Match data interaction selection method for MOBA game based on storyline visualization

Abstract

The application provides a MOBA game match data interaction selection method based on story line visualization, comprising: (1) match event extraction and data preprocessing; (2) match similarity calculation and match level analysis; (3) story line layout calculation and user round level understanding using a story line visual analysis method; (4) user exploration of event and hero detail information and data selection combined with visual elements. The method provided by the application supports analysis of complex data in a MOBA game, the whole method can efficiently extract key events of a match and analyze multiple matches, and rich visual elements created on this basis can help users visually and intuitively visualize the match process, and various user interactions contained can facilitate users to easily explore detailed information of each hero, thereby improving research efficiency and reducing operation burden.

Description

Technical Field

[0001] This invention relates to the fields of visualization and human-computer interaction, and specifically proposes a method for selecting interactive MOBA game match data based on storyline visualization. Background Technology

[0002] MOBA games are an emerging form of competitive sports, widely engaging in entertainment and social activities. Games like League of Legends, Dota 2, and Honor of Kings have become incredibly popular, attracting a massive player base. Each game is renowned for its deep gameplay mechanics, extensive roster of heroes, and vibrant esports environment. For MOBA game developers, a crucial task is creating adaptive training models tailored to the diverse characteristics of human players. For instance, players of varying skill levels may require different types of agents, such as beginner, intermediate, and advanced. Reinforcement learning algorithms are commonly used for agent training. Deep reinforcement learning combines neural networks with reinforcement learning frameworks, allowing agents to learn optimal actions in complex environments by maximizing cumulative rewards through trial and error. The reward function is critical in reinforcement learning because it provides immediate feedback to the agent, guiding its behavior in the desired direction and measuring the quality of behavioral strategies by quantifying the value of each decision in the environment. However, the long rounds, high task complexity, and diverse strategies inherent in MOBA games make designing comprehensive reward functions challenging. It's difficult to design diverse strategies for a wide range of players, and therefore, existing methods often fall short in developing agents with the diverse strategies required for MOBA game scenarios. Imitation learning offers an alternative approach to addressing the challenges of training deep reinforcement learning models. It focuses on learning and improving from existing policies rather than designing complex reward functions to characterize behavior across different policies. Focusing on selecting representative policy samples simplifies a significant amount of human design and is generally easier than defining complex reward mechanisms. However, a key challenge is obtaining suitable training data. Current methods typically select samples based on broad metrics such as reward value or win rate. However, complex sequential decision data with numerous events cannot be measured and differentiated using only one or two final numerical metrics. Furthermore, relying solely on metrics can lead to biased evaluations that do not necessarily align with user needs. Therefore, visualization analytics methods are needed to integrate human insights into the data selection process. However, due to contextual differences, existing sports and gaming visual analytics techniques are not directly applicable to MOBA games. Current sequential data visualization techniques also struggle to simultaneously represent the evolution of individual characters and overall game dynamics in both temporal and spatial aspects. Therefore, effectively visualizing and selecting MOBA game data remains a significant challenge. Summary of the Invention

[0003] To address the shortcomings and deficiencies of existing technologies, this invention aims to provide a method for selecting MOBA game match data based on storyline visualization. This is a visualization analysis method tailored for MOBA data selection and strategy analysis, designed to help game AI designers effectively collect target samples to train the agent. Guided by user design requirements, this invention develops a data selection visualization analysis method from macro to micro levels. This method first presents an overview of multiple matches, aiming to enable data analysts to efficiently identify matches of interest for more targeted analysis. To further understand the development of events in each match, this invention introduces a novel visualization technique called v-storyline. This technique encodes the positional attributes of characters on the vertical axis of the layout, allowing the abstraction and extraction of spatiotemporal sequence data from match events, effectively illustrating the evolution of match rounds. Furthermore, this method provides detailed hero statistics for comparison, thereby supporting data analysts in selecting data at the fine-grained hero level. It helps users intuitively and conveniently understand match strategy patterns and efficiently select data from multiple stages.

