DETAILED ACTION
Receipt of Applicant’s Amendment, filed May 20, 2026 is acknowledged.
Claims 1-4, 7-8 were amended.
Claims 5-6, 9-20 were canceled.
Claims 21-34 were newly added
Claims 1-4, 7-8, 21-34 are pending in this office action.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 3, 4, 8, 26, 27, and 29 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
With regard to claims 3, and 26 the claim recites “labeling each respective pair of training passages with a label.” This claim language lacks antecedent basis. The parent claims recite “wherein each respective pair of training passages has a corresponding label indicating that a first training passage is relatively more convincing than a second training passage of the respective pair”.
The parent claim recites language indicating that labeling has already occurred and that the pair of training passages already have a corresponding label. The distinction between this corresponding label and the “label” that is applied during the labeling process of the pair of training passages is unclear. Furthermore, it is unclear how many labeling processes are being claimed. Is the labeling process recited in claim 3, the process that resulted in the pair of training passages having the corresponding label? Does the claim require the previously labeled training passages to be relabeled? For examination purposes the labeling process recited in claim 3 has been construed as being the process that results in the pair of training passages having the corresponding label, and the claimed ‘a label’ has been construed to mean --the corresponding label--.
With regard to claims 4, 8, 27, and 29, claim 4 recites “training the convincingness ranking neural network model on remaining training passages in the directed graph after eliminating the loop. ” Claim 8 recites “wherein the convincingness ranking neural network model is based on a feed forward neural network that has previously been trained with back propagation.” This claim language lacks antecedent basis. The parent claim recites “a convincingness ranking neural network model that is trained on one or more pairs of training passages”.
It is unclear if the training recited in these claims are a distinct training process recited in the parent claim, or intended to further define the training process recited in the parent train. For examination purposes these claim limitation have been construed as referring to the previous training operation, for example claim 4 has been construed to mean –wherein the training of the convincingness ranking neural network model is performed based on the remaining training passages in the directed graph after eliminating the loop--.
With regard to claims 21, 22, 30, 31, and 34 these claims recite the term “that” in a manner in which it is unclear what is being referenced.
Claim 21 recites “the convincingness ranking neural network having separate fully-connected layers that generate the separate convincingness scores”. Within this language, the term “that” may reasonable be interpreted as referring to the convincingness ranking neural network or to the separate fully-connected layers. For examination purposes the term ‘that’ has been construed as referring to --the seperate fully-connected layers--.
Claim 22 recites “the convincingness ranking neural network having a softmax layer that generates relative convincingness probabilities for the two or more passages”. Within this language, the term “that” may reasonable be interpreted as referring to the convincingness ranking neural network or to the softmax layer. For examination purposes the term ‘that’ has been construed as referring to --softmax layer--.
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.
Claims 1-8, 24-29, and 33 are rejected under 35 U.S.C. 103 as being unpatentable over Habernal [Which argument is more convincing? Analysis and predicting convincingness of Web arguments using bidirectional LSTM] in view of Comar [10776847].
With regard to claim 1 Habernal teaches A computer-implemented method, comprising:
[a] receiving two or more input passages as the text that make up Argument 1 and Argument 2 (Habernal, Page 1591, Section 3.1 “First, we must be sure that the obtained texts are actual arguments. Second, the context of the argument should be known (the prompt and the stance).”; Figure 1, see the text which make up the actual arguments 1 and 2);
[b] inputting the two or more input passages as the text that make up Argument 1 and Argument 2 (Habernal, Page 1591, Section 3.1 “First, we must be sure that the obtained texts are actual arguments. Second, the context of the argument should be known (the prompt and the stance).”; Figure 1, see the text which make up the actual arguments 1 and 2) to a convincingness ranking neural network model as the BLSTM (Habernal, Page 1590; Section 1 “We propose a novel task of predicting convincingness of arguments in an argument pair, as well as ranking arguments related to a certain topic. Since no data for such a task are available, we create a new annotated corpus. We employ SVM model with rich linguistic features as well as bidirectional Long Short-Term Memory (BLSTM) neural networks because of their excellent performance across various end-to-end NLP tasks (Goodfellow et al., 2016; Piech et al., 2015; Wen et al., 2016; Dyer et al., 2015; Rocktaschel et al., 2016).”) that is trained (Habernal, Page 1594 Section 3.4.3 “We also release the full dataset UKPConvArgAll. In this data, no global filtering using graph construction methods is applied, only the local pre-filtering using MACE.We believe this dataset can be used as a supporting training data for some tasks that do not rely on the property of total ordering.”) on one or more pairs of training passages as the text that make up Argument 1 and Argument 2 (Habernal, Page 1591, Section 3.1 “First, we must be sure that the obtained texts are actual arguments. Second, the context of the argument should be known (the prompt and the stance).”; Figure 1, see the text which make up the actual arguments 1 and 2), wherein each respective pair of training passages has a corresponding label as the annotations of the pairs of arguments (Habernal, Page 1590 Section 2, “Our newly created corpus of annotated pairs of arguments might resemble recent large-scale corpora for textual inference”) indicating (Id; Please note this claim limitation has been identified as an intended use of the pre-labels, as well as non-functional descriptive material. The meaning that a human may derive from the pre-label does not invoke a functionality relationship on the claimed device. Within the device this label appears to have been human provided intelligence (See Paragraph [0035])) that a first training passage as A1 (Habemal Page 1589 Section 1: “one can decide that Al is probably more convincing than A2”) is relatively more convincing as probably more convincing (Id) than a second training passage as A2 (Id) of the respective pair as the A1 and A2 pair (Id);
