DETAILED ACTION
This nonfinal action is in response to the amendment filed 12/02/2025, and remarks filed 11/18/2025 and 12/02/2025, for application 17/649,970.
Claims 1-2, 4, 9, 16, and 21-22 have been amended.
Claims 1-14, 16, and 19-23 remain pending in the application. Claims 1, 9, and 16 are the pending independent claims.
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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous office action mailed 09/03/2025 has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed 12/02/2025 has been entered.
Response to Amendment
The amendment filed 12/02/2025 has been entered.
Applicant’s amendment to the claims with respect to resolving claim objections and indefiniteness rejections under 35 U.S.C. 112(b) has been considered, and the
objections and 112(b) rejections set forth in the office action mailed 09/30/2025 are
consequently withdrawn.
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 1-14, 16, and 19-23 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.
Regarding claim 1, it recites the following limitations:
“converting, by the computing system, the tracking data to a plurality of two-dimensional graph-based representations, wherein each two-dimensional graph-based representation corresponds to a frame of the plurality of frames, wherein each two-dimensional graph-based representation has a plurality of nodes corresponding to a plurality of players of the first team and the second team, wherein a first node of the plurality of nodes has at least one arrow directed towards a second node of the plurality of nodes, and wherein a feature of the arrow indicates a likelihood of a player corresponding to the second node becoming the pass receiver within the possession; and
modeling, by the computing system via the prediction engine, defensive behavior of the first team in the event based on the plurality of two-dimensional graph-based representations, the modeling further comprising:
outputting, via the first graph neural network, the second team's likelihood of completing a pass within the possession:
outputting, via the second graph neural network, the second team's likelihood of a shot occurring within the possession:
outputting, via the third graph neural network, the second team's likelihood of each player becoming a pass receiver within the possession…”.
As per the above limitations, the claims appear to recite, prior to receiving outputs of the graph neural networks of the prediction engine, a step of converting tracking data to graph-based representations, wherein the graph-based representations include a feature value representing “a likelihood of a player [corresponding to the second node] becoming the pass receiver within the possession”.
However, the claims later recite “outputting, via the third graph neural network, [the second team’s] likelihood of each player becoming a pass receiver within the possession”. Because the step of converting tracking data is recited prior to deployment of the prediction engine (including the third graph neural network), the claims thereby appear to recite that rather than the “likelihood” value in the converting step being obtained from the prediction engine output, that it is instead a separately obtained “likelihood” value.
However, it is unclear how conversion of the tracking data, prior to deployment of the prediction engine, would be able to calculate and include feature values indicating “likelihood of a player becoming the pass receiver within the possession” without first receiving output from the prediction engine. The specification also does not provide appear to provide support for such a configuration. While constructed graph-based representations of tracking data are explained to potentially include node features and/or edge features [¶ 0067-0068], these features do not include predicted likelihood values. In fact, the recited “likelihood” value is only calculated upon inputting these graph-based representation into the prediction engine [¶ 0069-0071]. Further while the specification does further disclose modeling “likelihood” values in certain graphical representations, those graphical representations are explained to be a representation of the outputs produced by the prediction engine [¶ 0073-0074] (i.e., generated based on output from the prediction engine, concurrent to the disruption map [¶ 0076-0077]), and therefore are distinct from the graph-based representations which are drawn from tracking data and input to the prediction engine.
Ultimately, the apparent inconsistencies between the specification and claims, and failure to clearly interrelate essential claim elements, renders their scope uncertain. Consequently, one of ordinary skill in the art would not be reasonably apprised of the scope of the invention.
For purposes of examination and as best understood in light of the instant specification [¶ 0067-0077] the examiner has interpreted the claims as follows:
“converting, by the computing system, the tracking data to a plurality of two-dimensional graph-based representations, wherein each two-dimensional graph-based representation corresponds to a frame of the plurality of frames, wherein each two-dimensional graph-based representation has a plurality of nodes corresponding to a plurality of players of the first team and the second team, wherein a first node of the plurality of nodes has at least one arrow directed towards a second node of the plurality of nodes; and
modeling, by the computing system via the prediction engine…
outputting, via the first graph neural network…
outputting, via the second graph neural network…
outputting, via the third graph neural network, the second team's likelihood of each player becoming a pass receiver within the possession; and
generating a graph-based representation based on outputs of the prediction engine, wherein a feature of an arrow directed from a first node to a second node in the graph-based representation indicates a likelihood of a player corresponding to the second node becoming the pass receiver within the possession; and
generating a disruption map…”
Regarding claims 9 and 16, they have the same deficiencies as those found in claim 1 above. Consequently, they are rejected for the same reasons and are likewise interpreted as detailed above.
Regarding claims 2-8, 10-14, and 19-23, they inherit the deficiencies of their parent claims. Consequently, they are also rejected under 35 U.S.C. 112(b) as being indefinite for depending on an indefinite parent claim.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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 1-14, 16, and 19-22 are rejected under 35 U.S.C. 103 as being unpatentable over Chang et al., (Pub. No. US 20200074181 A1, “Data Processing Systems and Methods for Generating Interactive User Interfaces and Interactive Game Systems Based on Spatiotemporal Analysis of Video Content”, filed 11/06/2019, cited in IDS), hereinafter Chang, further in view of Qi et al., (“stagNet: An Attentive Semantic RNN for Group Activity and Individual Action Recognition", available 2020), hereinafter Qi, Spearman et al., (“Physics-Based Modelling of Pass Probabilities in Soccer”, available 2017), hereinafter Spearman, and Fernandez et al., (“SoccerMap: A Deep Learning Architecture for Visually-Interpretable Analysis in Soccer”, available arXiv 20 Oct 2020), hereinafter Fernandez.