[0004] The method for selecting game data interaction in MOBA games based on storyline visualization includes the following four steps: (1) game event extraction and data preprocessing; (2) game similarity calculation and game level analysis; (3) storyline layout calculation and user understanding of the round level using the visual analysis method of storyline; (4) user exploration of event and hero details and multi-round data selection by combining visualization elements.

[0005] Specifically:

[0006] (1) Game event extraction and data preprocessing:

[0007] To help users efficiently complete data selection tasks, this method first extracts events from a large amount of game data, game by game, and simultaneously preprocesses and integrates some important information for easy use in subsequent visualization. The game event extraction step can be further subdivided in the workflow as follows:

[0008] First, behavioral information is read from the data for each hero individually. Then, based on different evaluation strategies such as development, damage, and kills, an index representing the state is calculated. This is a quantity that changes over time. Using this quantity, a change point detection algorithm is used to divide the entire game round into several time periods. Each time period corresponds to an event. Considering the background knowledge of MOBA games, five types of events are defined: team fights, jungling, pushing towers, movement, and death.

[0009] On the other hand, to analyze the spatial information of events, it is necessary to clearly define the location of the events. In this step, firstly, based on key locations in the MOBA game map (i.e., buildings of both sides, elite monsters in the jungle), a Veno diagram is calculated to divide the scene space. Then, to avoid overly fragmented area division, the areas are merged again based on background knowledge from the MOBA game, retaining a total of 7 areas corresponding to both sides' bases, top lane, mid lane, bottom lane, and top and bottom jungle areas. Finally, based on the time period information and spatial area information after segmentation, the entire game data is converted into storyline data. For a storyline data set, it contains a different number of events per hero. Each event is a triple, consisting of start time, end time, and event number (where the event type and location information are implicitly encoded in the event number).

[0010] The game event extraction and data preprocessing process primarily transforms round data, which contains a large amount of complex information, into concise storyline data for subsequent analysis. This approach has two advantages: First, it allows for efficient understanding of the game, eliminating the need for users to waste time searching frame by frame. Instead, users can efficiently grasp the development and changes in the situation over a longer timeframe. Second, it facilitates analysis and selection. Events provide a macroscopic portrayal of a game's characteristics, helping users choose from a large pool of games.

[0011] (2) Game similarity calculation and game level analysis:

[0012] In this step, the present invention proposes a method for calculating match similarity and designs corresponding visualization elements. First, based on the events extracted in step (1), the distribution of events in a match is statistically analyzed. The horizontal axis of the distribution represents the match time, and the vertical axis represents the number of heroes involved in an event. Since movement events occur frequently and are considered as a guaranteed event, they are not considered in the statistics. Death events, which are accompanied by team fight events, are not separately classified. When depicting a match, considering that the strategies of the winning and losing sides are usually different in the game context, the two sides are statistically analyzed separately in this step. Then, by calculating the Cramer distance between the distributions and using the TSNE algorithm, the points representing the match are projected into a two-dimensional space, thereby obtaining a scatter plot representing the match in a plane.

[0013] The following visual elements and user interactions are defined in the scatter plot. First, the colors represent the opposing factions in a match; the direction of the triangles indicates the win / loss relationship; and the size corresponds to the match duration. Second, the visualization interface of this method provides swipe, zoom, and click interactions. When users use it, they can click on matches of interest and further analyze and select matches using the methods and visualization elements in steps (3) and (4).

[0014] (3) Storyline layout calculation and user understanding of turn-level using visual analysis methods of storylines:

[0015] This step proposes a turn-based event display based on storyline visualization, which can efficiently and intuitively show the development and evolution of a selected game situation. The specific steps of the storyline layout generation algorithm are as follows:

[0016] First, the storyline event data generated in step (1) is read, and then the number of points in the entire layout space is calculated based on the number of key moments, the number of heroes, and the number of regions in the data. For points in the storyline layout, their position on the x-axis represents the time point, and their relative position on the y-axis represents the positional relationship of the regions in the method of this invention. Then, after calculating the number of points in the layout, each point is treated as a variable, and a layout target based on aesthetic principles is defined. In the formula, y represents the vertical position of the layout point, i and j are loop variables, representing the statistics of all points in the layout, c is the sum of the number of heroes and the number of regions, and t is the time of the round data. The first term of the formula represents minimizing the swing of the points before and after, and the second term represents minimizing the blank space in the layout.

