Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
DETAILED ACTION
Claims 14-17, 19-25, 27-33, and 35 are pending. Claims 14, 22, and 30 are independent.
This application was published as US 20230315990.
Apparent priority is 24 July 2020.
The instant application is directed to a method of text evaluation, using attribute features of the text as well as attribute features of related entities, in combination.
Applicant’s amendments and arguments are considered but are either unpersuasive or moot in view of the new grounds of rejection that, if presented, were necessitated by the amendments to the Claims.
This action is Final.
Response to Arguments
35 USC 103
Applicant’s arguments with respect to 35 USC 103 have been considered but are not persuasive. Applicant argues that Ball’s use of social-graph affinity is used to represent the strength of a relationship between objects, whereas the instant application is directed to screening out low-quality texts.
MPEP 2111.01.I. states that “Under a broadest reasonable interpretation (BRI), words of the claim must be given their plain meaning, unless such meaning is inconsistent with the specification.”
Ball discloses: “[0004] Content on a social-network page, such as a post or article, can have numerous user-contributed comments, some of which may be of greater interest to a viewing user than others. In particular embodiments, the user experience for reading and posting comments may be improved by identifying comments that are likely to be of interest to viewing users, and displaying those comments in a prominent location, such as near the top of the comment list. The comments of interest may be, for example, comments that people are reading and replying to, that provide interesting context for and information relevant to the post, are well-written, are written by authors known to write good comments, and so on. Comments of interest may be identified by determining scores for the comments based on signal values, which may quantify features of the comments and/or features of other entities in the social-networking system, such as posts or users. The comments of interest may be presented in an order according to their scores, with the highest-scoring comments being presented at the top of the comment list. The comments may also be filtered, in which case comments that satisfy filtering criteria, e.g., having scores above a threshold value, are presented.” Comments of interest would be considered high-quality comments. Further, Ball includes examples of well-written comments, and comments by authors known to write good comments which are similar to attribute features described in the specification of the instant application.
Ball further discloses: “[0064] In particular embodiments, social-networking system 160 may perform particular actions with respect to a user based on coefficient information. Coefficients may be used to predict whether a user will perform a particular action based on the user's interest in the action. A coefficient may be used when generating or presenting any type of objects to a user, such as advertisements, search results, news stories, media, messages, notifications, or other suitable objects. The coefficient may also be utilized to rank and order such objects, as appropriate. In this way, social-networking system 160 may provide information that is relevant to user's interests and current circumstances, increasing the likelihood that they will find such information of interest…” The coefficient are not only used to predict user actions, but to rank and order objects as described above.
Under the broadest reasonable interpretation, the quality of a text would include whether a comment is of interest to the reader. The system of Ball uses the coefficient in ranking comments which are relevant to a user; Ball also discloses that comments may be filtered based on a threshold. Therefore, the rejection is maintained.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 14, 20-22, and 28-30 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ball et al. (US 20170139921 A1) in view of Veličković et al. (“Graph Attention Networks”) and Zeng et al. ("GraphSAINT: GRAPH SAMPLING BASED INDUCTIVE LEARNING METHOD").
Regarding claim 14, Ball et al. (US 20170139921 A1) discloses: A text detection method, comprising: ("Comments of interest may be identified by determining scores for the comments based on signal values" [0004] – comments are text, and the method detects comments of interest)
determining a first attribute feature of a to-be-detected text ("One of the signals may be based on how many times the comment has been (a) liked, (b) hidden, (c) marked as spam, or (d) replied to within a specified period of time." Abstract. A-d are all attributes of the comment (to-be-detected text).)
and a second attribute feature of elements each having an association relationship with the to-be-detected text; ("Author-related features may be used, for example, to generate an author reputation signal based on a number of times content attributed to the author has been (a) liked, (b) hidden, (c) marked as spam, or (d) replied to within a specified period of time." [0006] A-d are attributes of the author (element having an association relationship with the comment).)