Regarding claim 1, Chang teaches A method, comprising:
retrieving, by a computing system, training tracking data from a data store, the training tracking data comprising a plurality of training frames of data for a plurality of events across a plurality of seasons; ("In operation, a system 4800 (e.g., the system described herein) may output one or more content feeds 4802-1, 4802-2 . . . 4802-N. The content feeds may include video, audio, text, and/or data (e.g., statistics of a game, player names). In some embodiments, the system 4800 may output a first content feed 4802-1 that includes a video and/or audio that is to be output (e.g., displayed) by a client media player 4808. The client media player 4808 may be executed by a user device (e.g., a mobile device, a personal computing device, a tablet computing device, and the like). The client media player 4808 is configured to receive the first content feed 4802 and to output the content feed 4802 via a user interface (e.g., display device and/or speakers) of the user device...Regardless of the source, a content feed 4802-2 or 4812 may include timestamps or other suitable temporal indicia to identify different positions (e.g., frames or chunks) in the content feed" [Chang ¶ 0281]; "In the illustrated example, the client device 5100 may include a processing device 5102, a storage device 5104, a communication unit 5106 that effectuates communication between the client device and other devices via one or more communication networks (e.g., the Internet and/or a cellular network), and a user interface 5108" [Chang ¶ 0308]; "FIGS. 6A and 6B show a set of filters in the UI, which can be used to filter particular items to obtain greater levels of detail or selected sets of results. Filters may exist for seasons, games...offensive team, defensive team, [p]layers on the court for offense/defense, players off court for offense/defense...offensive or defensive statistics...and various other features" [Chang ¶ 0147]; Chang discloses a client device that receives a content feed of data via an included storage, wherein the data is outputted by a system and includes game tracking data. Data can be filtered by the UI of the client device for a plurality of games (i.e., events) across a plurality of seasons)
converting, by the computing system, the training tracking data into a plurality of training two-dimensional graph-based representations; ("The visualizations layer 108 may allow dynamic visualizations of patterns and analytics developed from the data obtained from the real-time event...The visualizations layer 108 may use various types of visualizations and graphical tools for creating visual depictions. The visuals may include various types of interactive charts, graphs, diagrams, comparative analytical graphs, and the like" [Chang ¶ 0070]; Real-time event data (i.e., tracking data) can be converted into graph-based representations via a visualizations layer)
learning, by a prediction engine, to model defensive behavior by: (“"In embodiments, the event analytics and/or location-based games may include prediction-based scoring including generating or contributing to a user score based on the accuracy of an outcome prediction associated with the user. By way of this example, the outcome prediction may be associated with outcomes of individual offensive and defensive plays in the games and/or may be associated with scoring and/or individual player statistics at predetermined time intervals" [Chang ¶ 0283]; “...outputs generated by the computer-controlled intelligent systems may enable spatiotemporal analysis of various game attributes and elements for exploring, learning, analyzing such sporting events and utilize analytics results to generate predictive models and predictive analytics for gaming strategy” [Chang ¶ 0447]; Chang discloses learning, via predictive models and predictive analytics, to model game elements, including defensive behavior, via spatiotemporal analysis)
learning to predict a likelihood of a pass being completed at any moment within a possession, ("A spatiotemporal event may include, for example: (1) a particular play during the sporting event; (2) a particular time period during the sporting event (e.g., a quarter, half, etc.)…(4) a particular action by one or more players during the sporting event (e.g., a pass, an attempted shot, a scored shot...a particular movement, a particular off-the-ball movement, a run...and/or any other suitable potential action which a player may make during the course of any suitable sporting event)" [Chang ¶ 0503]; "By way of this example, the outcome prediction may be associated with outcomes of individual offensive and defensive plays in the games and/or may be associated with scoring and/or individual player statistics at predetermined time intervals...In embodiments, the event analytics and/or location-based games may include…analysis of instantaneous game state and/or comparison with evolution of game state such as maximum value or realized value of the game state in a given chance or possession" [Chang ¶ 0283]; "In still other embodiments, the spatiotemporal event data may include any other suitable data related to each discrete spatiotemporal event such as, for example....(4) a likelihood of the spatiotemporal event having occurred (e.g., a probability that a player will make a particular shot that makes up the spatiotemporal event)" [Chang ¶ 0522]; "In one example, a spatiotemporal event may include a completed pass in a football game. In this example, the spatiotemporal event data may include, for example:...7) probability data related to the caught pass" [Chang ¶ 0526]; Chang discloses predicting likelihood of a spatiotemporal event having occurred during a particular time period or game state (such as a possession), wherein a spatiotemporal event can be any particular action by a player (such as a completed pass))
learning to predict a likelihood of a shot occurring within the possession, ([Chang ¶ 0503, 0283, 0522]; Chang discloses predicting likelihood of a spatiotemporal event having occurred during a particular time period or game state (such as a possession), wherein a spatiotemporal event can be any particular action by a player (such as an attempted shot or scored shot))
receiving, by the computing system, tracking data for an event including a first team and a second team, the tracking data comprising a plurality of frames; ([Chang ¶ 0281, 0308]; "The machine learning tools may further allow to generate outputs based on a user query input such as to determine various predictive analytics for a particular team player in view of historical shots and screens in a particular context, determine possibilities of success and failures in particular zones and game scenarios conditioned to particular user inputs, and the like" [Chang ¶ 0446]; Chang discloses a client device that receives a content feed of data, wherein the data is outputted by a system and includes game (i.e., event with two teams) tracking data. Data can be marked by temporal indicia to identify different frames, and data can also be queried by the user based on a particular context).
converting, by the computing system, the tracking data to a plurality of two-dimensional graph-based representations; ([Chang ¶ 0070]; Tracking data can be converted into graph-based representations via a visualizations layer)
and modeling, by the computing system via the trained prediction engine, defensive behavior of the first team in the event ([Chang ¶ 0446, 0283, 0447]; Chang discloses modelling particular analytics for particular players, including defensive plays, via a trained predictive model), the modeling further comprising:
outputting, via the first graph neural network, the second team's likelihood
of completing a pass within the possession ([Chang ¶ 0503, 0283, 0522] as detailed above; “In various embodiments, the system is configured to determine the spatiotemporal event score (e.g., for any suitable player in the event) based on any suitable factor (e.g., piece of spatiotemporal event data) and/or scoring criteria described herein” [Chang ¶ 0509])
outputting, via the second graph neural network, the second team's
likelihood of a shot occurring within the possession ([Chang ¶ 0503, 0283, 0522, 0509] as detailed above)
However, Chang does not explicitly teach learning spatiotemporal event likelihoods by a [first/second/third] graph neural network, or
learning to model defensive behavior or modeling defensive behavior based on the plurality of [training] two-dimensional graph-based representations, wherein each two-dimensional graph-based representation corresponds to a frame of the plurality of frames, and wherein each two-dimensional graph-based representation has a plurality of nodes corresponding to a plurality of players of the first team and the second team, wherein a first node of the plurality of nodes has at least one arrow directed towards a second node of the plurality of nodes.
