Ex Parte Ozonat et alDownload PDFPatent Trial and Appeal BoardMay 26, 201613563030 (P.T.A.B. May. 26, 2016) Copy Citation UNITED STA TES p A TENT AND TRADEMARK OFFICE APPLICATION NO. FILING DATE FIRST NAMED INVENTOR 13/563,030 07 /31/2012 Mehmet Kivanc Ozonat 56436 7590 05/31/2016 Hewlett Packard Enterprise 3404 E. Harmony Road Mail Stop 79 Fort Collins, CO 80528 UNITED STATES DEPARTMENT OF COMMERCE United States Patent and Trademark Office Address: COMMISSIONER FOR PATENTS P.O. Box 1450 Alexandria, Virginia 22313-1450 www .uspto.gov ATTORNEY DOCKET NO. CONFIRMATION NO. 83013403 7115 EXAMINER PHAM, KHANH B ART UNIT PAPER NUMBER 2166 NOTIFICATION DATE DELIVERY MODE 05/31/2016 ELECTRONIC Please find below and/or attached an Office communication concerning this application or proceeding. The time period for reply, if any, is set in the attached communication. Notice of the Office communication was sent electronically on above-indicated "Notification Date" to the following e-mail address( es): hpe.ip.mail@hpe.com mkraft@hpe.com chris.mania@hpe.com PTOL-90A (Rev. 04/07) UNITED STATES PATENT AND TRADEMARK OFFICE BEFORE THE PATENT TRIAL AND APPEAL BOARD Ex parte MEHMET KIVANC OZONAT and CLAUDIO BARTOLINI Appeal2014-009536 Application 13/563,030 Technology Center 2100 Before JOHN A. JEFFERY, BRADLEY W. BAUMEISTER, and DENISE M. POTHIER, Administrative Patent Judges. JEFFERY, Administrative Patent Judge. DECISION ON APPEAL Appellants appeal under 35 U.S.C. § 134(a) from the Examiner's decision to reject claims 1-15. We have jurisdiction under 35 U.S.C. § 6(b). We affirm-in-part. Appeal2014-009536 Application 13/563,030 STATEMENT OF THE CASE Appellants' invention extracts tags and keywords from social network content. Spec. i-f 8. For example, a company's employees can share opinions, knowledge, and expertise on an enterprise social network. Id. i-f 6. But they may not always tag and thematize content. Id. i-f 7. Tags and keywords, however, are useful for searching. Id. i-f 8. So Appellants' invention uses pattern recognition and a semantics graph to extract tags and keywords from content. Id. i-f 9. Claim 1, reproduced below with our emphasis, is illustrative: 1. A computer-implemented method for keyword extraction, comprising: extracting a number of keywords from a portion of content inside an enterprise social network, wherein the portion is determined using a pattern recognition model; constructing a semantics graph based on the number of keywords; and determining content themes within the enterprise social network based on the constructed semantics graph. THE REJECTION The Examiner rejected claims 1-15 under 35 U.S.C. § 103(a) as unpatentable over Jure Leskovec et al., Learning Sub-structures of Document Semantic Graphs for Document Summarization (Aug. 2004) ("Leskovec") and Kivanc Ozonat, Keyword Extraction and Multi-view 2 Appeal2014-009536 Application 13/563,030 Clustering Trees for Search and Retrieval in Customer Product Forums (Jan. 2010) ("Ozonat"). Final Act. 2-9. 1 CLAIMS 1-11AND13 Contentions The Examiner finds that Leskovec discloses every recited element of claims 1, 5, and 11 except for the recited social network content and the portion from which the keywords are extracted, but cites Ozonat as teaching this feature in concluding that the claim would have been obvious. Final Act. 2-3 (rejecting claim 1), 4--5 (rejecting claim 5), 7-8 (rejecting claim 11). In the combination, the Examiner finds that Ozonat's co- occurrence model corresponds to the recited pattern-recognition model. Ans. 2-3. Likewise, the Examiner finds that Ozonat's most frequently observed "M" words are the recited portion. Id. at 3. Regarding claim 1, Appellants argue that Ozonat lacks the recited portion from which the keywords are extracted. App. Br. 7-8; Reply Br. 2-5. In Appellants' view, the recited portion refers to structures such as sections, titles, sub-titles, lists, tables, or link descriptors within a corpus. App. Br. 8 (citing Spec. i-fi-133-34); see also Reply Br. 3-5. According to Appellants, Ozonat extracts keywords from a company- supported forum. App. Br. 7; Reply Br. 2. But Appellants contend that 1 Throughout this Opinion, we refer to (1) the Final Office Action mailed February 11, 2014 ("Final Act."); (2) the Appeal Brief filed June 6, 2014 ("App. Br."); (3) the Examiner's Answer mailed July 3, 2014 ("Ans."); and (4) the Reply Brief filed September 3, 2014 ("Reply Br."). 