[c] receiving, from the convincingness ranking neural network model as the UKPConvArgRank is implemented with BLSTM (Habemal, Page 1596 Section 4.2 Ranking arguments: “We use the UKPConvArgRank data,… Regarding the BLSTM, we only replace the output layer with a linear activation function and optimize mean absolute error loss”), separate convincingness scores as the assigned real value score to each argument, e.g. the weights (Page 1596 Section 4.2 “We address this problem as a regression task. We use the UKPConvArgRank data, in which a real value score is assigned to each argument so the arguments can be ranked by their convincingness (for each topic independently).”; Page 1593 Section 3.4.2 “Then the weight w of edge ei is computed as follows: [Equation 1]) for each of the two or more input passages as each argument is assigned a real value score (Id);
[d] ranking the two or more input passages (Page 1596 Section 4.2 “We address this problem as a regression task. We use the UKPConvArgRank data, in which a real value score is assigned to each argument so the arguments can be ranked by their convincingness (for each topic independently).”) based at least on the separate convincingness scores as their assigned scores, e.g. the computed weights (Id; Page 1593 Section 3.4.2 “Then the weight w of edge ei is computed as follows: [Equation 1]);
[e] receiving a query as obtaining text containing a prompt (Habernal, Page 1591, Section 3.1 “Sampling large sets of arguments for annotation from the Web poses several challenges. First, we must be sure that the obtained texts are actual arguments. Second, the context of the argument should be known (the prompt and the stance).”) via a [[as sampling arguments from the Web (Id);
[f] based at least on the ranking (Page 1596 Section 4.2 “We address this problem as a regression task. We use the UKPConvArgRank data, in which a real value score is assigned to each argument so the arguments can be ranked by their convincingness (for each topic independently).”), choosing as predicting (Habernal, Page 1595 Section 4.1 Predicting convincingness of pairs: “Their output is then concatenated into a single drop-out layer and passed to the final sigmoid layer for binary predictions”), from the two or more input passages as the ranked arguments (Id), a [[as the highest ranked argument (Id) that is responsive to the query as the prompt (Id); and
[g] outputting the [[ as HIT workers were presented with an argument pair, the prompt, and the stance as in Figure 1 (Habernal, Page 1591 Section 3.2 Crowdsourcing annotations: “In the HIT, workers were presented with an argument pair, the prompt, and the stance as in Figure 1.”).
Habernal does not explicitly teach [e] receiving a query via a graphical user interface; [f] … choosing… a selected passage that is responsive to the query; and [g] outputting the selected passage on the graphical user interface.
Comar teaches [e] receiving a query as receiving a request 152 (Comar, Column 14, lines 9-11 “Client devices used to 10 interact with various embodiments can include any appropriate device operable to send and receive requests”; as the selected search results (Comar, Column 3, lines 61-63 “Similarly, the example display 150 of FIG. 1B illustrates a set of search results 154 presented for a different submitted query 152, here the query "women's apparel."”; Figure 1B 152 See query “women’s apparel”) via a graphical user interface (Figure 1B, see display);
[f] based at least on the ranking as ranking based on intent (Column 10, lines 6-9 “the intent logic 320 can attempt to determine the intent behind the request, which can then be used to adjust the ranking and/or selection of content to be provided based at last in part upon the determined intent”), choosing (Column 13, lines 19-22 “This can include, as discussed elsewhere herein, generating an ordered ranking and then selecting at least a subset of highest-ranked content items to be selected for display”), from the two input passages as highest-ranked content items (Id), a selected passage as selecting at least a subset (Id) that is responsive as for display (Id) to the query as based upon the specific intent of the query (Column 13, lines 2-4 “If bias information is available, such as whether a current user or query is associated with a specific intent”); and
[g] outputting the selected passage as the selected search results 154 (Comar, Column 3, lines 61-63 “Similarly, the example display 150 of FIG. 1B illustrates a set of search results 154 presented for a different submitted query 152, here the query "women's apparel."”; Figure 1B, 154) on the graphical user interface (Figure 1B, see display).
It would have been obvious to one of ordinary skill to which said subject matter pertains at the time the invention was filed to have implemented the selecting and ranking model taught by Comar, to select and present search results including ranked arguments as taught by Habernal as it yields the predictable results of displaying content to the user (Comar Column 12, lines 60-67) which may be considered more convincing to the human (Habernal, Page 1594 Section 3.4.2 Transitivity evaluation using argument graphs “We interpret this finding as that it is easier for humans to judge A> B than A=B consistently across all possible pairs of arguments from a given topic.”). One of ordinary skill in the art would recognize that Habernal presents a means of generating a Machine Learning model that is capable of ranking the convincingness of arguments. One of ordinary skill in the art would recognize that this may be implemented within a search recommendation system such as the one taught by Comar, to enable the system to determine which search results the user would find the most convincing. This may be used to determine what would be able to convince or persuade a user to make a particular purchase or prefer a specific item recommendation.