In the same field of endeavor, Qi teaches a method of modeling and predicting spatiotemporal events within sports events using tracking data ("Understanding dynamic scenes in sports games and surveillance videos encompasses a wide range of applications, such as sports team tactics analysis...we present a novel attentive semantic recurrent neural network, called stagNet for group activity and personal action recognition, which combines spatial-temporal attention and semantic graph. Specifically, individual activities and their spatial relations are inferred and depicted by an explicit semantic graph, and their temporal interactions are integrated by a structural-RNN model" [Qi Introduction pages 1-2]) that
learns likelihoods of spatiotemporal events by a [first/second/third] graph neural network, ("In [35], multi-class object detection [19], [39] and fully convolutional network [40] were adopted to capture multi-scale features for estimating individual actions and collective activities with probability inference...However, most of these works either extracted individual features in spite of the scene context or captured the context in an implicit manner without any semantic information. In this paper, we conceive to explicitly capture the semantic context of the scene by an expressive spatio-temporal semantic graph [43] through RNNs" [Qi Group Activity Recognition page 3]; "We inference the semantic graph to predict person’s affiliations based on their positions and visual appearance...A total of three kinds of information are inferred by modeling the graph: (1) the individual action label, (2) the inter-group relationships, and (3) the group activity label...Given a set of the scene labels (i.e. group activity label) Cscene, and individual action labels set Caction, we define...xi-->j ∈ R as the predicted relationship between the i -th and j -th person proposal boxes....In particular, the semantic graph is constructed by seeking out the optimal yt* and x* that maximize probability function as follows: [equation 1]" [Qi Semantic Graph page 4]; Qi discloses describing scene content by creating semantic graphs through RNNs (i.e., graph neural networks), and using the semantic graphs to predict the occurrence of spatiotemporal events, including individual events (such as a shot) or inter-group/inter-personal events (such as a pass between players). Qi performs the prediction by choosing event labels that maximize a probability (i.e., likelihood) function),
and learns to model defensive behavior and models defensive behavior based on the plurality of [training] graph-based representations ("Specifically, individual activities and their spatial relations are inferred and depicted by an explicit semantic graph...We introduce a semantic graph to describe explicitly all the content in the scene, i.e. group activity, individuals’ actions and their spatial relations" [Qi Introduction page 2]; "Volleyball Dataset [1] consists of 55 video clips of volleyball games with 4,830 labeled frames in total... we select 2/3 of the dataset as training set and the rest 1/3 as the testing set…[we] define four team-level activities additionally: attack, defense, win and lose in our experiments" [Qi Experiments page 8]; Qi discloses using semantic graphs (i.e., graph-based representations) to model scene content including group activity; Qi further discloses an experiment using a volleyball dataset wherein defense (i.e. defensive behavior) can be a group activity; furthermore, part of the data is used as training data for the StagNet model. and part is used as test data (i.e., learning to model behavior via training, and then further modeling behavior during testing)), wherein each two-dimensional graph-based representation corresponds to a frame of the plurality of frames of the tracking data ("We extend our semantic graph model to the temporal dimension between frames in a video via a structural-RNN, which is achieved by adopting the ‘factor sharing’ mechanism" [Qi Introduction page 2]; "With the semantic graph of a frame, temporal factors are integrated to construct the spatio-temporal semantic graph (see Fig. 2(c)) with the structural-RNN [12]" [Qi Integrating Temporal Factors page 6]; Fig. 2(b) [see Fig. 2 on page 4] demonstrates each frame having a corresponding semantic graph (i.e. graph-based representation), and Fig. 2(c) demonstrates extending semantic graphs to the temporal dimension by integrating across multiple frames), and wherein each two-dimensional graph-based representation has a plurality of nodes corresponding to a plurality of players of the first team and the second team, wherein a first node of the plurality of nodes has at least one arrow directed towards a second node of the plurality of nodes ("We define the semantic graph as G (S, V, E), where S is the scene node, and V and E are the object nodes and edges respectively. Concretely, S represents the global representation of a frame in a video, an object node vi ∈ V (i = 1,...,K) refers to the person-level proposal (where i = 1,...,K corresponds to the totally K persons in the scene), and the edge E corresponds to the spatial configuration of object nodes V in the frame" [Qi Graph Inference page 4]; Qi discloses the semantic graph representation of a frame having K number of object nodes, wherein each object node corresponds to a person (i.e., player) in the frame, and further discloses edges (i.e., arrows) E that connect object nodes V of the semantic graph representation based on a spatial configuration)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have incorporated learning spatiotemporal event likelihoods by a [first/second/third] graph neural network and learning to model defensive behavior or modeling defensive behavior based on the plurality of [training] two-dimensional graph-based representations, wherein each two-dimensional graph-based representation corresponds to a frame of the plurality of frames, and wherein each two-dimensional graph-based representation has a plurality of nodes corresponding to a plurality of players of the first team and the second team, wherein a first node of the plurality of nodes has at least one arrow directed towards a second node of the plurality of nodes as taught by Qi into Chang because both of these methods (and corresponding systems) are directed to modeling and predicting spatiotemporal events within sports events using tracking data. Incorporating the teachings of Qi into Chang would enable the system to more successfully capture structural semantic information and spatio-temporal information for each frame of video (“…our model outperforms other RNN-based approaches (e.g. HDTM/CERN/SRNN) by about 5 8% w.r.t. group activity recognition, because semantic graph with structural-RNN in our model can extract and model better spatio-temporal relationships…all the other state-of-the-art are unable to capture the semantic structural information for describing the scene” [Qi Results on the Volleyball Dataset page 10]), expanding possible applications of the system to other related areas, such as sports video captioning ("Besides, the structural semantic output is beneficial for lots of other tasks like dense video captioning [82], sports video captioning [76] and visual question answering [83] as it provides mid-level relationships for fine-grained recognition" [Qi Discussion page 15]).
