Pub Date : 1900-01-01DOI: 10.54364/aaiml.2022.1130
Edgar Daniel Rodriguez Velasquez, O. Kosheleva, V. Kreinovich
In many application areas, there are effective empirical formulas that need explanation. In this paper, we focus on two such challenges: neural networks, where a so-called softplus activation function is known to be very efficient, and pavement engineering, where there are empirical formulas describing the dependence of the pavement strength on the properties of the underlying soil. We show that similar scale-invariance ideas can explain both types of formulas – and, in the case of pavement engineering, invariance ideas can lead to a new formula that combines the advantages of several known ones.
{"title":"Invariance-Based Approach Explains Empirical Formulas from Pavement Engineering to Deep Learning","authors":"Edgar Daniel Rodriguez Velasquez, O. Kosheleva, V. Kreinovich","doi":"10.54364/aaiml.2022.1130","DOIUrl":"https://doi.org/10.54364/aaiml.2022.1130","url":null,"abstract":"In many application areas, there are effective empirical formulas that need explanation. In this paper, we focus on two such challenges: neural networks, where a so-called softplus activation function is known to be very efficient, and pavement engineering, where there are empirical formulas describing the dependence of the pavement strength on the properties of the underlying soil. We show that similar scale-invariance ideas can explain both types of formulas – and, in the case of pavement engineering, invariance ideas can lead to a new formula that combines the advantages of several known ones.","PeriodicalId":373878,"journal":{"name":"Adv. Artif. Intell. Mach. Learn.","volume":"189 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121081092","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.54364/aaiml.2021.1106
Javier Viaña, Stephan Ralescu, Kelly Cohen, V. Kreinovich, A. Ralescu
{"title":"Why Cauchy Membership Functions: Efficiency","authors":"Javier Viaña, Stephan Ralescu, Kelly Cohen, V. Kreinovich, A. Ralescu","doi":"10.54364/aaiml.2021.1106","DOIUrl":"https://doi.org/10.54364/aaiml.2021.1106","url":null,"abstract":"","PeriodicalId":373878,"journal":{"name":"Adv. Artif. Intell. Mach. Learn.","volume":"96 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125298040","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.54364/aaiml.2022.1123
Supun Sudaraka Manathunga, Ishanya I. Abeyagunawardena, Raahya Lafir, S. Dharmaratne
Background: The magnitude of the impact of COVID-19 is dependent on social, demographic, health, nutrition and even environmental factors. These factors act individually and synergistically to impact the incidence, mortality and morbidity of COVID-19. We aimed to evaluate the variables contributing individually to COVID-19 incidence and mortality utilizing techniques to minimize the effects of interaction between these factors. Method: Data regarding 88 variables for 195 countries over three years were extracted from The Health Nutrition and Population Statistics database and aggregated into a consolidated median. Outliers were eliminated and variables having a completeness of more than 70% were selected. The analysis was done separately for the incidence and mortality of COVID19. Principal component Analysis (PCA) and Elastic net regression were used to identify the most important single variables. The significant variables of the PCA which explained the most variance were identified. Subsequently, variables with the highest importance (using normalized ranked regression coefficients) in the Elastic Net model were selected and the intersecting set of variables common to both models was considered as predictors affecting incidence and mortality of COVID-19. Result: The study revealed communities with a high prevalence of anaemia has a negative correlation with COVID-19 incidence which was furthermore, interestingly seen in multiple age groups. Diphtheria, Tetanus and Pertussis (DTP) Immunization in children was also found to have a negative linear correlation. Conclusion: A negative individual association was seen between anaemia (in multiple age groups) and DTP immunization in children with the incidence and mortality of COVID 19.