With regard to claims 2 and 25 the proposed combination further teaches wherein the two or more input passages as the Argument pair Ai and A2 (Page 1591 “Argument pair is an ordered set of two arguments (Al and A2) belonging to the same topic; see Figure 1”) share a particular stance as a stance (Page 1589 Figure 1; Page 1591 “We will use the following terminology. We use topic to refer to a subset of an on-line debate with a given prompt and a certain stance (for example, "Should physical education be mandatory in schools? - yes" is considered as a single topic).”) with respect to a topic as topic (ID) of the query as a given prompt (Id).
With regard to claims 3 and 26 the proposed combination further teaches the computer-implemented method further comprising:
labeling as adding the annotations to the pairs of arguments, e.g. the real value scores assigned to each argument so the arguments can be ranked by their convincingness (Habernal, Page 1590 Section 2, “Our newly created corpus of annotated pairs of arguments might resemble recent large-scale corpora for textual inference”) each respective pair of training passages as the pairs of arguments (Id) with a label as the annotation (Id);
generating a directed graph representing the training passages(Habernal, Page 1593 Algorithm 1: “Building DAG from sorted argument pairs.”);
detecting a cyclic relationship when the directed graph forms a loop (Habernal, Page 1593, Section 3.4.2 “We use Johnson's algorithm for finding all elementary cycles in DAG (Johnson, 1975).”; “By building argument graph from all pairs, introducing cycles into the graph seems to be inevitable”; In Algorithm, 1 “if hasCycles(g) then”); and
eliminating the loop by removing, from the directed graph, a particular node representing a particular training passage (Habernal, Page 1593 Algorithm 1 “report about breaking DAG”; Page 1594 Section 3.4. 3 “We discard the equal argument pairs in advance and filter out argument pairs that break the DAG properties.”).
With regard to claims 4 and 27 the proposed combination further teaches
training (Habernal, Page 1594 Section 3.4.3 “We also release the full dataset UKPConvArgAll. In this data, no global filtering using graph construction methods is applied, only the local pre-filtering using MACE. We believe this dataset can be used as a supporting training data for some tasks that do not rely on the property of total ordering.”) the convincingness ranking neural network model as the BLSTM (Habernal, Page 1590; Section 1 “We propose a novel task of predicting convincingness of arguments in an argument pair, as well as ranking arguments related to a certain topic. Since no data for such a task are available, we create a new annotated corpus. We employ SVM model with rich linguistic features as well as bidirectional Long Short-Term Memory (BLSTM) neural networks because of their excellent performance across various end-to-end NLP tasks (Goodfellow et al., 2016; Piech et al., 2015; Wen et al., 2016; Dyer et al., 2015; Rocktaschel et al., 2016).”) on remaining training passages (Habernal, Page 1593 “Here we assume that the graph is a DAG”) in the directed graph after eliminating the loop (Habernal, Page 1593 Algorithm 1 “report about breaking DAG”; Page 1594 Section 3.4. 3 “We discard the equal argument pairs in advance and filter out argument pairs that break the DAG properties.”).
With regard to claims 7 and 28, the proposed combination further teaches receiving, via the graphical user interface, user input indicating feedback as workers applying annotations (Harnal, Page 1591 Section 3.2 “All 16,927 argument pairs were annotated by five workers each (85k assignments in total).”) with respect to convincingness of the selected passage as the workers chose a particular stance for the prompt as being the most convincing (Habernal, Page 1591 “In the HIT, workers were presented with an argument pair, the prompt, and the stance as in Figure 1. They had to choose either "Al is more convincing than A2" (Al> A2), "Al is less convincing than A2" (Al <A2), or "Al and A2 are convincing equally" (Al=A2).”);
updating the convincingness ranking neural network model (Habernal, Page 1595 “the embedding weights are further updated during training.”) based at least on the feedback as the weight is computed based on the votes of the workers (Habernal, Page 1593 “Let v be a single worker's vote and cv a global worker's competence score. Then the weight w of edge ei is computed as follows: [See Equation 1 based on cv]).”; See Equation 1).
With regard to claims 8 and 29, the proposed combination further teaches wherein the convincingness ranking neural network model as the BLSTM (Habernal, Page 1590; Section 1 “We propose a novel task of predicting convincingness of arguments in an argument pair, as well as ranking arguments related to a certain topic. Since no data for such a task are available, we create a new annotated corpus. We employ SVM model with rich linguistic features as well as bidirectional Long Short-Term Memory (BLSTM) neural networks because of their excellent performance across various end-to-end NLP tasks (Goodfellow et al., 2016; Piech et al., 2015; Wen et al., 2016; Dyer et al., 2015; Rocktaschel et al., 2016).”) is based on a feed forward neural network as BLSTM is inherently a neural network that has both feed forwarding and back propagation and SVM model is inherently a feed forward ML model (Id; Page 1595 “The core of the model consists of two bi-directional LSTM networks with 64 output neurons each.”) that has previously been trained (Habernal, Page 1595 Section 4.1 “The input layer relies on pre-trained word embeddings, in particular Glo Ve (Pennington et al., 2014) trained on 840B tokens from Common Crawl”) with back propagation as BiLSTM has Back propagation (Habernal, Page 1590; Section 1).