The combination of Chang and Qi further discloses determining "probability data related to [a] caught pass" based on "a continuously determined position of each player during the course of the play", "relative positioning data for one or more players in the game", and "information associated with a player that caught the pass" including "a pass catch rate" [Chang ¶ 0526].
However, the combination of Chang and Qi does not explicitly teach [learning to predict / outputting] a likelihood of each player becoming a pass receiver within the possession.
In the same field of endeavor, Spearman teaches a method of modeling and predicting spatiotemporal events within sports events using tracking data (“We use this model to quantify the likelihood that a given pass will succeed we determine the free parameters of the model using tracking and event data from the 2015-2016 Premier League season" [Spearman Abstract]; "Predictive: We require that the pass probability model be a predictive one. In other words, the model must be able to predict the probability of the pass using only information from the moment the pass occurs" [Spearman Introduction page 1]; "For each game, we have access to tracking data which gives the position of every player and the ball at 25 frames per second for the entire match. Additionally, we have access to event data which records events that occur during the match such as goals, passes, fouls, and many other soccer specific events" [Spearman Data page 7]) that learn[s] to predict a likelihood of each player becoming a pass receiver within the possession ("We use the following event types as indicative of a controlled touch and therefore, the receiver: pass, tackle…This means that events that do not indicate clear possession such as participation in an aerial duel are not considered when determining the eventual pass recipient" [Spearman Processing and Selection page 7]; "The probability that a player j will receive the pass within time t is given by: [equation 6]...This gives us the total probability Pj that a specific player j will receive the pass. To determine the probability of a successful pass, we sum the probabilities for each player on the passing team, excluding the passer" [Spearman Model pages 5-6]; Spearman discloses calculating a probability for each specific player on the passing team that they will receive the pass within the time frame of a possession event (e.g. a pass)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have incorporated [learning to predict / outputting] a likelihood of each player becoming a pass receiver within the possession as taught by Spearman into the combination of Chang and Qi because all of these methods (and corresponding systems) are directed to modeling and predicting spatiotemporal events within sports events using tracking data. Incorporating the teachings of Spearman into the combination of Chang and Qi would improve the combination by providing further metrics for quantifying the value of passes, analyzing the skill of receivers and defenders, and performing post-match analysis and player scouting [Spearman Abstract].
However, the combination of Chang, Qi, and Spearman does not explicitly teach generating a graph-based representation based on outputs of the prediction engine, wherein a feature of an arrow directed from a first node to a second node in the graph-based representation indicates a likelihood of a player corresponding to the second node becoming the pass receiver within the possession (examiner note: see 112(b) rejection of claim 1 above) or generating a disruption map having a compact weighted two-dimensional representation of the first team’s effect on the second team’s attacking strategy, the attacking strategy including predicted likelihoods (as outputted by the [first/second/third] graph neural network of Chang).
In the same field of endeavor, Fernandez teaches a method of modeling and predicting spatiotemporal events within sports events using tracking data (“From an applied standpoint, our work is related to several other approaches aimed at estimating pass probabilities and other performance metrics derived from spatiotemporal data in soccer” [Fernandez page 2 Related Work]) that generat[es] a graph-based representation based on outputs of the prediction engine, wherein a feature of an arrow directed from a first node to a second node in the graph-based representation indicates a likelihood of a player corresponding to the second node becoming the pass receiver within the possession (see Fig. 5 – “Fig. 5. In the left column, we present a game-state where red circles represent the optimal passing location for each teammate, and the expected pass probability. In the
right column, the green circles represent the optimal positioning of players increasing
the expected pass probability if the players were placed in those locations at that time.
(Color figure online)” [Fernandez page 13]) and generat[es] a disruption map having a compact weighted two-dimensional representation of the first team’s effect on the second team’s attacking strategy, the attacking strategy including predicted likelihoods (“The pass selection adaptation of SoccerMap, presented in Section 5.1, provides a fine-grained evaluation of the passing likelihood in different situations. However, it is clear to observe that passing selection is likely to vary according to a team’s player style and the specific game situation…Once we train a SoccerMap network to obtain this league-wide model, we can fine-tune the network with passes from each team to grasp team-specific behavior… In Figure 6 we compare the pass selection tendencies between Liverpool (left column) and Burnley (right column). On the top left corner of both columns, we show a 2D plot with the difference between the league mean passing selection heatmap, and each team’s mean passing selection heatmap, when the ball is within the green circle area.…In the two plots of Figure 6 we show over each players’ location the percentage increase in passing likelihood compared with the league’s mean value. In this situation, we can observe that when a left central defender has the ball during a buildup, Liverpool will tend to play short passes to the closest open player, while Burnley has a considerably higher tendency to play long balls to the forwards, especially if forwards are starting a run behind the defender’s backs, such as in this case. Through a straightforward fine-tuning of the SoccerMap-based model, we can provide detailed information to the coach for analyzing specific game situations” [Fernandez pages 14-15 Team-Based Passing Selection Tendencies]; see team-respective plots (Liverpool (left column) and Burnley (right column)) in Fig. 6 – “A game state representation of a real game situation in soccer” [Fernandez page 15]; The SoccerMap model can be adapted to generate two-dimensional plot representations (Fig 6) of the soccer field (i.e., maps having a compact two-dimensional representation) that model the effect (i.e., disruption) of one team’s defensive behavior (e.g., left central defender) on the opposition’s behavior (e.g., Liverpool having tendency to play short passes) based on the mean passing selection of the opposition (i.e., global averages), and translate said behavior to pass likelihood values (i.e., weights) corresponding to each player mapped on the plot of the field)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have incorporated generating a graph-based representation based on outputs of the prediction engine, wherein a feature of an arrow directed from a first node to a second node in the graph-based representation indicates a likelihood of a player corresponding to the second node becoming the pass receiver within the possession and generating a disruption map having a compact weighted two-dimensional representation of the first team’s effect on the second team’s attacking strategy, the attacking strategy including predicted likelihoods as taught by Fernandez into the combination of Chang, Qi, and Spearman because they are all directed towards modeling and predicting spatiotemporal events within sports events using tracking data. Given that Chang already produces predicted likelihood values corresponding to game events (passes/shots/etc,.), incorporating the teachings of Fernandez would thereby allow for visual representation of the inferences made from the performed analytics, enabling complex information to be made digestible to users (e.g., soccer coaches) (“The presented architecture allows generating visual tools to help coaches perform fine-tuned analysis of opponents and own-team performance, derived from low-level spatiotemporal soccer data. We show how this network can be easily adapted to many other challenging related problems in soccer, such as the estimation of pass selection likelihood and pass value, and that can perform remarkably well at estimating the probability of observed passes…This framework of analysis derived from spatiotemporal data could also be applied directly in many other team sports, where the visual representation of complex information can bring the coach and the data analyst closer” [Fernandez page 15 Discussion and Future Work]).