[0017]

[0018] Secondly, to introduce the relative positional relationship in the spatial region, the variables corresponding to the storyline characters (i.e., heroes and region boundaries) at each moment are sorted according to the layout number obtained in (1). Next, in order to reduce the overlap of lines within the region, the sorting relationship is adjusted again according to the principle of minimizing intersection. Then, in order to improve the alignment relationship of corresponding points between two consecutive moments to enhance readability, the similarity is calculated according to the formula for the sequences l and r between two consecutive moments, where LCS represents the calculation of the maximum common subsequence of the two sequences, m and n represent the number of story points contained in the two sequences, p is an indicator function with a value of 1 for the storyline points representing the boundary, and α, β, and γ are parameters.

[0019]

[0020] Next, the constraints on the y-values ​​at each time step are obtained based on the alignment relationships. Finally, a mathematical solver is used to solve the constrained quadratic optimization problem, resulting in a series of layout points, where AO represents the alignment order.

[0021]

[0022] Compared to previous methods, the method of this invention introduces relative positional relationships on the y-axis, which helps users better understand events in MOBA scenarios and reduces potential ambiguity caused by the lack of physical encoding of the y-axis in previous methods. At the same time, compared to a completely correspondent mapping of spatial positions, it reduces tediousness and improves readability.

[0023] After completing the layout calculation step, this invention designs the following visualization elements and provides users with various interactions. First, based on the layout points calculated in this section, a continuous storyline is obtained by connecting them using Bézier curves. Each colored line represents a hero, and gray dashed lines mark the area boundaries. The convergence and deepening of lines indicate the occurrence of events, and the movement of lines indicates hero changes in area and entry and exit events. Dashed lines indicate hero deaths. On the other hand, the visualization interface provides users with the opportunity to interact with the storyline. Users can adjust the granularity of storyline events, zoom in on the storyline display by scrolling the timeline, and drag the scroll bar to move the storyline display range. At the same time, each storyline supports floating highlighting and click interaction. Clicking can select a hero and further explore and select them in step (4).

[0024] (4) Users combine visual elements to explore event and hero details and select data:

[0025] Users utilize this invention and the constructed visualization elements and user interactions to explore information and select data from a macro to a micro perspective. The process is as follows:

[0026] First, the given dataset is read and the game projection data obtained in step (2) is displayed to the user. The user can then choose one of the games to explore in depth according to their own needs.

[0027] Then, read the user's interaction, and use the methods and data provided in step (3) to show the evolution and development of this round. The user selects the appropriate storyline granularity and the time period to be explored according to their own needs.

[0028] Next, detailed data is read to provide users with a statistical information view. In this view, the headers display the heroes played and their performance in team fights, jungling, tower pushing, etc. This visualization element includes a user-friendly sorting feature, allowing users to click on the headers to sort the data in descending order.

[0029] Secondly, the entire visualization method provides a hero comparison view. By default (when the user has not made a selection), the visualization elements include data such as the position, health, mana, actions performed, and rewards obtained by both heroes, displayed in the form of snapshots, pie charts, bar charts, and shaded charts. When the user makes a selection, a transition animation changes the displayed data to the selected object in the corresponding faction's visual element area.

[0030] Meanwhile, when users perform interactions such as brushing, dragging, or selecting units to change the range and objects of the displayed data, the visual elements in the corresponding storyline module, comparison view section, heatmap, etc. in the visualization are updated synchronously and the newly selected data is displayed.

[0031] Finally, based on the information provided by the visual elements in the entire process, users select events and heroes of interest to analyze their strategies, collect valuable and noteworthy data as training data, and then select new matches to explore and analyze again. This process in step 4 is repeated multiple times to complete the entire data selection task.