{"title":"Association of Social, Demographic, Health, Nutritional and Environmental Factors With the Incidence and Death Rates of COVID-19; a Global Cross-Sectional Analytical Study","authors":"Supun Sudaraka Manathunga, Ishanya I. Abeyagunawardena, Raahya Lafir, S. Dharmaratne","doi":"10.54364/aaiml.2022.1123","DOIUrl":"https://doi.org/10.54364/aaiml.2022.1123","url":null,"abstract":"Background: The magnitude of the impact of COVID-19 is dependent on social, demographic, health, nutrition and even environmental factors. These factors act individually and synergistically to impact the incidence, mortality and morbidity of COVID-19. We aimed to evaluate the variables contributing individually to COVID-19 incidence and mortality utilizing techniques to minimize the effects of interaction between these factors. Method: Data regarding 88 variables for 195 countries over three years were extracted from The Health Nutrition and Population Statistics database and aggregated into a consolidated median. Outliers were eliminated and variables having a completeness of more than 70% were selected. The analysis was done separately for the incidence and mortality of COVID19. Principal component Analysis (PCA) and Elastic net regression were used to identify the most important single variables. The significant variables of the PCA which explained the most variance were identified. Subsequently, variables with the highest importance (using normalized ranked regression coefficients) in the Elastic Net model were selected and the intersecting set of variables common to both models was considered as predictors affecting incidence and mortality of COVID-19. Result: The study revealed communities with a high prevalence of anaemia has a negative correlation with COVID-19 incidence which was furthermore, interestingly seen in multiple age groups. Diphtheria, Tetanus and Pertussis (DTP) Immunization in children was also found to have a negative linear correlation. Conclusion: A negative individual association was seen between anaemia (in multiple age groups) and DTP immunization in children with the incidence and mortality of COVID 19.","PeriodicalId":373878,"journal":{"name":"Adv. Artif. Intell. Mach. Learn.","volume":"46 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125280211","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.54364/aaiml.2022.1133
G. Sirbiladze, Teimuraz Mandjaparashvili, B. Midodashvili, B. Ghvaberidze, David Mikadze
Expert knowledge representations often fail to determine compatibility levels on all objects, and these levels are represented for a certain sampling of universe. The samplings for the fuzzy terms of the linguistic variable, whose compatibility functions are aggregated according to a certain problem, may also be different. In such a case, neither L.A. Zadeh’s analysis of fuzzy sets and even the dual forms of developing today R.R. Yager’s q-rung orthopair fuzzy sets cannot provide the necessary aggregations. This fact, as a given, can be considered as a source of new types of information, in order to obtain different levels of compatibility according to Zadeh, presented throughout the universe. This source of information can be represented as a pair ⟨A, fA⟩, where there is some crisp subset of the universe A that determines the sampling of objects from the universe, and a function fA determines the compatibility levels of the elements of that sampling. It is a notion of split fuzzy set, constructed in this article, that allows for the semantic representation and aggregation of such information. This notion is again and again based on the notion of Zadeh fuzzy set. In particular, the operation of splitting a crisp subset into dual fuzzy sets is introduced. Definitions of set operations on split dual fuzzy-sets are presented in the paper. The proofs are also presented that follow naturally from definitions and previous results. An example of MADM is presented for illustration of the application of splitting operation.