inputting the first attribute feature, the second attribute feature, association relationships between the to-be-detected text and the elements, and association relationships between the elements into a trained network model to obtain a detection result of the to-be-detected text; ("These factors may include, for example, user actions, types of relationships between objects, location information, other suitable factors, or any combination thereof" … "Any type of process or algorithm may be employed for assigning, combining, averaging, and so forth the ratings for each factor and the weights assigned to the factors. In particular embodiments, social-networking system 160 may determine coefficients using machine-learning algorithms trained on historical actions and past user responses, or data farmed from users by exposing them to various options and measuring responses" [0060])
calculating a loss value based on the detection result and a sample labeling result, and back-propagating the loss value to adjust model parameters; and (not explicitly disclosed by Ball)
outputting the detection result of the to-be-detected text to indicate whether the to-be-detected text is a low-quality text; in a case where an output result is 1, indicating that the to-be-detected text is the low-quality text, screening out the low- quality text and preventing the low-quality text from being displayed; in a case where the output result is 0, indicating that the to-be-detected text is not the low-quality text, (“[0004]… Comments of interest may be identified by determining scores for the comments based on signal values, which may quantify features of the comments and/or features of other entities in the social-networking system, such as posts or users. The comments of interest may be presented in an order according to their scores, with the highest-scoring comments being presented at the top of the comment list. The comments may also be filtered, in which case comments that satisfy filtering criteria, e.g., having scores above a threshold value, are presented.”- scores indicate a quality of the text for a user. Meeting a threshold is a Boolean function, and in a digital system this implies an output of 0 or 1.)
wherein the inputting the first attribute feature, the second attribute feature, association relationships between the to-be-detected text and the elements, and association relationships between the elements into a trained network model to obtain a detection result of the to-be-detected text comprises:
aggregating, by combining an attention mechanism, a (K-1)-order feature vector of the node corresponding to the to-be-detected text and (K-1)-order feature vectors of the neighbor nodes of the node corresponding to the to-be-detected text, to obtain a K-order feature vector of the node corresponding to the to-be-detected text; and (not explicitly disclosed by Ball)
predicting, based on the K-order feature vector, the detection result of the to-be-detected text; and “Once the machine-learning model 1204 has been trained, it may be supplied with a comment 1212 for which a ranking is being determined, and predicts one or more labels 1216 that apply to the comment 1212 with confidence 1218. The confidence 1218 may be, e.g., a floating-point value in the range 0 . . . 1, or a number in the range 1 . . . 5, or the like.” [0146])
wherein K is a hyperparameter of the network model, and is determined by pre-training the network model. (not explicitly disclosed by Ball)
Ball does not disclose an attention mechanism, a feature vector, or the order of the feature vector based on a hyperparameter. Ball also does not disclose training by backpropagation.
Veličković discloses: aggregating, by combining an attention mechanism, a (K-1)-order feature vector of the node corresponding to the to-be-detected text and (K-1)-order feature vectors of the neighbor nodes of the node corresponding to the to-be-detected text, to obtain a K-order feature vector of the node corresponding to the to-be-detected text; (Pg. 4, Figure 1 shows the attention mechanism. The feature vectors are aggregated by a weight vector to create the new feature vector (which could be denoted as K order, meaning the feature vectors that are inputs are inherently (K-1) order before the aggregation).
wherein K is a hyperparameter of the network model, and is determined by pre-training the network model. (As shown in Fig. 7 of Applicant’s disclosure, K is directly related to the number of layers in the network. Velickovic discloses that in certain embodiments the Graph Attention Networks consists of 2 or 3 layers (Velickovic Section 3.3). Selecting the number of layers in a neural network according to trial and error (which would include pre-training) is a known method. (see Stathakis (“How many hidden layers and nodes?”) Pg. 1, Section 1, para 1-2.) Therefore, it would have been obvious to one of ordinary skill in the art to pre-train both a 2 layer and 3 layer model as taught by Velickovic and select the better performing architecture. This combination falls under combining prior art elements according to known methods to yield predictable results or use of known technique to improve similar devices (methods, or products) in the same way. See MPEP 2141, KSR, 550 U.S. at 418, 82 USPQ2d at 1396.)