3 Appeal2014-009536 Application 13/563,030 Ozonat' s forum is not the recited portion of content, and it is not determined using a pattern recognition model. App. Br. 7; Reply Br. 2. Appellants further contend that Ozonat's co-occurrence matrix is also not the recited portion. Reply Br. 3. According to Appellants, Ozonat's value "M" is a pre-selected integer less than the total number of the repository's keywords. Id. Appellants argue that Ozonat extracts keywords from this repository based on two approaches: word frequency and term co- occurrence. App. Br. 8; Reply Br. 4. But in Appellants' view, neither approach determines a portion using a pattern recognition model. App. Br. 8; Reply Br. 4. Regarding claims 5 and 11, Appellants argue that Leskovec does not ( 1) automatically assign tags from within a portion of the repository to a number of keywords within received content, as recited in claim 5, or (2) determine words from within a portion of the retrieved content, as recited in claim 11. App. Br. 8-9; Reply Br. 5-7. According to Appellants, Leskovec assigns semantic tags to nodes, not the recited keywords and words. App. Br. 9. Moreover, Appellants argue that Leskovec does not describe the portion recited in claims 5 and 11 for the above-discussed reasons. Reply Br. 5-7. Issues I. Under§ 103, has the Examiner erred in rejecting claims 1, 5, and 11 by finding that Leskovec and Ozonat collectively would have taught or suggested a portion determined using a pattern recognition model? 4 Appeal2014-009536 Application 13/563,030 II. Under§ 103, has the Examiner erred in rejecting claims 5 and 11 by finding that Leskovec and Ozonat collectively would have taught or suggested (1) assigning tags and words from a portion of the content, as recited in claim 5? (2) determining words relevant to a domain of interest from within a portion of the content, as recited in claim 11? Analysis I On this record, we find no error in the Examiner's obviousness rejection of representative claim 1, which recites, in pertinent part, "the portion is determined using a pattern recognition model." "Although an inventor is indeed free to define the specific terms used to describe his or her invention, this must be done with reasonable clarity, deliberateness, and precision." In re Paulsen, 30 F.3d 1475, 1480 (Fed. Cir. 1994). Here, Appellants have not shown, and we do not find, that the term "portion" is defined anywhere in the Specification. See App. Br. 7-8; Reply Br. 2-5. Unlike other terms, 2 Appellants do not define the term "portion." The Specification merely provides examples of "a portion of content" inside an enterprise social network-i.e., sections, titles, sub-titles, lists, tables, or link descriptors within a corpus. Spec. i-fi-1 33-34 (cited at App. Br. 8). 2 See, e.g., Spec. i-f 12 (defining "a number of'), i-f 56 (defining "logic"). 5 Appeal2014-009536 Application 13/563,030 These non-limiting, exemplary embodiments do, however, inform our construction of the limitation at issue. Based upon the Specification's disclosure, we understand that the recited portion encompasses a variety of different portions of content inside an enterprise social network. This breadth is significant, for nothing on this record precludes the recited "portion" from constituting, among other things, Ozonat' s set of M words corresponding to the rows of the co-occurrence matrix. Specifically, Ozonat constructs a co-occurrence matrix with N rows and M columns. 3 Ozonat § 3 .1. Each row represents one word from all distinct words in a repository of words. Id. Each column represents one word from a subset of all distinct words. Id. Specifically, this subset contains the M most frequently observed words. Id. Typically, Ozonat preselects M to be 500. Id. The matrix element in the nth row and mth column is the number of times that words n and m occur together-i.e., their "co-occurrence." Id. Because Ozonat's M most frequently observed words are a subset of all distinct words, the Examiner's finding that these M words are a "portion" of all distinct words (Ans. 3) is reasonable. And these words are from an enterprise social network, as required by the claim. In particular, these words come from a repository ( Ozonat § 3 .1 ), as noted by Appellants (App. Br. 8; Reply Br. 4). This repository contains threads from an online product discussion forum. Ozonat § 1. We see no reason, and Appellants do not argue persuasively, why a "company-supported discussion forum" cannot be an enterprise social network. See, e.g., App. Br. 7 (referring to Ozonat' s "[ o ]nline product discussion forum[]" as a "company supported 3 "N" and "M" are variables representing numbers. Ozonat § 3 .1. 