With regard to claim 24 Habernal teaches A system, comprising:
[[
[a] receive two or more input passages as the text that make up Argument 1 and Argument 2 (Habernal, Page 1591, Section 3.1 “First, we must be sure that the obtained texts are actual arguments. Second, the context of the argument should be known (the prompt and the stance).”; Figure 1, see the text which make up the actual arguments 1 and 2);
[b] input the two or more input passages as the text that make up Argument 1 and Argument 2 (Habernal, Page 1591, Section 3.1 “First, we must be sure that the obtained texts are actual arguments. Second, the context of the argument should be known (the prompt and the stance).”; Figure 1, see the text which make up the actual arguments 1 and 2) to a convincingness ranking neural network model as the BLSTM (Habernal, Page 1590; Section 1 “We propose a novel task of predicting convincingness of arguments in an argument pair, as well as ranking arguments related to a certain topic. Since no data for such a task are available, we create a new annotated corpus. We employ SVM model with rich linguistic features as well as bidirectional Long Short-Term Memory (BLSTM) neural networks because of their excellent performance across various end-to-end NLP tasks (Goodfellow et al., 2016; Piech et al., 2015; Wen et al., 2016; Dyer et al., 2015; Rocktaschel et al., 2016).”) that is trained (Habernal, Page 1594 Section 3.4.3 “We also release the full dataset UKPConvArgAll. In this data, no global filtering using graph construction methods is applied, only the local pre-filtering using MACE.We believe this dataset can be used as a supporting training data for some tasks that do not rely on the property of total ordering.”) on one or more pairs of training passages as the text that make up Argument 1 and Argument 2 (Habernal, Page 1591, Section 3.1 “First, we must be sure that the obtained texts are actual arguments. Second, the context of the argument should be known (the prompt and the stance).”; Figure 1, see the text which make up the actual arguments 1 and 2), wherein each respective pair of training passages has a corresponding label as the annotations of the pairs of arguments (Habernal, Page 1590 Section 2, “Our newly created corpus of annotated pairs of arguments might resemble recent large-scale corpora for textual inference”) indicating (Id; Please note this claim limitation has been identified as an intended use of the pre-labels, as well as non-functional descriptive material. The meaning that a human may derive from the pre-label does not invoke a functionality relationship on the claimed device. Within the device this label appears to have been human provided intelligence (See Paragraph [0035])) that a first training passage as A1 (Habemal Page 1589 Section 1: “one can decide that Al is probably more convincing than A2”) is relatively more convincing as probably more convincing (Id) than a second training passage as A2 (Id) of the respective pair as the A1 and A2 pair (Id);
[c] receive, from the convincingness ranking neural network model as the UKPConvArgRank is implemented with BLSTM (Habemal, Page 1596 Section 4.2 Ranking arguments: “We use the UKPConvArgRank data,… Regarding the BLSTM, we only replace the output layer with a linear activation function and optimize mean absolute error loss”), separate convincingness scores as the assigned real value score to each argument, e.g. the weights (Page 1596 Section 4.2 “We address this problem as a regression task. We use the UKPConvArgRank data, in which a real value score is assigned to each argument so the arguments can be ranked by their convincingness (for each topic independently).”; Page 1593 Section 3.4.2 “Then the weight w of edge ei is computed as follows: [Equation 1]) for each of the two or more input passages as each argument is assigned a real value score (Id);
[d] rank the two or more input passages (Page 1596 Section 4.2 “We address this problem as a regression task. We use the UKPConvArgRank data, in which a real value score is assigned to each argument so the arguments can be ranked by their convincingness (for each topic independently).”) based at least on the separate convincingness scores as their assigned scores, e.g. the computed weights (Id; Page 1593 Section 3.4.2 “Then the weight w of edge ei is computed as follows: [Equation 1]);
[e] receive a query as obtaining text containing a prompt (Habernal, Page 1591, Section 3.1 “Sampling large sets of arguments for annotation from the Web poses several challenges. First, we must be sure that the obtained texts are actual arguments. Second, the context of the argument should be known (the prompt and the stance).”) via a [[as sampling arguments from the Web (Id);
[f] based at least on the ranking (Page 1596 Section 4.2 “We address this problem as a regression task. We use the UKPConvArgRank data, in which a real value score is assigned to each argument so the arguments can be ranked by their convincingness (for each topic independently).”), choose as predicting (Habernal, Page 1595 Section 4.1 Predicting convincingness of pairs: “Their output is then concatenated into a single drop-out layer and passed to the final sigmoid layer for binary predictions”), from the two or more input passages as the ranked arguments (Id), a [[as the highest ranked argument (Id) that is responsive to the query as the prompt (Id); and
[g] display the [[ as HIT workers were presented with an argument pair, the prompt, and the stance as in Figure 1 (Habernal, Page 1591 Section 3.2 Crowdsourcing annotations: “In the HIT, workers were presented with an argument pair, the prompt, and the stance as in Figure 1.”).
Habernal does not explicitly teach A system, comprising: a processor; and a memory storing computer executable instructions, which, when executed, cause the processor to: …[e] receive a query via a graphical user interface; [f] … choose… a selected passage that is responsive to the query; and [g] display the selected passage on the graphical user interface.