Ball and Veličković are considered analogous art to the claimed invention because they discuss graph networks. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the comment evaluation system of Ball with the teaching of Veličković to use an attention mechanism to aggregate feature vectors. Doing so would have been beneficial in order to “allow for dealing with variable sized inputs,” (Veličković Pg. 2, paragraph 3).
Veličković does not disclose training by backpropagation.
Zeng discloses: calculating a loss value (“λ-normalized loss L”) based on the detection result and a sample labeling result, (“Input: Training graph G (V, E, X); Labels Ȳ ; Sampler SAMPLE;” ; see also “In Algorithm 1, node representation is learned by performing node classification in the supervised setting,” pg. 4 para 1) and back-propagating the loss value to adjust model parameters (“Backward propagation from λ-normalized loss L(yv, ȳv). Update weights.”).
PNG
media_image1.png
275
915
media_image1.png
Greyscale
Zeng, Algorithm 1, Pg. 4
Ball, Velickovic and Zeng are considered analogous art to the claimed invention because they disclose graphs models. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the comment evaluation system of Ball in view of Velickovic with the teaching of Zeng to perform a back-propagation training algorithm. Doing so would have been beneficial because it in order to extract appropriately connected subgraphs so that little information is lost when propagating within the subgraphs, and combine information of many
subgraphs together so that the training process overall learns good representation of the full graph (Zeng Pg. 3, Section 3.1). This combination falls under combining prior art elements according to known methods to yield predictable results or use of known technique to improve similar devices (methods, or products) in the same way. See MPEP 2141, KSR, 550 U.S. at 418, 82 USPQ2d at 1396.
Regarding claim 20, Ball discloses: The method according to claim 14, wherein the element comprises at least one of: an author, a reader, and comment information; the type of the association relationship comprises at least one of: a reading relationship, a releasing relationship, a liking relationship, a commenting relationship, and a forwarding relationship. ("Author-related features may be used" [0006] Author is a releasing relationship.)
Regarding claim 21, Ball discloses: The method according to claim 14, wherein the first attribute feature comprises at least one of: a text feature, a picture feature, a soundtrack feature, a number-of-likes feature, a number-of-forwarding feature, a number-of-comments feature, a comment information feature, a number-of-views feature, and an online time feature; ("One of the signals may be based on how many times the comment has been (a) liked, (b) hidden, (c) marked as spam, or (d) replied to within a specified period of time." Abstract. A-d are all attributes of the comment (to-be-detected text).) the second attribute feature comprises at least one of: a reader portrait, an author portrait and a release time feature. ("As an example and not by way of limitation, a user may provide his or her name, profile picture, contact information, birth date, sex, marital status, family status, employment, education background, preferences, interests, or other demographic information. In particular embodiments, a user node 202 may be associated with one or more data objects corresponding to information associated with a user." [0041] a user is an author or a reader, or both.)
Claim 22 is an apparatus claim with limitations corresponding to the limitations of Claim 14 and is rejected under similar rationale. Additionally, Ball teaches “one or more processors” and “a memory” (Fig. 14 shows a computer system with “PROCESSOR” and “STORAGE”)
Claim 28 is an apparatus claim with limitations corresponding to the limitations of Claim 20 and is rejected under similar rationale.
Claim 29 is an apparatus claim with limitations corresponding to the limitations of Claim 21 and is rejected under similar rationale.
Claim 30 is a computer-readable medium product claim with limitations corresponding to the limitations of Claim 14 and is rejected under similar rationale. Additionally, Ball teaches “A non-transitory storage medium, comprising computer executable instructions” (Fig. 14 shows a computer system with “PROCESSOR,” “MEMORY,” and “STORAGE”)
Claim(s) 15-16, 23-24, 31-32, and 35 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ball in view Velickovic and Zeng, in further view of Johansson et al. ("Sharding the LDBC Social Network").