{"title":"Representation of a Crisp Set as a Pair of Dual Fuzzy Sets","authors":"G. Sirbiladze, Teimuraz Mandjaparashvili, B. Midodashvili, B. Ghvaberidze, David Mikadze","doi":"10.54364/aaiml.2022.1133","DOIUrl":"https://doi.org/10.54364/aaiml.2022.1133","url":null,"abstract":"Expert knowledge representations often fail to determine compatibility levels on all objects, and these levels are represented for a certain sampling of universe. The samplings for the fuzzy terms of the linguistic variable, whose compatibility functions are aggregated according to a certain problem, may also be different. In such a case, neither L.A. Zadeh’s analysis of fuzzy sets and even the dual forms of developing today R.R. Yager’s q-rung orthopair fuzzy sets cannot provide the necessary aggregations. This fact, as a given, can be considered as a source of new types of information, in order to obtain different levels of compatibility according to Zadeh, presented throughout the universe. This source of information can be represented as a pair ⟨A, fA⟩, where there is some crisp subset of the universe A that determines the sampling of objects from the universe, and a function fA determines the compatibility levels of the elements of that sampling. It is a notion of split fuzzy set, constructed in this article, that allows for the semantic representation and aggregation of such information. This notion is again and again based on the notion of Zadeh fuzzy set. In particular, the operation of splitting a crisp subset into dual fuzzy sets is introduced. Definitions of set operations on split dual fuzzy-sets are presented in the paper. The proofs are also presented that follow naturally from definitions and previous results. An example of MADM is presented for illustration of the application of splitting operation.","PeriodicalId":373878,"journal":{"name":"Adv. Artif. Intell. Mach. Learn.","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129135088","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.54364/aaiml.2022.1128
A. Aminpour, Mehran Ebrahimi, E. Widjaja
Focal cortical dysplasia (FCD) is one of the most common lesions responsible for drug-resistant epilepsy, and is frequently missed by visual inspection. FCD may be amenable to surgical resection to achieve seizure freedom. By improving lesion detection the surgical outcome of these patients can be improved. Image processing techniques are a potential tool to improve the detection of FCD prior to epilepsy surgery. In this research, we propose and compare the performance of two type of models, Fully Convolutional Network (FCN) and a multi-sequence FCN to classify and segment FCD in children with drug-resistant epilepsy. This experiment utilized the volumetric T1-weighted, T2 weighted and FLAIR sequences. The whole slice FCN models were applied to each sequence separately while the multi-sequence model leverages combined information of all three sequences simultaneously. A leave-one-subject-out technique was utilized to train and evaluate the models. We evaluated subjectwise sensitivity and specificity, which corresponds to the ability of the model to classify those with or without a lesion. We also evaluated lesional sensitivity and specificity, which expresses the ability of the model to segment the lesion and the dice coefficient to evaluate lesion coverage. Our data consisted of 80 FCD subjects (56 MR-positive and 24 MR-negative) and 15 healthy controls. Performance of whole slice FCN was best on T1-weighted, followed by T2-weighted and lowest with FLAIR sequences. Multi-sequence model performed better than the T1 whole slice FCN, and detected 98% vs. 93% respectively MR-positive cases, and 92% vs. 88% respectively MR-negative cases, as well as achieved lesion coverage of 74% vs. 67% respectively for MR-positive cases and 68% vs. 64% for MR negative cases. The dice coefficient for the multi-sequence model was 57% and for whole slice FCN was 56% for MR-positive cases. In the test cohort of six new cases, the multi-sequence model detected 4 out of 6 cases where the predicted lesion had 56% overlap with the actual lesion. This work showed that deep learning methods in particular fully convolutional networks are a promising tool for classification and segmentation of FCD. Additional work is required to further improve lesion classification and segmentation, particularly for small lesions, as well as to train and test optimal algorithms on a larger multi-center dataset.
{"title":"Lesion Segmentation in Paediatric Epilepsy Utilizing Deep Learning Approaches","authors":"A. Aminpour, Mehran Ebrahimi, E. Widjaja","doi":"10.54364/aaiml.2022.1128","DOIUrl":"https://doi.org/10.54364/aaiml.2022.1128","url":null,"abstract":"Focal cortical dysplasia (FCD) is one of the most common lesions responsible for drug-resistant epilepsy, and is frequently missed by visual inspection. FCD may be amenable to surgical resection to achieve seizure freedom. By improving lesion detection the surgical outcome of these patients can be improved. Image processing techniques are a potential tool to improve the detection of FCD prior to epilepsy surgery. In this research, we propose