6 Appeal2014-009536 Application 13/563,030 [discussion] forum"). Accordingly, we see no error in the Examiner's finding that the M words are a portion of the N distinct words, and thus, also a portion of an enterprise social network. Because the Examiner finds that the M words in Ozonat' s co- occurrence matrix corresponds to "the recited portion" (Ans. 3), Appellants' contention, then, that Ozonat's company-supported forum (App. Br. 7; Reply Br. 2) or the integer M (Reply Br. 3) is not "the recited portion" does not squarely address-let alone persuasively rebut-the Examiner's position. We are also unpersuaded by Appellants' argument (App. Br. 8; Reply Br. 4) that Ozonat does not determine a portion using a pattern recognition model. Specifically, Appellants argue that neither of Ozonat's approaches use a pattern recognition model. App. Br. 8; Reply Br. 4. The claim, however, does not recite detecting a portion by its pattern. Rather, the claim only requires that the portion is determined using a model. Nor does the claim specify how the model is used to make this determination. On this record, we see nothing that precludes the recited determination from constituting, among other things, Ozonat's word-based analysis ( Ozonat § 3 .1) described above. Here, the Examiner finds, and we agree, that Ozonat's co-occurrence model corresponds to the recited pattern- recognition model. Ans. 2-3. This co-occurrence model determines co- occurrence values for the M most frequently occurring words. Ozonat § 3 .1. As discussed above, the Examiner finds that Ozonat's M words correspond to the recited portion. Ans. 3. Because the portion's co-occurrence values are determined in Ozonat' s co-occurrence model ( Ozonat § 3 .1 ), the 7 Appeal2014-009536 Application 13/563,030 Examiner did not err in finding that the portion is determined using this model (Final Act. 2-3; Ans. 3). Accordingly, we sustain the Examiner's rejection of claim 1, and claims 2--4, not argued separately with particularity. See App. Br. 1 O; Reply Br. 8. II We also sustain the Examiner's rejection of claims 5 and 11. By reiterating previous arguments about the recited "portion" (see App. Br. 8-9; Reply Br. 5-7), Appellants do not persuasively rebut the Examiner's position for the above-discussed reasons. To the extent that Appellants separately contest the Examiner's findings regarding claims 5 and 11 (see id.), we are unpersuaded of error. Specifically, we are not persuaded by Appellants' argument that Leskovec assigns semantic tags to nodes, not keywords and words as recited in claims 5 and 11. See App. Br. 9. Leskovec assigns tags to keywords extracted from a document. Leskovec § 3 .1 (cited at Ans. 5-6). And the nodes of the graph represent these words and keywords. See, e.g., Leskovec, Figs. 4, 5. So by assigning tags to nodes, we agree with the Examiner's finding that Leskovec also assigns tags to the keywords and words (Ans. 5-6), as recited in claims 5 and 11. Although the Examiner finds that Leskovec' s document is a portion of the retrieved content (Ans. 5), the Examiner nevertheless relies on Ozonat, not Leskovec, as Appellants argue (App. Br. 9), for the recited portion. See Ans. 5---6. Because the Examiner relies on Ozonat for the social network and recited portion (see Final Act. 4--5, 7-8; Ans. 5---6), Appellants' arguments 8 Appeal2014-009536 Application 13/563,030 about Leskovec's deficiencies in this regard (App. Br. 9) are not germane to the Examiner's position. Appellants also refer to a domain of interest, as recited in claim 11. Reply Br. 6-7. The Examiner, however, has made findings regarding the domain of interest-namely, that Leskovec teaches this limitation in Section 3 .1. Final Act. 7. But apart from reiterating the argument about the recited portion, Appellants