Regarding claim 2, the combination of Chang, Qi, Spearman, and Fernandez teaches the limitations of parent claim 1 and detecting, by the computing system, active runs of offensive players based on the modeled defensive behavior of the first team in the event ([Chang ¶ 0283, 0446]; "A spatiotemporal event may include, for example:...(4) a particular action by one or more players during the sporting event (e.g.,...a particular movement, a particular off-the-ball movement, a run...and/or any other suitable potential action which a player may make during the course of any suitable sporting event)" [Chang ¶ 0503]; Examiner has interpreted "active runs" as described in instant specification ("In some embodiments, prediction engine 120 may utilize the various GNNs (e.g., pass model 126, threat model 128, and/or receiver model 130) to detect "active runs" of offensive players. An "active run" may refer to a run (i.e., detected as a run or movement based on velocity)" [specification ¶ 0044]). Chang discloses that a spatiotemporal event may include a detected movement or run (i.e., an active run) of any player (including offensive players), and may be associated with predicted defensive plays, wherein the defensive plays may be from any particular players in any particular context (e.g., team, event)). Qi further teaches modeling defensive behavior based on the plurality of graph-based representations ([Qi Introduction page 2]; [Qi Experiments page 8]; Qi discloses using semantic graphs (i.e., graph-based representations) to model scene content including group activity; Qi further discloses an experiment using a volleyball dataset wherein defense (i.e. defensive behavior) can be a group activity), as detailed in parent claim 1 above.
Regarding claim 3, the combination of Chang, Qi, Spearman, and Fernandez teaches the limitations of parent claim 2 and breaking down, by the computing system, the active runs of the offensive players based on various metrics associated therewith ("In various embodiments, the system may be configured to determine and/or receive data related to each of the plurality of discrete events (e.g., spatiotemporal events) over the course of the overall sporting event" [Chang ¶ 0503]; Chang discloses the system being configured to determine related data for any spatiotemporal event (i.e., break down the event based on various metrics), wherein a spatiotemporal event can be an active run).
Regarding claim 4, the combination of Chang, Qi, Spearman, and Fernandez teaches the limitations of parent claim 1 and assigning, by the computing system, a value to each player of the first team in the event based on the modeled defensive behavior ("...as the interactive game system receives spatiotemporal event data regarding events (e.g., actions, activities, etc.) that occur during the course of the sporting event, the system may be configured to determine a score for any particular player involved in a particular event" [Chang ¶ 0508]; Chang discloses the system being configured to determine a score (i.e., value) for any particular player in an event, including every player on a particular team, based on the modeled spatiotemporal events (including defensive behavior [¶ 0283])).
Regarding claim 5, the combination of Chang, Qi, Spearman, and Fernandez teaches the limitations of parent claim 1 and wherein converting, by the computing system, the tracking data into the plurality of graph-based representations, comprises: for each frame, generating a node representation of each player in the frame ("We define the semantic graph as G (S, V, E), where S is the scene node, and V and E are the object nodes and edges respectively. Concretely, S represents the global representation of a frame in a video, an object node vi ∈ V (i = 1,...,K) refers to the person-level proposal (where i = 1,...,K corresponds to the totally K persons in the scene), and the edge E corresponds to the spatial configuration of object nodes V in the frame" [Qi Graph Inference page 4]; Qi discloses the semantic graph representation of a frame having K number of object nodes, wherein each object node corresponds to a person (i.e., player) in the frame).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have incorporated for each frame, generating a node representation of each player in the frame as taught by Qi into Chang because both of these methods (and corresponding systems) are directed to modeling and predicting spatiotemporal events within sports events using tracking data. Generating node representations is intrinsic to generating semantic graphs; therefore, incorporating the teachings of Qi into the system of Chang would further enable the system to successfully capture structural semantic information for each frame of video [Qi Results on the Volleyball Dataset page 10], and expand possible applications of the system to other related areas, such as sports video captioning [Qi Discussion page 15].
Regarding claim 6, the combination of Chang, Qi, Spearman, and Fernandez teaches the limitations of parent claim 5 and features compris[ing] at least one of player (x, y) position, speed of the player, acceleration of the player, a first angle of motion of the player, a first distance from the player to an attacking goal or a basket, a second angle between the player and the attacking goal or the basket, a second distance from the player to a ball carrier, a difference in the first angle of motion between the player and the ball carrier, and a flag that indicates whether the player is the ball carrier (“The spatiotemporal event data may include any other suitable data related to each discrete spatiotemporal event such as, for example…2) a location of each player participating in the sporting event during the spatiotemporal event…(3) a relative position of each player participating in the sporting event (e.g., a distance between at least two players during the spatiotemporal event)…(5) a movement speed of each particular participant in the sporting event during any particular spatiotemporal event” [Chang ¶ 0522]; Chang discloses spatiotemporal event data (i.e., features) comprising player position and player movement speed). Qi further teaches for each node, storing a plurality of node features therein, ("With the semantic graph of a frame, temporal factors are integrated to construct the spatio-temporal semantic graph (see Fig. 2(c)) with the structural-RNN [12]...We define the node label as ytv and corresponding feature vectors for node and edge are referred to ftv , fte at time t, respectively" [Qi Integrating Temporal Factors page 6]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have incorporated for each node, storing a plurality of node features therein, as taught by Qi into Chang because both of these methods (and corresponding systems) are directed to modeling and predicting spatiotemporal events within sports events using tracking data. Generating nodes with stored features is intrinsic to generating semantic graphs; therefore, incorporating the teachings of Qi into the system of Chang would further enable the system to successfully capture structural semantic information for each frame of video [Qi Results on the Volleyball Dataset page 10], and expand possible applications of the system to other related areas, such as sports video captioning [Qi Discussion page 15].
