A sports event commentary real-time generation method

Abstract

The present application relates to the field of intelligent content generation, in particular to a sports event commentary real-time generation method, comprising constructing an event commentary trigger monitoring sequence, generating a commentary candidate semantic unit set, screening a target commentary semantic based on a commentary continuity constraint, determining a commentary trigger time and generating a commentary instruction, generating a commentary text based on a commentary language organization strategy, outputting and dynamically updating the commentary content, through constructing the commentary trigger monitoring sequence, introducing an instant semantic activation amount judgment mechanism and a commentary continuity maintenance mechanism, accurate identification of key changes in the event and continuity control of the commentary content are realized, so as to generate event commentary content in line with real-time commentary habits.

Description

Technical Field

[0001] This invention relates to the field of intelligent content generation, specifically to a method for real-time generation of sports event commentary. Background Technology

[0002] With the rapid development of live sports broadcasts, online streaming, and intelligent media platforms, higher demands are placed on the real-time performance, continuity, and accuracy of sports commentary. Current sports commentary methods primarily rely on human commentary, generating content based on fixed event triggering rules and directly generating commentary text by analyzing the event footage within a single time window or based on a single state change. This approach suffers from the following problems:

[0003] (1) The commentary triggering mechanism is simple and it is difficult to accurately depict the key changes in the event. Existing technologies mostly rely on a single event or a simple threshold as the commentary triggering condition. They fail to comprehensively analyze the changes in the participants, behaviors, and the state of the event process. This can easily lead to the commentary missing key moments or frequently triggering the commentary at non-key moments, affecting the accuracy of the commentary and the viewing experience.

[0004] (2) Lack of continuous control of explanation, the explanation is prone to fragmentation or repetition. Existing automatic explanation methods usually treat each detected event as an independent explanation generation unit, lacking the constraint of the continuity of adjacent explanation content in time and semantics, which easily causes the explanation content to break, repeat or become unbalanced, making it difficult to form a continuous explanation output that conforms to real explanation habits. Summary of the Invention

[0005] To address the above issues and overcome the problems of monotone commentary triggering mechanisms and insufficient commentary continuity in existing technologies, this invention provides a real-time sports commentary generation method. By constructing a commentary triggering monitoring sequence, introducing an instant semantic activation determination mechanism, and a commentary continuity maintenance mechanism, it achieves accurate identification of key changes in the event and continuous control of the commentary content, thereby generating event commentary content that conforms to real-time commentary habits.

[0006] The technical solution adopted by this invention is as follows: This invention provides a method for real-time generation of sports event commentary, which includes the following steps:

[0007] Step S1: Construct a commentary trigger monitoring sequence, obtain real-time event data streams of sports events, including event video data, time information, and event progress status information, set commentary trigger monitoring rules to continuously monitor the event data stream, and generate corresponding commentary trigger monitoring sequences;

[0008] Step S2: Generate a set of candidate semantic units for commentary, set a trigger frequency threshold for the trigger monitoring sequence, and when the trigger frequency in the commentary trigger monitoring sequence exceeds the trigger frequency threshold, analyze the event data within the corresponding time window and construct candidate semantic units for commentary, including event participant identification information, behavior change indication information, status change indication information, and time identification information associated with the event timeline.

[0009] Step S3: Filter target interpretive semantics based on interpretive continuity constraints. Introduce an interpretive continuity maintenance mechanism to the set of candidate interpretive semantic units, filter the candidate interpretive semantic units, and obtain the target interpretive semantic units.

[0010] Step S4: Determine the commentary trigger time and generate commentary instructions. Based on the target commentary semantic unit and combined with the current event time information, determine the corresponding commentary trigger time and generate commentary instructions at the corresponding commentary trigger time, including the commentary start identifier, the corresponding target commentary semantic unit identifier, and the commentary output duration or priority information.

[0011] Step S5: Generate commentary text based on commentary language organization strategy. Preset sports event commentary language organization strategy, organize the target commentary semantic units according to the commentary instructions, and generate commentary text output that conforms to real-time commentary habits;

[0012] Step S6: Output and dynamically update the narration content. Output the narration text that conforms to the real-time narration habits in the form of audio and text in real time. During the output process, continuously update the trigger monitoring sequence and the set of narration candidate semantic units.