Comar teaches A system, comprising: a processor; and
a memory storing computer executable instructions, which, when executed, cause the processor (Comar, Claim 11 “A system, comprising: at least one processor; and memory including instructions that, when executed by the at least one processor, cause the system to:”) to:
[e] receive a query as receiving a request 152 (Comar, Column 14, lines 9-11 “Client devices used to 10 interact with various embodiments can include any appropriate device operable to send and receive requests”; as the selected search results (Comar, Column 3, lines 61-63 “Similarly, the example display 150 of FIG. 1B illustrates a set of search results 154 presented for a different submitted query 152, here the query "women's apparel."”; Figure 1B 152 See query “women’s apparel”) via a graphical user interface (Figure 1B, see display);
[f] based at least on the ranking as ranking based on intent (Comar, Column 10, lines 6-9 “the intent logic 320 can attempt to determine the intent behind the request, which can then be used to adjust the ranking and/or selection of content to be provided based at last in part upon the determined intent”), choose (Column 13, lines 19-22 “This can include, as discussed elsewhere herein, generating an ordered ranking and then selecting at least a subset of highest-ranked content items to be selected for display”), from the two input passages as highest-ranked content items (Id), a selected passage as selecting at least a subset (Id) that is responsive as for display (Id) to the query as based upon the specific intent of the query (Column 13, lines 2-4 “If bias information is available, such as whether a current user or query is associated with a specific intent”); and
[g] display the selected passage as the selected search results 154 (Comar, Column 3, lines 61-63 “Similarly, the example display 150 of FIG. 1B illustrates a set of search results 154 presented for a different submitted query 152, here the query "women's apparel."”; Figure 1B, 154) on the graphical user interface (Figure 1B, see display).
It would have been obvious to one of ordinary skill to which said subject matter pertains at the time the invention was filed to have implemented the selecting and ranking model taught by Comar, to select and present search results including ranked arguments as taught by Habernal as it yields the predictable results of displaying content to the user (Comar Column 12, lines 60-67) which may be considered more convincing to the human (Habernal, Page 1594 Section 3.4.2 Transitivity evaluation using argument graphs “We interpret this finding as that it is easier for humans to judge A> B than A=B consistently across all possible pairs of arguments from a given topic.”). One of ordinary skill in the art would recognize that Habernal presents a means of generating a Machine Learning model that is capable of ranking the convincingness of arguments. One of ordinary skill in the art would recognize that this may be implemented within a search recommendation system such as the one taught by Comar, to enable the system to determine which search results the user would find the most convincing. This may be used to determine what would be able to convince or persuade a user to make a particular purchase or prefer a specific item recommendation.
With regard to claim 33, Habernal does not explicitly teach A [[
[a] receiving two or more input passages as the text that make up Argument 1 and Argument 2 (Habernal, Page 1591, Section 3.1 “First, we must be sure that the obtained texts are actual arguments. Second, the context of the argument should be known (the prompt and the stance).”; Figure 1, see the text which make up the actual arguments 1 and 2);
[b] inputting the two or more input passages as the text that make up Argument 1 and Argument 2 (Habernal, Page 1591, Section 3.1 “First, we must be sure that the obtained texts are actual arguments. Second, the context of the argument should be known (the prompt and the stance).”; Figure 1, see the text which make up the actual arguments 1 and 2) to a convincingness ranking neural network model as the BLSTM (Habernal, Page 1590; Section 1 “We propose a novel task of predicting convincingness of arguments in an argument pair, as well as ranking arguments related to a certain topic. Since no data for such a task are available, we create a new annotated corpus. We employ SVM model with rich linguistic features as well as bidirectional Long Short-Term Memory (BLSTM) neural networks because of their excellent performance across various end-to-end NLP tasks (Goodfellow et al., 2016; Piech et al., 2015; Wen et al., 2016; Dyer et al., 2015; Rocktaschel et al., 2016).”) that is trained (Habernal, Page 1594 Section 3.4.3 “We also release the full dataset UKPConvArgAll. In this data, no global filtering using graph construction methods is applied, only the local pre-filtering using MACE.We believe this dataset can be used as a supporting training data for some tasks that do not rely on the property of total ordering.”) on one or more pairs of training passages as the text that make up Argument 1 and Argument 2 (Habernal, Page 1591, Section 3.1 “First, we must be sure that the obtained texts are actual arguments. Second, the context of the argument should be known (the prompt and the stance).”; Figure 1, see the text which make up the actual arguments 1 and 2), wherein each respective pair of training passages has a corresponding label as the annotations of the pairs of arguments (Habernal, Page 1590 Section 2, “Our newly created corpus of annotated pairs of arguments might resemble recent large-scale corpora for textual inference”) indicating (Id; Please note this claim limitation has been identified as an intended use of the pre-labels, as well as non-functional descriptive material. The meaning that a human may derive from the pre-label does not invoke a functionality relationship on the claimed device. Within the device this label appears to have been human provided intelligence (See Paragraph [0035])) that a first training passage as A1 (Habemal Page 1589 Section 1: “one can decide that Al is probably more convincing than A2”) is relatively more convincing as probably more convincing (Id) than a second training passage as A2 (Id) of the respective pair as the A1 and A2 pair (Id);