[0032] The method described in this invention supports the analysis of complex data in MOBA games. The entire method can efficiently extract key events in a match and analyze multiple games. Based on this, rich visualization elements help users intuitively visualize the match process, and various user interactions allow users to easily explore detailed information about each hero. Specifically, the effects of this invention include:

[0033] Enhancing users' ability to efficiently extract key events in a game: In step (1), the present invention supports automated data processing, finding key time periods and summarizing them into events, avoiding the time-consuming process of watching video replays and reducing the hassle of users thinking frame by frame.

[0034] Reduce the difficulty for users to select data from multiple levels: In step (2), the present invention combines the process results in step (1), integrates the distance calculation method, and designs visualization elements and user interaction to provide users with a starting point for exploration. Users do not need to conduct aimless sequential analysis and random sampling, which improves the efficiency of exploration and reduces the operational burden.

[0035] Enhancing users' understanding of the game's development: In step (3), this invention uses a new storyline visualization format, which can intuitively and efficiently display the key events obtained in step (1). It can more intuitively display the development of multiple heroes and events in the spatiotemporal dimension, making it easier to analyze changes in the situation and providing users with a variety of interactive features for exploration. It is an important part of realizing hierarchical exploration.

[0036] To alleviate the difficulty for users in comparing hero strategy information: In step (4), the visualization of this invention provides comparisons of hero levels, within factions, and between different factions. At the same time, the user interaction implied in steps (3) and (4) allows users to freely choose the event period and hero content to explore, reducing the time spent by users searching and locating from a large amount of statistical data, and reducing the difficulty for users to compare different hero strategies. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating the selection of MOBA game match data interaction from macro to micro levels in a specific embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram illustrating the steps of game event extraction and data preprocessing in a specific embodiment of the present invention;

[0039] Figure 3 The following are the representational meanings and main visual elements of the storyline method in specific embodiments of the present invention;

[0040] Figure 4 This refers to the specific page involved in the human-computer interaction visual analysis in the specific embodiments of the present invention;

[0041] Figure 5 This is a logic block diagram in a specific embodiment of the present invention. Detailed Implementation

[0042] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0043] Figure 1 This invention illustrates a flowchart of the macro-to-micro MOBA game match data interaction selection process in a specific embodiment: First, data preprocessing and event extraction are required, such as... Figure 2 Then, users will proceed through three progressive exploration stages: game-level analysis, event-level understanding, and hero-level comparison. Figure 1 The document showcases the analytical methods and corresponding visualization elements provided in the workflow. Users progressively complete the interaction according to their needs, selecting a match of interest and understanding the match strategy patterns through the events present in the match, comparing the similarities and differences between different hero strategies. Users continuously cycle through this process, exploring new data and completing the entire MOBA game match data interaction and selection process.

[0044] Figure 2This diagram illustrates the steps of game event extraction and data preprocessing in a specific embodiment of the present invention. The entire event extraction and preprocessing process consists of three main steps, corresponding to the three parts from left to right in the diagram. First, data is extracted from the original data by hero, and the original behavior is represented by calculated state variables based on factors such as development, KDA, and damage. Then, a model-free change point detection algorithm is used to segment the entire game round, resulting in several time segments. Finally, category labels are added to the events, and region representations are added based on the location of occurrence.

[0045] Figure 3 This diagram illustrates the representation and main visual elements of the storyline method in a specific embodiment of the present invention. The horizontal gray dashed lines divide the entire storyline layout into several regions, corresponding to the map space divided using a Veno diagram. Each colored solid line represents a hero, and the convergence of lines corresponds to a specific time and region, indicating that related heroes are jointly involved in an event. Depending on the event, these events can primarily include team fights, jungling, and tower pushing.

[0046] Figure 4 The specific page for the human-computer interaction visual analysis involved in this invention is shown.

[0047] (a) The view is the game mode selection view, which is the first step in each user's exploration process. The visualization elements in it show multiple games and project them into a two-dimensional space based on their characteristics. Users can freely select one or a group of games to explore through interaction.

[0048] (b) is a hero heatmap view, where dark lines represent the spatial division based on the Veno diagram, and darker black lines represent the boundaries of merged regions. The heatmap in the view shows the position distribution of selected heroes within a selected time period, and by default displays the position statistics of the entire faction over the entire round.