and compare the performance of two type of models, Fully Convolutional Network (FCN) and a multi-sequence FCN to classify and segment FCD in children with drug-resistant epilepsy. This experiment utilized the volumetric T1-weighted, T2 weighted and FLAIR sequences. The whole slice FCN models were applied to each sequence separately while the multi-sequence model leverages combined information of all three sequences simultaneously. A leave-one-subject-out technique was utilized to train and evaluate the models. We evaluated subjectwise sensitivity and specificity, which corresponds to the ability of the model to classify those with or without a lesion. We also evaluated lesional sensitivity and specificity, which expresses the ability of the model to segment the lesion and the dice coefficient to evaluate lesion coverage. Our data consisted of 80 FCD subjects (56 MR-positive and 24 MR-negative) and 15 healthy controls. Performance of whole slice FCN was best on T1-weighted, followed by T2-weighted and lowest with FLAIR sequences. Multi-sequence model performed better than the T1 whole slice FCN, and detected 98% vs. 93% respectively MR-positive cases, and 92% vs. 88% respectively MR-negative cases, as well as achieved lesion coverage of 74% vs. 67% respectively for MR-positive cases and 68% vs. 64% for MR negative cases. The dice coefficient for the multi-sequence model was 57% and for whole slice FCN was 56% for MR-positive cases. In the test cohort of six new cases, the multi-sequence model detected 4 out of 6 cases where the predicted lesion had 56% overlap with the actual lesion. This work showed that deep learning methods in particular fully convolutional networks are a promising tool for classification and segmentation of FCD. Additional work is required to further improve lesion classification and segmentation, particularly for small lesions, as well as to train and test optimal algorithms on a larger multi-center dataset.","PeriodicalId":373878,"journal":{"name":"Adv. Artif. Intell. Mach. Learn.","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114568918","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.54364/aaiml.2022.1137
Seth Reine, Holden Archer, Ahmed Alshaikhsalama, J. Wells, Ajay Kohli, L. Vazquez, A. Hummer, M. Difranco, R. Ljuhar, Yin Xi, A. Chhabra
Background: Hip dysplasia (HD) causes accelerated osteoarthrosis of the acetabulum and is diagnosed through radiographic evaluation. An artificial intelligence (AI) program capable of measuring the necessary anatomical landmarks relevant to HD could reduce resource utilization, increase standardized HD screenings, and form HD outcome models. The study’s aim was to evaluate the relationship between AI measurements of dysplastic hips on initial presentation and changes in patient-reported outcome measures following surgical intervention for HD. Methods: One hundred nine patients with HD and planned surgical intervention obtained preoperative anterior-posterior pelvic radiographs which were measured by the HIPPO AI for lateral center edge angle, Tönnis angle, Sharp angle, Caput-Collum-Diaphyseal angle, femoral coverage, femoral extrusion, and pelvic obliquity. Patients completed a preoperative survey containing the 12-Item Short Form, EuroQol Visual Analog Scale (EQVAS), International Hip Outcome Tool (iHOT-12), Harris Hip Score, and Visual Analog Pain Scales. Patients were recommended to follow up at four months and one year to complete the same survey. Changes in outcome measures were evaluated with paired t-tests for each follow-up interval. Partial Spearman Rank-order correlations were performed between radiographic measures and changes in outcome measures at each follow-up interval controlling for age, BMI, and follow-up time. Results: Patients had significant improvement in all outcome measures at four months (N=46, pvalues<0.05) and one year (N=49,p-values<0.001), except one-year EQVAS (p-value=0.090). Significant positive correlation of moderate strength existed between the Sharp angle and iHOT-12 at four months postoperatively (r𝑠=0.472,p-value=0.044). No other significant correlations were found at either follow-up interval between HIPPO measures and outcome measures. Conclusion: Correlations between deep learning radiographic measurements of dysplastic hips and improvements in postoperative outcomes as evaluated by outcome measures lacked any significant relationships in this study. Physicians treating HD patients can augment care with AI tools but outcomes are likely more multi-factorial and require multi-disciplinary patient care.