[0013] Furthermore, step S2 specifically includes the following steps:

[0014] Step S21: Determine the semantic analysis time anchor point. Based on the commentary-triggered monitoring sequence, parse the state change index and map the state change index to the event timeline corresponding to the event data stream to obtain the semantic analysis time anchor point. The semantic analysis time anchor point is determined by the following relationship:

[0015] ;

[0016] in, This indicates the index number of the state change recorded in the monitoring sequence that triggered the explanation. As a temporal anchor point for semantic analysis, The event timeline is generated from official event timing information and is an index-time mapping function provided by the event data stream.

[0017] Step S22: Construct a time-coupled data window. Centered on the semantic analysis time anchor point, extract event data from consecutive time periods before and after the event data stream to construct a time-coupled data window, as shown below:

[0018] ;

[0019] in, , The time span parameter is determined by the historical distribution statistics of similar indices in the explanatory trigger monitoring sequence. Represents a time-coupled data window. For time Corresponding event data;

[0020] Step S23: Generate instantaneous semantic activation values. Perform state change analysis on the event data within the time-coupled data window, calculate the change measures of event participants, behavioral states, and event progress states respectively, and obtain the instantaneous semantic activation values. The formula used is as follows:

[0021] ;

[0022] in, Represents a single semantic analysis time anchor. Real-time semantic activation volume, Measure the change in the identifier of participants within the window. Measured for changes in behavioral state. To measure changes in the status of the competition process, For the weighting coefficients, satisfying ;

[0023] Step S24: Determine the instantaneous semantic activation criteria. Compare the instantaneous semantic activation volume corresponding to the current semantic analysis time anchor point with the statistical distribution of the instantaneous semantic activation volume generated within the same event phase. When the following relationship is satisfied, the event state change corresponding to the time anchor point is determined to have commentary generation value. The relationship is expressed by the following formula:

[0024] ;

[0025] in, , These represent the mean and standard deviation of real-time semantic activations during the current stage of the competition. This is an adjustment factor that automatically adjusts according to changes in the pace of the competition;

[0026] Step S25: Construct candidate semantic units for interpretation and form a set. When the semantic analysis time anchor meets the instant semantic activation judgment condition, construct candidate semantic units for interpretation bound to the time anchor to form a set of candidate semantic units for interpretation.

[0027] Furthermore, step S3 specifically includes the following steps:

[0028] Step S31: Calculate the continuity association weight. Based on the set of interpretive candidate semantic units, calculate the continuity association weight for each interpretive candidate semantic unit, as shown below:

[0029] ;

[0030] in, Representing candidate semantic units A set of candidate units that are temporally close. This represents the candidate semantic units in the set of candidate semantic units for explanation. for Semantic analysis time anchor points, For time decay parameters, This is a semantic similarity function that calculates the multidimensional similarity of participants, their behavioral states, and the state of the event process. , These are similarity indicators for the participants, their behavioral status, and the status of the event's progress. These are the weighting coefficients;

[0031] Step S32: Integrate dynamic semantic scores, combining immediate semantic activation and continuous association weights to calculate the comprehensive score of candidate semantic units. The formula used is as follows:

[0032] ;

[0033] in, This is the continuity enhancement coefficient;

[0034] Step S33: Filter target explanatory semantic units. Sort the candidate semantic unit set according to the comprehensive score, set an adaptive filtering threshold, and select units that meet the adaptive filtering threshold to form the target explanatory semantic unit set. , means as follows:

[0035] ;

[0036] in, This represents the set of semantic units used to interpret the target. For adaptive filtering thresholds, and These are the mean and standard deviation of the overall score for the current stage of the competition, respectively. It is a regulating factor.

[0037] The beneficial effects achieved by the present invention using the above solution are as follows:

[0038] (1) To address the problem of difficulty in accurately depicting key changes in the competition, a semantic analysis time anchor and an instant semantic activation quantity judgment mechanism are introduced into the competition data stream to conduct multi-dimensional joint quantitative analysis of changes in the competition participants, changes in behavior status, and changes in the competition process status. Only when the instant semantic activation quantity exceeds the adaptive judgment condition is a candidate semantic unit for commentary generated. This effectively avoids the problem of commentary omissions or frequent triggering of irrelevant commentary caused by relying solely on a single event trigger, and improves the accuracy and relevance of commentary triggering.