Regarding claim 15, Ball discloses: The method according to claim 14, wherein before the inputting the first attribute feature, the second attribute feature, the association relationships between the to-be-detected text and the elements, and the association relationship between the elements into the trained network model, the method further comprises: determining the to-be-detected text and the elements as nodes respectively; ("In particular embodiments, the one or more signals may be based on existence of an edge between a user node representing the audience member and a user node representing an author of the comment in a social graph on the social-networking system." [0105]. Ball discloses that the elements are represented as nodes. Ball does not explicitly disclose that the comment is represented as a node.)
generating, according to types of the association relationships between the to-be-detected text and the elements, connection edges between the node corresponding to the to-be-detected text and nodes corresponding to the elements; generating, according to types of the association relationships between the elements, connection edges between the nodes corresponding to the elements; and determining, according to a structure diagram composed of the nodes and the connection edges, the association relationships between the to-be-detected text and the elements, and the association relationships between the elements. ("In particular embodiments, a pair of nodes in social graph 200 may be connected to each other by one or more edges 206. An edge 206 connecting a pair of nodes may represent a relationship between the pair of nodes." [0045] Fig. 2 further illustrates the edges, as well as association relationships, between nodes. [0105] provides further details on the association relationships.)
Ball teaches the comments appearing as “edges 206” such as the “like” comments in Figure 2. Thus, the comments are not shown as nodes.
Velickovic does not teach comments as nodes. Neither does Zeng.
Johansson discloses: determining the to-be-detected text and the elements as nodes respectively; (":Comment" Pg. 2, 1st Figure; The figure shows a comment in a social network represented as a node in a graph.)
Ball, Velickovic, Zeng, and Johansson are considered analogous art to the claimed invention because they disclose graphs models. Ball discloses in [0042] that a concept node can be any resource or concept within the social network, which would include a comment. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the comment evaluation system of Ball in view of Velickovic and Zeng with the teaching of Johansson to represent comments as nodes. Doing so would have been beneficial so that interactions between the author and audience of the comment (Ball [0104]) are represented in the graph.
Regarding claim 16, Ball in view of Velickovic, Zeng, and Johansson discloses: The method according to claim 15, wherein the determining, according to the structure diagram composed of the nodes and the connection edges, the association relationships between the to-be-detected text and the elements and the association relationships between the elements comprises: (disclosed in claim 15) performing a sampling operation on neighbor nodes of the node corresponding to the to-be- detected text, wherein the neighbor nodes are nodes each having a connection edge with the node corresponding to the to-be-detected text; and determining a structure diagram composed of the node corresponding to the to-be-detected text, neighbor nodes obtained through sampling, a node associated with the neighbor nodes obtained through sampling and the connection edges, as the association relationships between the to-be-detected text and the elements and the association relationships between the elements. (not explicitly disclosed by Ball)
Ball and Velickovic do not teach the specific features of Claim 16.
Zeng et al. discloses: performing a sampling operation on neighbor nodes of the node corresponding to the to-be- detected text, wherein the neighbor nodes are nodes each having a connection edge with the node corresponding to the to-be-detected text; and determining a structure diagram composed of the node corresponding to the to-be-detected text, neighbor nodes obtained through sampling, a node associated with the neighbor nodes obtained through sampling and the connection edges, as the association relationships between the to-be-detected text and the elements and the association relationships between the elements. ("we perform variance reduction analysis, and design light-weight sampling algorithms by quantifying “influence” of neighbors." Pg. 2; Fig. 1 shows an example of sampled nodes and relationships; Section 1. Sections 3.3 and 3.4 further discuss the methods of sampling.)
Ball/Velickovic/Johansson and Zeng are considered analogous art to the claimed invention because they disclose graphs models. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have further modified the comment evaluation system of Ball in view of Velickovic, Zeng and Johansson with the teaching of Zeng to perform a sampling operation on neighboring nodes. Doing so would have been beneficial because it "improves training efficiency and accuracy" (Zeng Pg. 1, Abstract).
Claim 23 is an apparatus claim with limitations corresponding to the limitations of Claim 15 and is rejected under similar rationale.
Claim 24 is an apparatus claim with limitations corresponding to the limitations of Claim 16 and is rejected under similar rationale.