Regarding claim 7, the combination of Chang, Qi, Spearman, and Fernandez teaches the limitations of parent claim 6 and connecting each node in the frame using one or more edges ("We define the semantic graph as G (S, V, E), where S is the scene node, and V and E are the object nodes and edges respectively. Concretely, S represents the global representation of a frame in a video, an object node vi ∈ V (i = 1,...,K) refers to the person-level proposal (where i = 1,...,K corresponds to the totally K persons in the scene), and the edge E corresponds to the spatial configuration of object nodes V in the frame" [Qi Graph Inference page 4]; Qi discloses edges E that connect object nodes V of the semantic graph representation based on a spatial configuration).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have incorporated connecting each node in the frame using one or more edges as taught by Qi into Chang because both of these methods (and corresponding systems) are directed to modeling and predicting spatiotemporal events within sports events using tracking data. Connecting nodes with edges is intrinsic to generating semantic graphs; therefore, incorporating the teachings of Qi into the system of Chang would further enable the system to successfully capture structural semantic information for each frame of video [Qi Results on the Volleyball Dataset page 10], and expand possible applications of the system to other related areas, such as sports video captioning [Qi Discussion page 15].
Regarding claim 8, the combination of Chang, Qi, Spearman, and Fernandez teaches the limitations of parent claim 7 and for each edge, storing a plurality of edge features therein, wherein the plurality of edge features comprises at least one of a flag defining a relationship between a starting node and an ending node, a distance between two players connected by the edge, and a difference in the angle of motion between the two players connected by the edge (“...We define the node label as ytv and corresponding feature vectors for node and edge are referred to ftv , fte at time t, respectively" [Qi Integrating Temporal Factors page 6]; "an object node vi ∈ V (i = 1,...,K) refers to the person-level proposal (where i = 1,...,K corresponds to the totally K persons in the scene), and the edge E corresponds to the spatial configuration of object nodes V in the frame...fvi is the feature of the i -th node, and feij is the feature of the edge connecting the i -th node and j -th node, which is the unified bounding box over two nodes. We compute feij using six features via calculating the basic distances and direction vectors" [Qi Graph Inference pages 4-5]; Qi discloses connecting object nodes V, wherein each object node corresponds to a person/player, with edges E, wherein features of an edge connecting two nodes can be calculated basic distances).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have incorporated for each edge, storing a plurality of edge features therein, wherein the plurality of edge features comprises at least one of a flag defining a relationship between a starting node and an ending node, a distance between two players connected by the edge, and a difference in the angle of motion between the two players connected by the edge as taught by Qi into Chang because both of these methods (and corresponding systems) are directed to modeling and predicting spatiotemporal events within sports events using tracking data. Edges that connect nodes having stored features provide a greater depth of contextual information on the relationship between nodes than a mere binary (yes/no indication of relationship) value (“As shown in Fig. 2, the edgeRNNs offer contextual information for the nodeRNNs…the edgeRNN passes the summation of all edge features that are connected to the identical node as the message. EdgeRNNS and nodeRNNS take the visual features as initial input and produce a collection of hidden states…Finally, the hidden states of the RNN are utilized to predict…inter-group relationships” [Qi Graph Inference page 5]); therefore, incorporating the teachings of Qi into the system of Chang would further enable the system to successfully capture structural semantic information for each frame of video [Qi Results on the Volleyball Dataset page 10], and expand possible applications of the system to other related areas, such as sports video captioning [Qi Discussion page 15].
Regarding claims 9 and 12-14, they are system/apparatus claims that correspond to the methods of claims 1 and 5-7 respectively. The combination of Chang, Qi, Spearman, and Fernandez further teaches A system, comprising: a processor; and a memory having programming instructions stored thereon, which, when executed by the processor, causes the system to perform one or more operations, comprising: ("In the illustrated example, the client device 5100 may include a processing device 5102...The processing device 5102 may include one or more processors and memory that stores computer-executable instructions that are executed by the one or more processors" [Chang ¶ 0308]). Consequently, claims 9 and 12-14 are rejected for the same reasons as claims 1 and 5-7 above.
Regarding claim 10, the combination of Chang, Qi, Spearman, and Fernandez teaches the limitations of parent claim 9 and wherein learning to predict the first likelihood of the pass being completed at any moment within the possession ([Chang ¶ 0503, 0283, 0522, 0526]; Chang discloses predicting likelihood of a spatiotemporal event having occurred during a particular time period or game state (such as a possession), wherein a spatiotemporal event can be any particular action by a player (such as a completed pass) comprises: identifying a subset of frames prior to the pass ("In these embodiments, a wide variety of elements may be indexed temporally (e.g., in relation to individual video frames)...Examples of elements that may be indexed include events (match/game identifier), objects (players, game objects, objects in the environment such as court or playing field) involved in an event, information and statistics relating to the event..." [Chang ¶ 0308]; "...a particular sporting event (e.g., or other event) may comprise a plurality of discrete events that occur over the course of the event...the system may be configured to determine and/or receive data related to each of the plurality of discrete events (e.g., spatiotemporal events) over the course of the overall sporting event. In particular embodiments, the spatiotemporal event data may include, for example, one or more particular actions undertaken by one or more players in the game, such as: (1) during each discrete event; (2) leading up to each discrete event; and/or (3) after each discrete event" [Chang ¶ 0503]; Chang discloses the system being configured to receive spatiotemporal event data for actions taken leading up to (i.e., prior to) a spatiotemporal event (such as a pass), and that spatiotemporal events may be indexed (i.e., identified) in relation to individual video frames). Qi further teaches learning likelihoods of spatiotemporal events by the first graph neural network ([Qi Group Activity Recognition page 3]; [Qi Semantic Graph page 4]) and identifying a subset of graph representations that correspond to a subset of frames ([Qi Introduction page 2]; [Qi Integrating Temporal Factors page 6]; [see Fig. 2 on page 4]), as detailed in claim 1 (which corresponds with parent claim 9) above.