[0039] (2) To address the problem of fragmentation or repetition in the commentary, a continuous association weight calculation and comprehensive scoring mechanism is introduced for the candidate semantic units of the commentary. During the commentary generation stage, the candidate semantic units that are close in time and related in meaning are subject to continuous constraints and screening. This ensures that the final generated commentary semantic units are consistent in time and meaning, avoiding the problem of fragmentation or repetitive output of the commentary content, thereby significantly improving the coherence of the automatic commentary and the overall viewing experience. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating a method for real-time generation of sports event commentary provided by the present invention.

[0041] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0042] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0043] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0044] Example 1, see Figure 1 The present invention provides a method for real-time generation of sports event commentary, the method comprising the following steps:

[0045] Step S1: Construct a commentary trigger monitoring sequence, obtain real-time event data streams of sports events, including event video data, time information, and event progress status information, set commentary trigger monitoring rules to continuously monitor the event data stream, and generate corresponding commentary trigger monitoring sequences;

[0046] Step S2: Generate a set of candidate semantic units for commentary, set a trigger frequency threshold for the trigger monitoring sequence, and when the trigger frequency in the commentary trigger monitoring sequence exceeds the trigger frequency threshold, analyze the event data within the corresponding time window and construct candidate semantic units for commentary, including event participant identification information, behavior change indication information, status change indication information, and time identification information associated with the event timeline.

[0047] Step S3: Filter target interpretive semantics based on interpretive continuity constraints. Introduce an interpretive continuity maintenance mechanism to the set of candidate interpretive semantic units, filter the candidate interpretive semantic units, and obtain the target interpretive semantic units.

[0048] Step S4: Determine the commentary trigger time and generate commentary instructions. Based on the target commentary semantic unit and combined with the current event time information, determine the corresponding commentary trigger time and generate commentary instructions at the corresponding commentary trigger time, including the commentary start identifier, the corresponding target commentary semantic unit identifier, and the commentary output duration or priority information.

[0049] Step S5: Generate commentary text based on commentary language organization strategy. Preset sports event commentary language organization strategy, organize the target commentary semantic units according to the commentary instructions, and generate commentary text output that conforms to real-time commentary habits;

[0050] Step S6: Output and dynamically update the narration content. Output the narration text that conforms to the real-time narration habits in the form of audio and text in real time. During the output process, continuously update the trigger monitoring sequence and the set of narration candidate semantic units.

[0051] Example 2, based on the above example, specifically includes the following steps in step S2:

[0052] Step S21: Determine the semantic analysis time anchor point. Based on the commentary-triggered monitoring sequence, parse the state change index and map the state change index to the event timeline corresponding to the event data stream to obtain the semantic analysis time anchor point. The semantic analysis time anchor point is determined by the following relationship:

[0053] ;

[0054] in, This indicates the index number of the state change recorded in the monitoring sequence that triggered the explanation. As a temporal anchor point for semantic analysis, The event timeline is generated from official event timing information and is an index-time mapping function provided by the event data stream.

[0055] Step S22: Construct a time-coupled data window. Centered on the semantic analysis time anchor point, extract event data from consecutive time periods before and after the event data stream to construct a time-coupled data window, as shown below:

[0056] ;

[0057] in, , The time span parameter is determined by the historical distribution statistics of similar indices in the explanatory trigger monitoring sequence. Represents a time-coupled data window. For time Corresponding event data;

[0058] Step S23: Generate instantaneous semantic activation values. Perform state change analysis on the event data within the time-coupled data window, calculate the change measures of event participants, behavioral states, and event progress states respectively, and obtain the instantaneous semantic activation values. The formula used is as follows:

[0059] ;

[0060] in, Represents a single semantic analysis time anchor. Real-time semantic activation volume, Measure the change in the identifier of participants within the window. Measured for changes in behavioral state. To measure changes in the status of the competition process, For the weighting coefficients, satisfying ;

[0061] Step S24: Determine the instantaneous semantic activation criteria. Compare the instantaneous semantic activation volume corresponding to the current semantic analysis time anchor point with the statistical distribution of the instantaneous semantic activation volume generated within the same event phase. When the following relationship is satisfied, the event state change corresponding to the time anchor point is determined to have commentary generation value. The relationship is expressed by the following formula:

[0062] ;

[0063] in, , These represent the mean and standard deviation of real-time semantic activations during the current stage of the competition. This is an adjustment factor that automatically adjusts according to changes in the pace of the competition;

[0064] Step S25: Construct candidate semantic units for interpretation and form a set. When the semantic analysis time anchor meets the instant semantic activation judgment condition, construct candidate semantic units for interpretation bound to the time anchor to form a set of candidate semantic units for interpretation.