Claim 31 is a computer-readable medium product claim with limitations corresponding to the limitations of Claim 15 and is rejected under similar rationale.
Claim 32 is a computer-readable medium product claim with limitations corresponding to the limitations of Claim 16 and is rejected under similar rationale.
Regarding claim 35, Ball and Velickovic do not disclose the additional limitations.
Zeng discloses: 35. The method according to claim 16, wherein the performing the sampling operation on neighbor nodes of the node corresponding to the to-be-detected text comprises: adopting a set sampling rule to sample the neighbor nodes of the node corresponding to the to-be-detected text. (“We define the optimal edge sampler to minimize variance for every dimension of s. We restrict ourselves to independent edge sampling. For each e E, we make independent decision on whether it should be in Gs or not. The probability of including e is pe. We further constrain P pe = m, so that the expected number of sampled edges equals to m. The budget m is a given sampling parameter.”)
See claim 16 for motivation statement.
Claim(s) 17, 19, 25, 27, and 33 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ball in view of Velickovic, Zeng, and Johansson, in further view of Lao et al. (“Multimodal Local Perception Bilinear Pooling for Visual Question Answering”).
Regarding claim 17, Ball in view of Velickovic, Zeng, and Johansson discloses: The method according to claim 15, wherein the determining the first attribute feature of the to-be-detected text (disclosed in claim 14) comprises: adopting different conversion algorithms for attribute information of different categories of the to-be-detected text, to obtain expression vectors of the attribute information of different categories; obtaining a zero-order feature vector of the node corresponding to the to-be-detected text, through a pooling operation on the expression vectors of the attribute information of different categories; and determining the zero-order feature vector as the first attribute feature. (not explicitly disclosed)
Ball, Velickovic, Zeng, and Johansson do not teach the specific features of Claim 17.
Lao discloses: adopting different conversion algorithms for attribute information of different categories of the to-be-detected text, to obtain expression vectors of the attribute information of different categories; (Pg. 4, Figure 2 shows different conversion algorithms for an image vs. a text question.) obtaining a zero-order feature vector of the node corresponding to the to-be-detected text, through a pooling operation on the expression vectors of the attribute information of different categories; and determining the zero-order feature vector as the first attribute feature. (Figure 2 shows “Multimodal Local Bilinear Pooling” to combine the feature vectors into one feature vector (feature fusion). It is then directly input into the Answer Prediction model, so it can be considered a zero-order feature vector before it is input to the model.
Ball/Velickovic/Zeng/Johansson and Lao are considered analogous art to the claimed invention because they discuss use of machine learning models. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the comment evaluation system of Ball in view of Velickovic, Zeng, and Johansson with the teaching of Lao to use multiple conversion algorithms and combine the feature vectors through a pooling operation. Doing so would have been beneficial in order to "comprehensively employ the information from both modals" (Lao Pg. 1, Abstract).
Regarding claim 19, Ball in view of Velickovic, Zeng, Johansson and Lao discloses: The method according to claim 17. Ball further discloses: wherein the attribute information of different categories of the to-be-detected text comprises at least one of: numerical-type attribute information, text-type attribute information, image-type attribute information and audio-type attribute information. (“One of the signals may be based on how many times the comment has been (a) liked, (b) hidden, (c) marked as spam, or (d) replied to within a specified period of time.” Abstract. This is numerical-type attribute as defined in [0062] of the instant application.)
Claim 25 is an apparatus claim with limitations corresponding to the limitations of Claim 17 and is rejected under similar rationale.
Claim 27 is an apparatus claim with limitations corresponding to the limitations of Claim 19 and is rejected under similar rationale.
Claim 33 is a computer-readable medium product claim with limitations corresponding to the limitations of Claim 17 and is rejected under similar rationale.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JON C MEIS whose telephone number is (703)756-1566. The examiner can normally be reached Monday - Thursday, 8:30 am - 5:30 pm EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Hai Phan can be reached on 571-272-6338. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/JON CHRISTOPHER MEIS/Examiner, Art Unit 2654
/HAI PHAN/Supervisory Patent Examiner, Art Unit 2654