Regarding claim 11, the combination of Chang, Qi, Spearman, and Fernandez teaches the limitations of parent claim 9 and identifying a subset of frames prior to the pass ([Chang ¶ 0308]; [Chang ¶ 0503]; Chang discloses the system being configured to receive spatiotemporal event data for actions taken leading up to (i.e., prior to) a spatiotemporal event (such as a pass), and that spatiotemporal events may be indexed (i.e., identified) in relation to individual video frames). Qi further teaches learning likelihoods of spatiotemporal events by the third graph neural network ([Qi Group Activity Recognition page 3]; [Qi Semantic Graph page 4]) and identifying a subset of graph based representations that correspond to a subset of frames [Qi Introduction page 2]; [Qi Integrating Temporal Factors page 6]; [see Fig. 2 on page 4]), as detailed in claim 1 (which corresponds with parent claim 9) above. Spearman further teaches learning to predict the third likelihood of each player becoming the pass receiver within the possession ([Spearman Processing and Selection page 7]; [Spearman Model pages 5-6]), as detailed in claim 1 (which corresponds with parent claim 9) above.
Regarding claims 16, 19, and 20, they are product claims that correspond to the method of claims 1, 6, and 8 respectively . The combination of Chang, Qi, Spearman, and Fernandez further teaches A non-transitory computer readable medium comprising one or more sequences of instructions, which, when executed by one or more processors, causes a computing system to perform operations, comprising: ([Chang ¶ 0308]; "Furthermore, particular embodiments may take the form of a computer program product stored on a computer-readable storage medium (e.g., a nontransitory computer-readable medium) having computer-readable instructions (e.g., software) embodied in the storage medium" [Chang ¶ 0479]). Consequently, claims 16, 19, and 20 are rejected for the same reasons as claims 1, 6, and 8 above.
Regarding claim 21, the combination of Chang, Qi, Spearman, and Fernandez teaches the limitations of parent claim 1 and wherein the training tracking data is extracted from a video feed (“The video transformation module 5208 receives the video player instance and obtains video feeds and/or additional content provided by a multimedia server (or analogous device) that may be displayed with the video encoded in the video feeds….In these embodiments, the spatio-temporal index identifies information associated with particular video frames of a video and/or particular locations depicted in the video frames. In some of these embodiments, the locations may be locations in relation to a playing surface (e.g., at the fifty yard line or at the free throw line) or defined in relation to individual pixels or groups of pixels. It is noted that the pixels may be two-dimensional pixels or three-dimensional pixels (e.g., voxels). The spatio-temporal index may index participants on a playing surface (e.g., players on a basketball court), statistics relating to the participants (e.g., Player A has scored 32 points), statistics relating to a location on the playing surface (e.g., Team A has made 30% of three-pointers from a particular area on a basketball court), advertisements, score bugs, graphics, and the like” [Chang ¶ 0313]).
Regarding claim 22, the combination of Chang, Qi, Spearman, and Fernandez teaches the limitations of parent claim 1 and the plurality of training graph-based representations being defined by at least nodes and edges (“...We define the node label as ytv and corresponding feature vectors for node and edge are referred to ftv , fte at time t, respectively" [Qi Integrating Temporal Factors page 6]; "an object node vi ∈ V (i = 1,...,K) refers to the person-level proposal (where i = 1,...,K corresponds to the totally K persons in the scene), and the edge E corresponds to the spatial configuration of object nodes V in the frame...fvi is the feature of the i -th node, and feij is the feature of the edge connecting the i -th node and j -th node, which is the unified bounding box over two nodes. We compute feij using six features via calculating the basic distances and direction vectors" [Qi Graph Inference pages 4-5]; Qi discloses connecting object nodes V, wherein each object node corresponds to a person/player, with edges E, wherein features of an edge connecting two nodes can be, e.g., calculated basic distances).
Claim 23 is rejected under 35 U.S.C. 103 as being unpatentable over the combination of Chang, Qi, Spearman, and Fernandez, as applied to claim 1 above, further in view of Rowlinson (“mplsoccer.pitch module” available Wayback Machine 1 Jun 2020, and “Basics” available Wayback Machine 1 Jun 2020), hereinafter Rowlinson, and Lucey et al., (“Assessing Team Strategy using Spatiotemporal Data”, available 2013), hereinafter Lucey.
Regarding claim 23, the combination of Chang, Qi, Spearman, and Fernandez teaches the limitations of parent claim 1, and generating a team identity map by creating a two-dimensional weighted probability distribution, (“On the top left corner of both columns, we show a 2D plot with the difference between the league mean passing selection heatmap, and each team’s mean passing selection heatmap, when the ball is within the green circle area” [Fernandez page 14 Team-Based Passing Selection Tendencies]) and generating a game identity map (see Fig. 6 – “A game-state representation of a real game situation in soccer” [Fernandez page 15]).
However, the combination does not explicitly teach wherein the two-dimensional probability distribution may be created by fitting a two-dimensional mesh grid over a dimension of a pitch.
In the same field of endeavor, Rowlinson teaches a method of modeling sports tracking data wherein the two-dimensional probability distribution may be created by fitting a two-dimensional mesh grid over a dimension of a pitch (“mplsoccer.pitch is a python module for plotting soccer / football pitches in Matplotlib” [Rowlinson page 1]; “mplsoccer supports 7 pitch types by specifying the pitch_type argument: ‘statsbomb’, ‘opta’,‘tracab’, ‘stats’, ‘wyscout’, ‘statsperform’, and ‘metricasports’. If you are using tracking data (‘metricasports’ or ‘tracab’), you also need to specify the pitch_length and pitch_width , which are typically 105 and 68 respectively” [Rowlinson page 16 Supported data providers]; “By default mplsoccer turns off the axis (border), ticks, and labels. You can use them by setting the axis, label, and tick arguments” [Rowlinson page 30 Axis]; The mplsoccer module can be used to plot tracking data (e.g., probability distribution) over the dimensions (e.g., pitch_length and pitch_width) of a pitch, wherein the x and y axes of the plot form a two-dimensional coordinate grid).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have incorporated wherein the two-dimensional probability distribution may be created by fitting a two-dimensional mesh grid over a dimension of a pitch as taught by Rowlinson into the combination of Chang, Qi, Spearman, and Fernandez because they are all directed towards modeling sports tracking data. Given that Fernandez already teaches methods of visualizing game data using 2D plots representing the field (i.e., soccer pitch), incorporating the teachings of Rowlinson would provide a pre-built means for generating said plots by using an already existing module in the Matplotlib python library [Rowlinson page 1].