[c] receiving, from the convincingness ranking neural network model as the UKPConvArgRank is implemented with BLSTM (Habemal, Page 1596 Section 4.2 Ranking arguments: “We use the UKPConvArgRank data,… Regarding the BLSTM, we only replace the output layer with a linear activation function and optimize mean absolute error loss”), separate convincingness scores as the assigned real value score to each argument, e.g. the weights (Page 1596 Section 4.2 “We address this problem as a regression task. We use the UKPConvArgRank data, in which a real value score is assigned to each argument so the arguments can be ranked by their convincingness (for each topic independently).”; Page 1593 Section 3.4.2 “Then the weight w of edge ei is computed as follows: [Equation 1]) for each of the two or more input passages as each argument is assigned a real value score (Id);
[d] ranking the two or more input passages (Page 1596 Section 4.2 “We address this problem as a regression task. We use the UKPConvArgRank data, in which a real value score is assigned to each argument so the arguments can be ranked by their convincingness (for each topic independently).”) based at least on the separate convincingness scores as their assigned scores, e.g. the computed weights (Id; Page 1593 Section 3.4.2 “Then the weight w of edge ei is computed as follows: [Equation 1]);
[e] receiving a query as obtaining text containing a prompt (Habernal, Page 1591, Section 3.1 “Sampling large sets of arguments for annotation from the Web poses several challenges. First, we must be sure that the obtained texts are actual arguments. Second, the context of the argument should be known (the prompt and the stance).”) via a [[as sampling arguments from the Web (Id);
[f] based at least on the ranking (Page 1596 Section 4.2 “We address this problem as a regression task. We use the UKPConvArgRank data, in which a real value score is assigned to each argument so the arguments can be ranked by their convincingness (for each topic independently).”), choosing as predicting (Habernal, Page 1595 Section 4.1 Predicting convincingness of pairs: “Their output is then concatenated into a single drop-out layer and passed to the final sigmoid layer for binary predictions”), from the two or more input passages as the ranked arguments (Id), a [[as the highest ranked argument (Id) that is responsive to the query as the prompt (Id); and
[g] outputting the [[ as HIT workers were presented with an argument pair, the prompt, and the stance as in Figure 1 (Habernal, Page 1591 Section 3.2 Crowdsourcing annotations: “In the HIT, workers were presented with an argument pair, the prompt, and the stance as in Figure 1.”).
Habernal does not explicitly teach A hardware computer-readable medium storing instructions which, when executed by a processor, cause the processor to perform acts comprising: [e] receiving a query via a graphical user interface; [f] … choosing… a selected passage that is responsive to the query; and [g] outputting the selected passage on the graphical user interface.
Comar teaches A hardware computer-readable medium storing instructions which, when executed by a processor, cause the processor (Comar, Column 15, lines 6-9 “typically will include a non-transitory computer-readable medium storing instructions that, when executed by a processor of the server, allow the server to perform its intended functions”) to perform acts comprising:
[e] receiving a query as receiving a request 152 (Comar, Column 14, lines 9-11 “Client devices used to 10 interact with various embodiments can include any appropriate device operable to send and receive requests”; as the selected search results (Comar, Column 3, lines 61-63 “Similarly, the example display 150 of FIG. 1B illustrates a set of search results 154 presented for a different submitted query 152, here the query "women's apparel."”; Figure 1B 152 See query “women’s apparel”) via a graphical user interface (Figure 1B, see display);
[f] based at least on the ranking as ranking based on intent (Column 10, lines 6-9 “the intent logic 320 can attempt to determine the intent behind the request, which can then be used to adjust the ranking and/or selection of content to be provided based at last in part upon the determined intent”), choosing (Column 13, lines 19-22 “This can include, as discussed elsewhere herein, generating an ordered ranking and then selecting at least a subset of highest-ranked content items to be selected for display”), from the two input passages as highest-ranked content items (Id), a selected passage as selecting at least a subset (Id) that is responsive as for display (Id) to the query as based upon the specific intent of the query (Column 13, lines 2-4 “If bias information is available, such as whether a current user or query is associated with a specific intent”); and
[g] outputting the selected passage as the selected search results 154 (Comar, Column 3, lines 61-63 “Similarly, the example display 150 of FIG. 1B illustrates a set of search results 154 presented for a different submitted query 152, here the query "women's apparel."”; Figure 1B, 154) on the graphical user interface (Figure 1B, see display).
It would have been obvious to one of ordinary skill to which said subject matter pertains at the time the invention was filed to have implemented the selecting and ranking model taught by Comar, to select and present search results including ranked arguments as taught by Habernal as it yields the predictable results of displaying content to the user (Comar Column 12, lines 60-67) which may be considered more convincing to the human (Habernal, Page 1594 Section 3.4.2 Transitivity evaluation using argument graphs “We interpret this finding as that it is easier for humans to judge A> B than A=B consistently across all possible pairs of arguments from a given topic.”). One of ordinary skill in the art would recognize that Habernal presents a means of generating a Machine Learning model that is capable of ranking the convincingness of arguments. One of ordinary skill in the art would recognize that this may be implemented within a search recommendation system such as the one taught by Comar, to enable the system to determine which search results the user would find the most convincing. This may be used to determine what would be able to convince or persuade a user to make a particular purchase or prefer a specific item recommendation.
