamBayes Tokenizes the header and body of each email before constructing token spam scores. Robinson’s method assumes that each token’s presence of absence in an email affects that email’s spam status independently from other tokens.”). 14 212-Romance was incorporated under the name Microvox Systems, Inc. 6 Case 2:15-cv-02049 Document 1 Filed 12/03/15 Page 7 of 64 PageID #: 7 14. In the mid-1990s, Mr. Robinson recognized that the growing adoption of the internet and increased computational power enabled collection and processing of data relating to customer and user preferences that, with proper data analytics processes, could provide accurate recommendations of products and content. 15. Mr. Robinson further recognized that the growth of the internet led to unique problems involving information overload that filtering techniques using specific new collaborative filtering technologies could solve. 16. At the time, existing recommendation technologies, discussed in the ‘282 patent, failed to teach a robust and accurate process for providing recommendations. A key insight of Mr. Robinson was that the input of buying habits and/or ratings information from multiple users over the internet allowed similarity values among users to be calculated based on identifying subgroups of similar users. 17. Mr. Robinson invented an automated collaborative filtering (“ACF”) system that received and stored data based on internet users’ purchasing history, preferences, and/or buying history. When a new user accessed the ACF system through a website (in one embodiment), the ACF system recommended further content (e.g., products) based on the similarity values for the first user as compared with other users that previously provided preference data to the ACF system. 18. Mr. Robinson worked to develop novel systems and processes designed to provide accurate content and product recommendations using data stored, collected, and computed on specific computer-based systems. Mr. Robinson’s insights led to the patent application resulting in the '282 patent. 19. The patent-in-suit - the ‘282 patent - is a pioneering patent in the field of data analytics. The ‘282 patent uses novel algorithmic approaches to provide accurate recommendations of products and content using data analysis specific to a computer system. 7 Case 2:15-cv-02049 Document 1 Filed 12/03/15 Page 8 of 64 PageID #: 8 Jonathan A. Zdziarski, ENDING SPAM: BAYESIAN CONTENT FILTERING AND THE ART OF STATISTICAL LANGUAGE CLASSIFICATION 269 (2005). 20. The ‘282 patent has been cited by over 443 United States patents and patent applications as prior art before the United States Patent and Trademark Office.15 Companies whose patents cite the ‘282 patent include: • OpenText S.A. • Accenture Global Services GMBH • YellowPages.com LLC • Nielsen Holdings N.V. • International Business Machines Corporation • Koninklijke Philips N.V. • Google, Inc. • Amazon.com, Inc. • Microsoft Technology Licensing LLC • Arbor Networks, Inc. • Johnson & Johnson Consumer Companies • S.C. Johnson & Son Inc. • Sony Electronics, Inc. • Infosys Ltd. • Parasoft Corporation • AT&T Intellectual Property LLP • Dish Network LLC • eBay, Inc. • Rovi Corporation • CBS Interactive, Inc. • American Express Company • Hewlett-Packard Company • Xerox Corp. • Capital One Financial Corporation • JDA Software Group, Inc. • State University of New York • Robert Bosch Healthcare System, Inc. • Netflix, Inc. 15 The 443 forward citations to the ‘282 patent do not include patent applications that were abandoned prior to publication in the face of the ‘282 patent. 