[0065] In this embodiment, taking a standard 90-minute football match as an example, the match data stream is output in real time by the official match system, including match timing information, player identification, ball possession, behavioral event records, and match stage status information. When the match reaches the 63rd minute, the commentary trigger monitoring module detects a significant status change index. This index corresponds to a composite status change of "the ball possession party changes and is accompanied by a change in the direction of attack". The status change index number is recorded as k = 1287 in the commentary trigger monitoring sequence.

[0066] Call the pre-established index-time mapping function in the event data stream Mapping the state change index k=1287 to the official event timeline yields the corresponding event time as 63:14. Based on this, the semantic analysis time anchor is determined as follows:

[0067] ;

[0068] Based on the statistical results of the index of historical similar state changes in the explanatory trigger monitoring sequence, it was found that such "attack-defense transition" events are usually accompanied by continuous behavioral changes within a few seconds before and after the trigger point;

[0069] For index k=1287, adaptively determine the time span parameter:

[0070] Forward time span: Second;

[0071] Backward time span: Second;

[0072] Centered on the semantic analysis time anchor point of 63:14, continuous event data within the time interval of 63:10 to 63:20 is extracted from the event data stream to form a time-coupled data window. This includes: player position changes, passing behavior records, ball possession status, and game phase markers;

[0073] Perform state change analysis on the event data within the time-coupled data window, and calculate three types of change metrics:

[0074] 1. Measure of change in the participants in the game ΔO(t_a): When the ball possession is detected to have been transferred from the defending defender to the midfielder and then quickly passed to the forward within the window, involving continuous changes in multiple key player identifiers, ΔO(t_a) takes the higher value;

[0075] 2. Measurement of behavioral state change ΔA(t_a): When the behavior type rapidly changes from "defensive recovery" to "vertical advancement and passing" and "dribbling breakthrough", the behavioral state undergoes a significant leap, and ΔA(t_a) increases significantly;

[0076] 3. Measurement of change in the state of the game: ΔP(t_a) is at a moderate level as the overall stage of the game is still in the middle of the second half, with no change in score or stage of the game.

[0077] According to weighting coefficients , , The three types of change measures are weighted and fused to calculate the instantaneous semantic activation corresponding to the semantic analysis time anchor point. ;

[0078] Further calculate the instantaneous semantic activation amount. Compare with the statistical distribution of instantaneous semantic activations generated within the same event phase;

[0079] During this phase of the competition, the average value of instantaneous semantic activations is maintained in real time. and standard deviation The adjustment factor is dynamically adjusted according to changes in the pace of the game. ;

[0080] Upon comparison, the immediate semantic activation value corresponding to the current time anchor point satisfies:

[0081] ;

[0082] Determining the change in the state of the event corresponding to this time anchor point has high value for generating commentary.

[0083] Once the semantic analysis time anchor point meets the instant semantic activation judgment condition, the time anchor point is bound to its corresponding event state change feature to construct a commentary candidate semantic unit.

[0084] The candidate semantic unit for commentary includes: time anchor information, change characteristics of participant identifiers, evolution characteristics of behavioral states, and contextual information of the event process. This semantic unit is added to the set of candidate semantic units for commentary and is used for subsequent continuous correlation analysis and commentary content generation.