Claims 21-23, 30-32, and 34 are rejected under 35 U.S.C. 103 as being unpatentable over Habernal, Comar and Gao [2020/0370110].
With regard to claims 21 and 30 the proposed combination further teaches the convincingness ranking neural network as the BLSTM (Habernal, Page 1590; Section 1 “We propose a novel task of predicting convincingness of arguments in an argument pair, as well as ranking arguments related to a certain topic. Since no data for such a task are available, we create a new annotated corpus. We employ SVM model with rich linguistic features as well as bidirectional Long Short-Term Memory (BLSTM) neural networks because of their excellent performance across various end-to-end NLP tasks (Goodfellow et al., 2016; Piech et al., 2015; Wen et al., 2016; Dyer et al., 2015; Rocktaschel et al., 2016).”) having separate (Habernal, Page 1595 Section 4.1 Predicting convincingness of pairs: “The core of the model consists of two bi-directional LSTM networks with 64 output neurons each”) [[ (Please note this claim limitation has been read in light of Paragraph [0049] and [0050]) that generate the separate convincingness scores as the assigned real value score to each argument, e.g. the weights (Page 1596 Section 4.2 “We address this problem as a regression task. We use the UKPConvArgRank data, in which a real value score is assigned to each argument so the arguments can be ranked by their convincingness (for each topic independently).”; Page 1593 Section 3.4.2 “Then the weight w of edge ei is computed as follows: [Equation 1]).
Habernal does is silent regarding the BiLSTM is having fully-connected layers.
Gao teaches the convincingness ranking neural network as the BiLSTM(Gao, ¶47 “After concatenating in step 520 the hidden representations of the feature matrices, the method feeds the concatenated representation 522 into a fully-connected layer 530 with 200 nodes, which is followed by a regression layer 540, after which a transformed signal 550 is generated.”; ¶48 “The encodings then go through BiLSTM layers 510, fully-connected layers 530 as well as the final regression layer 540 to generate the expected electrical signals 550.”) having … fully-connected layers as fully-connected layers 530 (Id) that generate the … scores as the expected electrical signals (Id).
It would have been obvious to one of ordinary skill to which said subject matter pertains at the time the invention was filed to have implemented each of the BiLSTM networks as fully-connected networks as taught by Gao as it yields the predictable results of generating a deep neural network that becomes context-dependent after training (Gao, ¶48). One of ordinary skill in the art would recognize the parallel between the BiLSTM taught by Habernal and Gao, and would identify the techniques taught by Gao as providing a more detailed explanation regarding how the BiLSTM may be implemented.
With regard to claims 22, 31 and 34 the proposed combination further teaches the convincingness ranking neural network as the BiLSTM (Habernal, Page 1590; Section 1 “We propose a novel task of predicting convincingness of arguments in an argument pair, as well as ranking arguments related to a certain topic. Since no data for such a task are available, we create a new annotated corpus. We employ SVM model with rich linguistic features as well as bidirectional Long Short-Term Memory (BLSTM) neural networks because of their excellent performance across various end-to-end NLP tasks (Goodfellow et al., 2016; Piech et al., 2015; Wen et al., 2016; Dyer et al., 2015; Rocktaschel et al., 2016).”) having a softmax layer as the final sigmoid layer (Habernal, Page 1595, Section 4.1 Predicting convincingness of pairs: “The core of the model consists of two bi-directional LSTM networks with 64 output neurons each. Their output is then concatenated into a single drop-out layer and passed to the final sigmoid layer for binary predictions.”; Please note the term “softmax” has been interpreted as the final layer as depicted in Figure 4E element 498 of the original specification) that generates relative convincingness probabilities as binary prediction (Id) for the two or more passages based at least on as the outputs of the two bi-directional LSTM networks are concatenated and passed to the final layer (Id) the separate convincingness scores generated by the separate [[(Habernal, Page 1595 Section 4.1 Predicting convincingness of pairs: “The core of the model consists of two bi-directional LSTM networks with 64 output neurons each”).
Habernal does is silent regarding the BiLSTM is having fully-connected layers.
Gao teaches the convincingness ranking neural network as the BiLSTM (Gao, ¶47 “After concatenating in step 520 the hidden representations of the feature matrices, the method feeds the concatenated representation 522 into a fully-connected layer 530 with 200 nodes, which is followed by a regression layer 540, after which a transformed signal 550 is generated.”; ¶48 “The encodings then go through BiLSTM layers 510, fully-connected layers 530 as well as the final regression layer 540 to generate the expected electrical signals 550.”) having a softmax layer as the final regression layer (Id) hat generate relative … probabilities as generated expected signals (Id) … by the… fully-connected layers as fully-connected layers 530 (Id).
It would have been obvious to one of ordinary skill to which said subject matter pertains at the time the invention was filed to have implemented each of the BiLSTM networks as fully-connected networks as taught by Gao as it yields the predictable results of generating a deep neural network that becomes context-dependent after training (Gao, ¶48). One of ordinary skill in the art would recognize the parallel between the BiLSTM taught by Habernal and Gao, and would identify the techniques taught by Gao as providing a more detailed explanation regarding how the BiLSTM may be implemented.