[0049] (c) The view is the storyline view. After the layout points are calculated by the v-storyline visualization method introduced in step (3), these layout points are smoothly connected by Bézier curves. This invention provides a variety of interactions such as granularity selection, storyline zoom, time period selection, and storyline click and hover.

[0050] (d) The view is the economic experience curve, which plots the difference between the economic experience curves of the two sides in a game in a discounted form.

[0051] (e) The view is a statistical comparison view that uses bar charts to represent the data for corresponding items in the table, with colors tied to camps. For ease of comparison, the view provides sorting functionality by row and column, supporting sorting data from highest to lowest.

[0052] (f) is the hero comparison view, which divides the time period into ten segments. This view provides detailed match data, including hero actions, spatial distribution, health and mana status, as well as differences between the two teams in resource acquisition and status metrics. This view consists of four parts.

[0053] (f1) shows a snapshot for each time interval. Each snapshot is represented by a gray circle showing the spatial distribution of heroes at a given time. The center and radius of these circles were calculated using a clustering algorithm based on the average hero position.

[0054] (f2) and (f4) display pie charts representing the action choices of the two teams at the same time: white indicates idle time, green indicates movement, pink indicates attack, and blue-purple indicates skill usage. The shaded areas below each pie chart show the fluctuations in health (pink) and mana (purple) over time, allowing users to track team actions and the status of each hero.

[0055] View (f3) uses a bar chart and shaded areas to illustrate the changes in resource acquisition and status metrics between the two teams. The height of the bars represents the total resources acquired by each team, while the shaded areas below show the real-time differences in status metrics. Positive (top) shading indicates that the red team is dominant, while negative (bottom) shading indicates that the blue team is dominant.

[0056] Figure 5 A logical block diagram of the present invention is shown. It corresponds to the proposed analysis steps and the above flowchart, and integrates the visual analysis method to form an interactive page for users. The working logic of this method in terms of process is shown in the figure, mainly divided into three steps, corresponding to the macro-to-micro selection of the game, event understanding, and hero-level comparison.

Claims

1. A method for selecting game data interaction in MOBA games based on storyline visualization, characterized in that, Includes the following steps: (1) Game event extraction and data preprocessing; (2) Calculation of game similarity and analysis of game hierarchy; The specific method is as follows: First, based on the events extracted in step (1), the distribution of events in a game is statistically analyzed; the horizontal axis of the distribution is the game time, and the vertical axis is the number of heroes involved in an event; when describing the game, the two sides are statistically analyzed separately. Then, by calculating the Cramer distance between the distributions and projecting the points representing the game into a two-dimensional space using the TSNE algorithm, a scatter plot representing the game is obtained in the plane. The following visual elements and user interactions are defined in the scatter plot visualization: First, color represents the opposing factions in the game; the direction of the triangles expresses the win / loss relationship in the game; and the size corresponds to the game duration. Secondly, the visualization interface provides swipe, zoom and click interactions; when users use it, they can click on the game they are interested in and use the methods and visualization elements in steps (3) and (4) for further analysis and selection. (3) Storyline layout calculation and user understanding of turn-level using visual analysis methods of storylines; The specific steps of the storyline layout generation algorithm are as follows: First, read the events and preprocessed data extracted in step (1), and then calculate the number of points in the entire layout space based on the number of key moments, the number of heroes, and the number of regions in the data; for points in the storyline layout, their position on the x-axis represents the time point, and their relative position on the y-axis represents the positional relationship of the regions; then, after calculating the number of points in the layout, treat each point as a variable, and then define the layout target according to aesthetic principles. In the formula, y represents the vertical position of the layout point, i and j are loop variables, representing the statistics of all points in the layout, c is the sum of the number of heroes and the number of regions, and t is the time of the round data; the first term of the formula represents minimizing the swing of the points before and after, and the second term represents minimizing the blank space in the layout; Secondly, based on the events and preprocessed data extracted in step (1), the variables corresponding to the storyline characters at each moment are sorted; then, the sorting relationship is adjusted again according to the principle of minimizing crossover; then, for the sequences l and r at two consecutive moments, the similarity is calculated according to the formula, where LCS represents the calculation of the maximum common subsequence of the two sequences, M and N represent the number of story points contained in the two sequences, p is an indicator function with a value of 1 for the storyline points representing the boundary, and α, β, and γ are parameters; Next, the constraints on the y-values ​​at each time step are obtained based on the alignment relationship; finally, the constrained quadratic optimization problem is solved using a mathematical solver to obtain a series of layout points, where AO represents the alignment relationship order. After completing the layout calculation steps, the following visualization elements are designed and multiple user interactions are provided. First, based on the layout points calculated in this section, a coherent storyline is obtained by connecting them using Bézier curves. Each colored line represents a hero, and the gray dashed line marks the boundary of the area. The convergence and deepening of the lines indicate the occurrence of events, and the movement of the lines indicates the hero changing areas and entering and leaving events. The dashed line indicates the hero's death. On the other hand, the visualization process provides users with the function of interacting with the storyline. Users can adjust the granularity of the storyline events, zoom in on the storyline display by brushing the timeline, and drag the scroll bar to move the display range of the storyline. At the same time, each storyline supports floating highlighting and click interaction. Clicking can select a hero and further explore and select in step (4). (4) Users combine visual elements to explore event and hero details and select data in multiple rounds.