[0085] Example 3, based on the above examples, specifically includes the following steps in step S3:

[0086] Step S31: Calculate the continuity association weight. Based on the set of interpretive candidate semantic units, calculate the continuity association weight for each interpretive candidate semantic unit, as shown below:

[0087] ;

[0088] in, Representing candidate semantic units A set of candidate units that are temporally close. This represents the candidate semantic units in the set of candidate semantic units for explanation. for Semantic analysis time anchor points, For time decay parameters, This is a semantic similarity function that calculates the multidimensional similarity of participants, their behavioral states, and the state of the event process. , These are similarity indicators for the participants, their behavioral status, and the status of the event's progress. These are the weighting coefficients;

[0089] Step S32: Integrate dynamic semantic scores, combining immediate semantic activation and continuous association weights to calculate the comprehensive score of candidate semantic units. The formula used is as follows:

[0090] ;

[0091] in, This is the continuity enhancement coefficient;

[0092] Step S33: Filter target explanatory semantic units. Sort the candidate semantic unit set according to the comprehensive score, set an adaptive filtering threshold, and select units that meet the adaptive filtering threshold to form the target explanatory semantic unit set. , means as follows:

[0093] ;

[0094] in, This represents the set of semantic units used to interpret the target. For adaptive filtering thresholds, and These are the mean and standard deviation of the overall score for the current stage of the competition, respectively. It is a regulating factor.

[0095] In this embodiment, the code used is as follows:

[0096] import numpy as np

[0097] def semantic_selection(U, t_a, eta_ta, sim_O, sim_A, sim_P,

[0098] w_O=0.4, w_A=0.3, w_P=0.3,

[0099] tau=2.0, gamma=0.5, lambda_psi=0.8, time_window=3.0):

[0100] """

[0101] U: The set of candidate semantic units for interpretation, in the form of a list [u_1, u_2, ..., u_n]

[0102] t_a: The semantic analysis time anchor for each semantic unit, dict {u_i: t_a(u_i)}

[0103] eta_ta: The instantaneous semantic activation η(t_a) at the current moment.

[0104] sim_O, sim_A, sim_P: Semantic similarity matrices, dict {(u_i, u_j): value}

[0105] """

[0106] C = {} # Continuity association weight

[0107] Psi = {} # Overall Rating

[0108] # ===== S31: Calculate the weight of continuous associations =====

[0109] for u_i in U:

[0110] N_i = [u_j for u_j in U

[0111] if u_j != u_i and abs(t_a[u_i] - t_a[u_j]) <= time_window]

[0112] if len(N_i) == 0:

[0113] C[u_i] = 0.0

[0114] continue

[0115] weight_sum = 0.0

[0116] for u_j in N_i:

[0117] delta_t = abs(t_a[u_i] - t_a[u_j])

[0118] temporal_decay = np.exp(-delta_t / tau)

[0119] semantic_similarity = (

[0120] w_O * sim_O[(u_i, u_j)] +

[0121] w_A * sim_A[(u_i, u_j)] +

[0122] w_P * sim_P[(u_i, u_j)] )

[0124] weight_sum += temporal_decay * semantic_similarity

[0125] C[u_i] = weight_sum / len(N_i)

[0126] # ===== S32: Fusion of dynamic semantic scores =====

[0127] for u_i in U:

[0128] Psi[u_i] = eta_ta * (1.0 + gamma * C[u_i])

[0129] # ===== S33: Adaptive threshold screening =====

[0130] psi_values = np.array(list(Psi.values()))

[0131] mu_psi = np.mean(psi_values)

[0132] sigma_psi = np.std(psi_values)

[0133] theta_psi = mu_psi + lambda_psi * sigma_psi

[0134] U_star = [u_i for u_i in U if Psi[u_i] > theta_psi]

[0135] return U_star, Psi, C。

[0136] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0137] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0138] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for real-time generation of sports event commentary, characterized in that: This invention provides a method for real-time generation of sports event commentary, which includes the following steps: Step S1: Construct a commentary trigger monitoring sequence, obtain real-time event data streams of sports events, including event video data, time information, and event progress status information, set commentary trigger monitoring rules to continuously monitor the event data stream, and generate corresponding commentary trigger monitoring sequences; Step S2: Generate a set of candidate semantic units for commentary, set a trigger frequency threshold for the trigger monitoring sequence, and when the trigger frequency in the commentary trigger monitoring sequence exceeds the trigger frequency threshold, analyze the event data within the corresponding time window and construct candidate semantic units for commentary, including event participant identification information, behavior change indication information, status change indication information, and time identification information associated with the event timeline. Step S3: Filter target interpretive semantics based on interpretive continuity constraints. Introduce an interpretive continuity maintenance mechanism to the set of candidate interpretive semantic units, filter the candidate interpretive semantic units, and obtain the target interpretive semantic units. Step S4: Determine the commentary trigger time and generate commentary instructions. Based on the target commentary semantic unit and combined with the current event time information, determine the corresponding commentary trigger time and generate commentary instructions at the corresponding commentary trigger time, including the commentary start identifier, the corresponding target commentary semantic unit identifier, and the commentary output duration or priority information. Step S5: Generate commentary text based on commentary language organization strategy. Preset sports event commentary language organization strategy, organize the target commentary semantic units according to the commentary instructions, and generate commentary text output that conforms to real-time commentary habits; Step S6: Output and dynamically update the narration content. Output the narration text that conforms to the real-time narration habits in the form of audio and text in real time. During the output process, continuously update the trigger monitoring sequence and the set of narration candidate semantic units.