With regard to claims 23 and 32 the proposed combination further teaches inputting one or more first word vectors representing a first input passage as a first word embedding (Habernal, Page 1595 Section 4.1 Predicting convincingness of pairs: “The input layer relies on pre-trained word embeddings, in particular Glo Ve (Pennington et al., 2014) trained on 840B tokens from Common Crawl; 10 the embedding weights are further updated during training.”) to a first set of layers of the convincingness ranking neural network model as the first of the two bi-directional LSTM networks (Habernal, Page 1595 Section 4.1 Predicting convincingness of pairs: “The core of the model consists of two bi-directional LSTM networks with 64 output neurons each. Their output is then concatenated into a single drop-out layer and passed to the final sigmoid layer for binary predictions.”), the first set of layers including multiple first [[ as the bi-directional LSTM must include multiple layers for it to be bi-directional (Id); and
inputting one or more second word vectors representing a second input passage as a first word embedding (Habernal, Page 1595 Section 4.1 Predicting convincingness of pairs: “The input layer relies on pre-trained word embeddings, in particular Glo Ve (Pennington et al., 2014) trained on 840B tokens from Common Crawl; 10 the embedding weights are further updated during training.”) to a second set of layers of the convincingness ranking neural network model as the second of the two bi-directional LSTM networks (Habernal, Page 1595 Section 4.1 Predicting convincingness of pairs: “The core of the model consists of two bi-directional LSTM networks with 64 output neurons each. Their output is then concatenated into a single drop-out layer and passed to the final sigmoid layer for binary predictions.”), the second set of layers including multiple second as the bi-directional LSTM must include multiple layers for it to be bi-directional (Id) that are separate from the one or more first [[ as the two bi-directional LSTM (Habernal, Page 1595 Section 4.1 Predicting convincingness of pairs: “The core of the model consists of two bi-directional LSTM networks with 64 output neurons each. Their output is then concatenated into a single drop-out layer and passed to the final sigmoid layer for binary predictions.”).
Habernal does is silent regarding the BiLSTM is having fully-connected layers.
Gao teaches a first set of layers of the convincingness ranking neural network model as the BiLSTM (Gao, ¶47 “After concatenating in step 520 the hidden representations of the feature matrices, the method feeds the concatenated representation 522 into a fully-connected layer 530 with 200 nodes, which is followed by a regression layer 540, after which a transformed signal 550 is generated.”; ¶48 “The encodings then go through BiLSTM layers 510, fully-connected layers 530 as well as the final regression layer 540 to generate the expected electrical signals 550.”), the first set of layers including multiple first [[ as the fully-connected layers 530 (Id) having a softmax layer as the final regression layer 542 (Id) that generate relative … probabilities as generated expected signals (Id) …
the second set of layers including multiple second as the fully-connected layers (Id) … the one or more first [[ as the fully-connected layers (Id).
It would have been obvious to one of ordinary skill to which said subject matter pertains at the time the invention was filed to have implemented each of the BiLSTM networks as fully-connected networks as taught by Gao as it yields the predictable results of generating a deep neural network that becomes context-dependent after training (Gao, ¶48). One of ordinary skill in the art would recognize the parallel between the BiLSTM taught by Habernal and Gao, and would identify the techniques taught by Gao as providing a more detailed explanation regarding how the BiLSTM may be implemented. It should be noted that Habernal teaches the BLSTM neural network having “to bi-directional LSTM networks with 64 output neurons each”. The outputs from these are passed to the final sigmoid layer for final prediction. Within the proposed combination, each of the two bi-directional LSTM networks have been modified in view of Gao to be fully-connected BiLSTM layers, which feed their output to the “Final Regression layer (e.g. the final sigmoid layer).
Response to Arguments
Applicant’s arguments have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. All the arguments regarding the newly added limitations are addressed in the above rejections.
With regard to the “separate scores” applicant argues that Habernal generates a binary prediction for each passage.
In response, Habernal explicitly states that it generates a convincingness score for each passage (Page 1596 Section 4.2 “We address this problem as a regression task. We use the UKPConvArgRank data, in which a real value score is assigned to each argument so the arguments can be ranked by their convincingness (for each topic independently).”; Page 1593 Section 3.4.2 “Then the weight w of edge ei is computed as follows: [Equation 1]). The real value score generated for each argument so that each argument can be ranked by their convincingness independently. The “binary relation” is the determination of if argument 1 is more convincing than argument 2 (See Habernal, Page 1952, Section 3.4.2 Transitivity evaluation using argument graphs “Considering Al is more convincing than A2 as a binary relation R”). This binary prediction of which argument is more convincing is generated based on the individual convincingness scores assigned to each argument.
Based upon the above reasoning and rational the applied art reads on the claim language.
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 AMANDA WILLIS whose telephone number is (571)270-7691. The examiner can normally be reached Monday-Friday 8am-2pm.
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, Ajay Bhatia can be reached at 571-272-3906. 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.
/AMANDA L WILLIS/ Primary Examiner, Art Unit 2156
Prosecution Insights