However, the combination of Chang, Qi, Spearman, Fernandez, and Rowlinson does not explicitly teach generating the disruption map by subtracting the game identity map from the team identity map.
In the same field of endeavor, Lucey teaches a method of modeling and predicting spatiotemporal events within sports events using tracking data (“By way of example, we present an approach which uses an entire season of ball tracking data from the English Premier League (2010-2011 season) to reinforce the common held belief that teams should aim to “win home games and draw away ones". We do this by: i) forming a representation of team behavior by chunking the incoming spatiotemporal signal into a series of quantized bins, and ii) generate an expectation model of team behavior based on a code-book of past performances” [Lucey Abstract]) that generat[es] the disruption map by subtracting the game identity map from the team identity map (“We also show that our approach can flag anomalous team behavior which has many potential applications” [Lucey Abstract]; “Even though not excessive, this drop in performance suggests there is a change in the spatial behavior between home and away performances. To explore this aspect further, we visualized the difference in occupancy between the home and away performances. To do this, we simply subtracted the home occupancy maps from the away maps and divided by the away occupancy. The difference maps for all twenty teams is given in Figure 7 and it makes for compelling viewing…. However, through the use of spatiotemporal data, we can provide evidence of behavioral differences which can aid in the analysis of performance and decision making. This approach can also be used to flag and predict individual team performances, and in the next section we show methods in which these can be applied” [Lucey page 6 Comparing Home vs Away Behavior]; “Having an measure which could indicate how variable a team's performance is would be quite useful. Given they have a feature representation of each of the previous performances of a team, our approach could be a method of determining the performance variance. To do this, the distance in feature space between each of the past performances, y, and the mean, ^y, can be calculated where the mean is
PNG
media_image1.png
68
137
media_image1.png
Greyscale
and where M is the number of previous performances. A distance measure such as the L2 norm could be used given that the input space has been scaled appropriately (as is the case in our work), which generates the distance measure via
PNG
media_image2.png
36
187
media_image2.png
Greyscale
where m refers to the game of interest… In terms of post-match analysis, a similar approach could be used to see if a team’s performance was within the expectation range. A good example was Fulham’s away performance against Manchester United. In this match, they lost 2-0 and conceded both goals in the first half (12th and 32nd minute). As can be seen by comparing both occupancy maps, in their match against Manchester United they occupied a lot more possession in the middle of the field then normal” [Lucey pages 6-7 Pre/Post Game Analysis]; Occupancy maps can be generated modeling team behavior across different sets of games (e.g., home games vs away games). and can be compared via calculating “difference maps” (i.e., subtracting maps). This technique can be further applied to compare average team behavior across multiple games (i.e., team identity map) to behavior in an individual game (i.e., game identity map) by calculating absolute distance between them in order to identify anomalous (i.e., disruptive) team behavior for the game of interest).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have incorporated generating the disruption map by subtracting the game identity map from the team identity map as taught by Lucey into the combination of Chang, Qi, Spearman, Fernandez, and Rowlinson because it is similarly directed towards modeling and predicting spatiotemporal events within sports events using tracking data. Incorporating the teachings of Lucey would enable further use cases (e.g., identifying variability of a team’s performance) of the behavior map outputted by the combination, providing additional useful information for pre or post-match analysis (“Given a coach or analyst is preparing for an upcoming match, having a measure of how variable a team’s performance is would be quite beneficial. For example, the coach or analyst may have viewed a previous match and formed a qualitative model based on their expert observation. However, this model is only formed by a single observation and may be subject to over-fitting. Having an measure which could indicate how variable a team’s performance is would be quite useful… In terms of post-match analysis, a similar approach could be used to see if a team’s performance was within the expectation range (i.e. ±σ)” [Lucey pages 6-7 Pre/Post Game Analysis]).
Response to Arguments
The remarks filed 11/18/2025 and 12/02/2025 have been fully considered.
Applicant’s remarks (see Remarks filed 11/18/2025 [pages 11-16]) traversing the non-eligible subject matter rejections under 35 U.S.C. 101 set forth in the office action mailed 09/30/2025, in view of claims 1-14, 16, and 19-23 as amended, have been considered and are persuasive.
In concordance with the discussion of previous 35 U.S.C. 101 issues held in the telephonic interview conducted 11/03/2025 (see Examiner Interview Summary Record mailed 11/06/2025) and portions of applicant’s remarks (see Remarks filed 11/18/2025 [page 14, para. 2 – page 16, para. 1]), the examiner further acknowledges that the amended claims now recite a specific, technical procedure that leverages advantageous properties of graph neural networks to improve existing techniques of modeling and analyzing sports tracking data, and thereby are no longer directed towards an abstract idea.
Consequently, the previous rejections under 35 U.S.C. 101 are withdrawn.
Applicant’s remarks (see Remarks filed 11/18/2025 [pages 16-19] and Remarks filed 12/02/2025 [pages 10-12]) traversing the obviousness rejections under 35 U.S.C. 103 set forth in the office action mailed 09/30/2025, in view of claims 1-14, 16, and 19-23 as amended, have been considered but are not persuasive.
Applicant alleges that reference Qi, in addition to all other references cited in the prior rejection of record, does not disclose the limitation wherein a first node of the plurality of nodes has at least one arrow directed towards a second node of the plurality of nodes, and wherein a feature of the arrow indicates a likelihood of a player corresponding to the second node becoming the pass receiver within the possession.
The examiner respectfully disagrees, and has found the cited references, when considered in combination, to still teach or suggest the limitations at issue. Applicant is directed towards the grounds of rejection under 35 U.S.C. 103 with respect to amended claims 1-14, 16, and 19-23 set forth above.
Applicant has not presented further arguments with respect to the dependent claims. As such, amended claims 1-14, 16, and 19-23 stand rejected under 35 U.S.C. 103.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to VIJAY M BALAKRISHNAN whose telephone number is (571) 272-0455. The examiner can normally be reached 10am-5pm EST Mon-Thurs.
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, JENNIFER WELCH can be reached on (571) 272-7212. 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.
/V.M.B./
Examiner, Art Unit 2143
/JENNIFER N WELCH/
Supervisory Patent Examiner, Art Unit 2143