provide no further reasons for error particular to the recited "domain of interest" and the Examiner's findings. Reply Br. 6-7. Therefore, we are unpersuaded by these arguments (id.) for the reasons stated previously. Accordingly, we sustain the Examiner's rejection of claims 5 and 11, and claims 6-10, and 13 not argued separately with particularity. See App. Br. 10; Reply Br. 8. CLAIM 12 We, however, do not sustain the Examiner's rejection of claim 12, which recites, in part, a module configured to coarsen the semantics graph, and partition the coarsened graph by computing an edge-cut bisection of the coarsened graph. In the proposed combination, the Examiner finds that Ozonat's hierarchical clustering coarsens the semantics graph. Ans. 6; see also Final Act. 8 (citing Ozonat § 5.2.2). Furthermore, the Examiner finds that Ozonat's determination of which of the feature vectors are closest uses the recited edge-cut bisection of the coarsened graph. Ans. 7. We are persuaded by Appellants' argument that Ozonat lacks an edge-cut bisection, as recited. App. Br. 9--10; Reply Br. 7-8. 9 Appeal2014-009536 Application 13/563,030 Even assuming, without deciding, that Ozonat's hierarchical clustering coarsens the semantics graph, the Examiner's finding in this regard is inconsistent with the finding pertaining to the recited edge-cut bisection. See Ans. 6-7. For example, claim 12 requires coarsening by collapsing. Yet Ozonat merges, or collapses, the nodes with closest feature vectors (Ozonat § 5.2.2}-i.e., the step that the Examiner maps to the edge- cut bisection (Ans. 7). That is, the Examiner's mapping seems to suggest that the claim requires coarsening-not partitioning-by computing an edge-cut bisection. See id. at 6-7. Apart from citing Ozonat's Section 5.2.2, the Examiner provides no further analysis to resolve this apparent inconsistency. See Final Act. 8; Ans. 6-7. Nor do we find anything in Ozonat' s hierarchical clustering that partitions by computing an edge-cut bisection. See Ozonat § 5.2.2. Rather, Ozonat only describes merging clusters, not partitioning them. See id. Accordingly, we are persuaded that the Examiner erred in rejecting claim 12. CLAIMS 14 AND 15 We also will not sustain the Examiner's rejection of claim 14, which recites, in pertinent part, a module configured to compute a minimum number of web links to reach from a seed page to each corpus page. In the proposed combination, the Examiner finds that Leskovec' s object-authority weight measures the recited computing minimum number of web links. Ans. 7; see also Final Act. 8-9 (citing Leskovec§ 4.5.2). But Appellants argue, and we agree, that Leskovec lacks this recited computing. App. Br. 10; Reply Br. 8. In particular, Leskovec's object- authority weight measures hub importance. Leskovec§ 4.5.2. Leskovec's 10 Appeal2014-009536 Application 13/563,030 nodes point to other nodes using links. See generally id., Fig. 7 (showing nodes connected by links). Subject-object nodes are "hubs." Id. § 4.5.2. A "good hub" points to nodes with "authoritative" content. Id. The Examiner, however, has not shown that these links are web-links from seed pages. Ans. 7; see also Final Act. 8-9. For example, the Specification describes the recited seed pages are related to websites. See, e.g., Spec. i-f 30. But the Examiner does not point to any websites or web- links in Leskovec. See Ans. 7; see also Final Act. 8-9. Nevertheless, even assuming, without deciding, that the above-discussed links are for web- pages, the Examiner has not shown how Leskovec computes a minimum number of them. See Ans. 7; see also Final Act. 8-9. Rather, Leskovec's authority weight uses the total number of incoming links to one node, rather than a minimum number of links between two nodes. See Leskovec § 4.5.2. Accordingly, we are persuaded that the Examiner erred in rejecting claim 14, and dependent claim 15 for similar reasons. CONCLUSION Under§ 103, the Examiner did not err in rejecting claims 1-11 and 13, but erred in rejecting claims 12, 14, and 15. DECISION The Examiner's decision rejecting claims 1-15 is affirmed-in-part. No time period for taking any subsequent action in connection with this appeal may be extended under 37 C.F.R. § 1.136(a)(l )(iv). AFFIRMED-IN-PART 11 Copy with citationCopy as parenthetical citation