8 Case 2:15-cv-02049 Document 1 Filed 12/03/15 Page 9 of 64 PageID #: 9 • Intel Corporation • Tribune Media Company • Ingenio, LLC • Recommend, Inc. • Dassault Systemes S.A. • Pandora Media, Inc. • Pace plc • Regents of the University of California • Facebook, Inc. • Numera, Inc. 21. Patents citing Mr. Robinson’s ‘282 patent as prior art have been asserted by Amazon.com, Inc. (“Amazon”) and Netflix, Inc. (“Netflix”) in patent infringement cases: • Amazon asserted U.S. Patent No. 6,266,649, entitled “Collaborative Recommendations Using Item-to-Item Similarity Mappings,” against Discovery Communications, Inc. (“Discovery”). The ‘649 patent claimed a priority date of September 1998 (subsequent to the ‘282 patent). Amazon’s ‘649 patent cited Mr. Robinson’s ‘282 patent as prior art during prosecution before the Patent and Trademark Office. After two years of litigation, Discovery took a license to Amazon’s ‘649 patent (prior to claim construction being adjudicated).16 • Netflix asserted U.S. Patent No. 7,024,381, claiming a priority date of April 2000, against Blockbuster LLC (“Blockbuster”). The ‘381 patent referenced the ‘282 patent as prior art. A settlement and license agreement was reached between Netflix and Blockbuster on the verge of trial.17 • Robert Bosch Healthcare Systems, Inc. (“Robert Bosch”) asserted U.S. Patent Nos. 7,223,235 & 7,223,236 against MedApps, Inc (“MedApps”). The ‘235 and ‘236 patents cite Mr. Robinson’s ‘282 patent as prior art. MedApps reached a settlement and license with Robert Bosch roughly one year after the infringement action was initiated.18 • Black Hills Media LLC (“Black Hills”) asserted U.S. Patent Nos. 8,028,323, 8.230,099, and 8,458,356. The ’323, ‘099, and ‘356 patents referenced Mr. Robinson’s ‘282 patent as prior art. Black Hills settled a majority of its cases following denial of summary judgment of invalidity.19 • i2 Technologies, Inc. (“i2”) asserted U.S. Patent No. 7,370,009 against Oracle in the Eastern District of Texas. Subsequently, Oracle asserted four patents against i2’s parent, 16 Amazon.com Inc v. Discovery Communications Inc., Case No. 09-cv-00681 Dkt. Nos. 122 & 166 (W.D. Wash.). 17 Netflix, Inc. v. Blockbuster, Inc., Case No. 06-cv-02361 Dkt. No. 239 (Cal. N.D.). 18 Robert Bosch Healthcare Systems, Inc. -v- MedApps, Inc. Case No. 12-cv-00113 Dkt. No. 64 (Cal. N.D.); US. Patent No. 8,028,323 Information Disclosure Statement (March 3, 2010). 19 Black Hills Media LLC v. Sonos, Inc., Case No. 14-cv-00486 Dkt. Nos. 129 & 169 (Cal. C.D.). 9 Case 2:15-cv-02049 Document 1 Filed 12/03/15 Page 10 of 64 PageID #: 10 JDA Software Group. Following a year of litigation, the parties reached a settlement in March 2011.20 22. Cases against Oracle, Discovery and Blockbuster underscore the inventive nature of the ‘282 patent, as the above asserted cases involve patents referencing Mr. Robinson’s ‘282 patent as prior art. 23. The claims in the ‘282 patent are directed at solving a problem that did not arise in prior art systems, i.e. generating preference data from large data sets. In prior art systems, the sample size of users was typically very small, and thus the need for a process that takes into account unusual similarities was not at issue. There is no question pre-electronic recommendation systems are significantly different from computer and/or internet-based recommendation systems. The speed, quantity, and variety of rating information markedly differ from the objectives and data available to recommendation systems existing before modern, computer and/or internet-based systems. Differences between the analog versions of preference systems and the invention disclosed in the ‘282 patent diverge significantly. 24. The use of ratings data and probability values to make recommendations over a computer network was not a longstanding or fundamental economic practice at the time of the invention disclosed in the ‘282 patent. Nor at the time was the use of ratings data and probability values to make recommendations a fundamental principle in ubiquitous use on the internet or computers in general. Dr. Zeynep Tufekci of Harvard University’s Berkman Center for Internet and Society described recommendation engine systems such as the systems disclosed in the ‘282 patent as being far from a “law of nature.” The fear I have is that every time this is talked about, people talk about it as if it's math or physics, therefore some natural, neutral world. And they're 20 i2 Technologies, Inc. et al v. Oracle Corporation et al., Case No. 10-cv-00284 Dkt. Nos. 85 & 130 (E.D.Tex.) (i2 asserted several predictive analytics patents against Oracle); Erin Coe, I2, Oracle Resolve Software Patent Battle, LAW360, March 4, 2011, http://www.law360.com/articles/229787/i2-oracle-resolve-software-patent-battle. 10