2. The method for real-time generation of sports event commentary according to claim 1, characterized in that: Step S2 specifically includes the following steps: Step S21: Determine the semantic analysis time anchor point. Based on the commentary-triggered monitoring sequence, parse the state change index and map the state change index to the event timeline corresponding to the event data stream to obtain the semantic analysis time anchor point. The semantic analysis time anchor point is determined by the following relationship: ； in, This indicates the index number of the state change recorded in the monitoring sequence that triggers the explanation. As a temporal anchor point for semantic analysis, The event timeline is generated from official event timing information and is an index-time mapping function provided by the event data stream. Step S22: Construct a time-coupled data window. Centered on the semantic analysis time anchor point, extract event data from consecutive time periods before and after the event data stream to construct a time-coupled data window, as shown below: ； in, , The time span parameter is determined by the historical distribution statistics of similar indices in the explanatory trigger monitoring sequence. Represents a time-coupled data window. For time Corresponding event data; Step S23: Generate instantaneous semantic activation values. Perform state change analysis on the event data within the time-coupled data window, calculate the change measures of event participants, behavioral states, and event progress states respectively, and obtain the instantaneous semantic activation values. The formula used is as follows: ； in, Represents a single semantic analysis time anchor. Real-time semantic activation volume, Measure the change in the identifier of participants within the window. Measured for changes in behavioral state. To measure changes in the status of the competition process, Let be the weighting coefficient, satisfying ; Step S24: Determine the instantaneous semantic activation criteria. Compare the instantaneous semantic activation volume corresponding to the current semantic analysis time anchor point with the statistical distribution of the instantaneous semantic activation volume generated within the same event phase. When the following relationship is satisfied, the event state change corresponding to the time anchor point is determined to have commentary generation value. The relationship is expressed by the following formula: ； in, , These represent the mean and standard deviation of real-time semantic activations during the current stage of the competition. This is an adjustment factor that automatically adjusts according to changes in the pace of the competition; Step S25: Construct candidate semantic units for interpretation and form a set. When the semantic analysis time anchor meets the instant semantic activation judgment condition, construct candidate semantic units for interpretation bound to the time anchor to form a set of candidate semantic units for interpretation.

3. The method for real-time generation of sports event commentary according to claim 1, characterized in that: Step S3 specifically includes the following steps: Step S31: Calculate the continuity association weight. Based on the set of interpretive candidate semantic units, calculate the continuity association weight for each interpretive candidate semantic unit, as shown below: ； in, Representing candidate semantic units A set of candidate units that are temporally close. This represents the candidate semantic units in the set of candidate semantic units for explanation. for Semantic analysis time anchor points, For time decay parameters, This is a semantic similarity function that calculates the multidimensional similarity of participants, their behavioral states, and the state of the event process. , These are similarity indicators for the participants, their behavioral status, and the status of the event's progress. These are the weighting coefficients; Step S32: Integrate dynamic semantic scores, combining immediate semantic activation and continuous association weights to calculate the comprehensive score of candidate semantic units. The formula used is as follows: ； in, This is the continuity enhancement coefficient; Step S33: Filter target explanatory semantic units. Sort the candidate semantic unit set according to the comprehensive score, set an adaptive filtering threshold, and select units that meet the adaptive filtering threshold to form the target explanatory semantic unit set. , means as follows: ； in, This represents the set of semantic units used to interpret the target. For adaptive filtering thresholds, and These are the mean and standard deviation of the overall score for the current stage of the competition, respectively. It is a regulating factor.

Images