2. The method for selecting game data interaction in MOBA games based on storyline visualization according to claim 1, characterized in that, The specific method for step (1) is as follows: First, events are extracted from a large amount of game data, game by game, while important information is preprocessed and integrated for use in subsequent visualization steps. The game event extraction step is broken down into the following steps: First, behavioral information is read from the data for each hero. Then, an index representing the state is calculated based on different aspects of the evaluation strategy. This is a quantity that changes over time. The index representing the state is used to divide the entire game round into several time periods through a change point detection algorithm. Each time period corresponds to an event. Considering the background knowledge of MOBA games, five types of events are defined: team fight, jungling, pushing towers, movement, and death. On the other hand, in order to analyze the spatial information of the event and clarify the location of the event, in this step, we first rely on the key locations in the MOBA game map to calculate the Veno diagram to divide the scene space; then, in order to avoid the area division being too fragmented, we rely on the background knowledge of the MOBA game to merge the areas again, retaining a total of 7 areas corresponding to the two bases, top lane, mid lane, bottom lane and top and bottom jungle areas respectively. Finally, based on the segmented time period information and spatial region information, the game data is integrated into storyline data. For a storyline data set, there are different numbers of events per hero. Each event is a triple, consisting of the start time, end time, and event number.

3. The method for selecting game data interaction in MOBA games based on storyline visualization according to claim 1, characterized in that, The specific method for step (4) is as follows: First, continue reading the given dataset and use the game projection data obtained in step (2) to display it to the user. The user can choose one of the games to explore in depth according to their own needs. Then, continue to read the user's interaction, and use the methods and data provided in step (3) to show the evolution and development of this round. The user selects the appropriate storyline granularity and the time period to be explored according to their own needs. Next, the system continues to read detailed data and provides a user statistics view. In the statistics view, the header displays the heroes in the game and their values ​​in team fights, jungling, and tower pushing. When constructing this visualization element, the system provides user sorting interaction, allowing users to click on the header to sort the data in descending order. Secondly, the entire visualization method provides a hero comparison view; by default, the visualization elements include the positions, health, mana, actions performed, and rewards obtained by both heroes, which are displayed in the form of snapshots, pie charts, bar charts, and shaded charts respectively; when the user makes a selection, the displayed data in the corresponding faction's visual element area is transformed to the object selected by the user through a transition animation; Meanwhile, when the user performs interactions such as swiping, dragging, or selecting units to change the range of displayed data and objects, the corresponding storyline module, comparison view section, and visual elements in the heatmap in the visualization method are updated synchronously and the new selected data is displayed. Finally, based on the information provided by the visual elements in the entire process, users select events and heroes of interest to analyze their strategies, collect valuable and noteworthy data as training data, and then select new matches to explore and analyze again. This process in step 4 is repeated multiple times to complete the entire data selection task.

Images