{"title":"Deep Learning-Generated Radiographic Hip Dysplasia Parameters: Relationship to Postoperative Patient-Reported Outcome Measures","authors":"Seth Reine, Holden Archer, Ahmed Alshaikhsalama, J. Wells, Ajay Kohli, L. Vazquez, A. Hummer, M. Difranco, R. Ljuhar, Yin Xi, A. Chhabra","doi":"10.54364/aaiml.2022.1137","DOIUrl":"https://doi.org/10.54364/aaiml.2022.1137","url":null,"abstract":"Background: Hip dysplasia (HD) causes accelerated osteoarthrosis of the acetabulum and is diagnosed through radiographic evaluation. An artificial intelligence (AI) program capable of measuring the necessary anatomical landmarks relevant to HD could reduce resource utilization, increase standardized HD screenings, and form HD outcome models. The study’s aim was to evaluate the relationship between AI measurements of dysplastic hips on initial presentation and changes in patient-reported outcome measures following surgical intervention for HD. Methods: One hundred nine patients with HD and planned surgical intervention obtained preoperative anterior-posterior pelvic radiographs which were measured by the HIPPO AI for lateral center edge angle, Tönnis angle, Sharp angle, Caput-Collum-Diaphyseal angle, femoral coverage, femoral extrusion, and pelvic obliquity. Patients completed a preoperative survey containing the 12-Item Short Form, EuroQol Visual Analog Scale (EQVAS), International Hip Outcome Tool (iHOT-12), Harris Hip Score, and Visual Analog Pain Scales. Patients were recommended to follow up at four months and one year to complete the same survey. Changes in outcome measures were evaluated with paired t-tests for each follow-up interval. Partial Spearman Rank-order correlations were performed between radiographic measures and changes in outcome measures at each follow-up interval controlling for age, BMI, and follow-up time. Results: Patients had significant improvement in all outcome measures at four months (N=46, pvalues<0.05) and one year (N=49,p-values<0.001), except one-year EQVAS (p-value=0.090). Significant positive correlation of moderate strength existed between the Sharp angle and iHOT-12 at four months postoperatively (r𝑠=0.472,p-value=0.044). No other significant correlations were found at either follow-up interval between HIPPO measures and outcome measures. Conclusion: Correlations between deep learning radiographic measurements of dysplastic hips and improvements in postoperative outcomes as evaluated by outcome measures lacked any significant relationships in this study. Physicians treating HD patients can augment care with AI tools but outcomes are likely more multi-factorial and require multi-disciplinary patient care.","PeriodicalId":373878,"journal":{"name":"Adv. Artif. Intell. Mach. Learn.","volume":"51 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114918845","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.54364/aaiml.2023.1150
Christopher Robinson, Joshua Lancaster
In this paper we present an algorithm, the Goal Agnostic Planner (GAP), which combines elements of Reinforcement Learning (RL) and Markov Decision Processes (MDPs) into an elegant, effective system for learning to solve sequential problems. The GAP algorithm does not require the design of either an explicit world model or a reward function to drive policy determination, and is capable of operating on both MDP and RL domain problems. The construction of the GAP lends itself to several analytic guarantees such as policy optimality, exponential goal achievement rates, reciprocal learning rates, measurable robustness to error, and explicit convergence conditions for abstracted states. Empirical results confirm these predictions, demonstrate effectiveness over a wide range of domains, and show that the GAP algorithm performance is an order of magnitude faster than standard reinforcement learning and produces plans of equal quality to MDPs, without requiring design of reward functions.
{"title":"Goal Agnostic Learning and Planning without Reward Functions","authors":"Christopher Robinson, Joshua Lancaster","doi":"10.54364/aaiml.2023.1150","DOIUrl":"https://doi.org/10.54364/aaiml.2023.1150","url":null,"abstract":"In this paper we present an algorithm, the Goal Agnostic Planner (GAP), which combines elements of Reinforcement Learning (RL) and Markov Decision Processes (MDPs) into an elegant, effective system for learning to solve sequential problems. The GAP algorithm does not require the design of either an explicit world model or a reward function to drive policy determination, and is capable of operating on both MDP and RL domain problems. The construction of the GAP lends itself to several analytic guarantees such as policy optimality, exponential goal achievement rates, reciprocal learning rates, measurable robustness to error, and explicit convergence conditions for abstracted states. Empirical results confirm these predictions, demonstrate effectiveness over a wide range of domains, and show that the GAP algorithm performance is an order of magnitude faster than standard reinforcement learning and produces plans of equal quality to MDPs, without requiring design of reward functions.","PeriodicalId":373878,"journal":{"name":"Adv. Artif. Intell. Mach. Learn.","volume":"115 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122004503","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 1900-01-01DOI: 10.54364/aaiml.2022.1117
R. Sear, R. Leahy, N. J. Restrepo, Y. Lupu, N. Johnson
Not only can online hate content spread easily between social media platforms, but its focus can also evolve over time. Machine learning and other artificial intelligence (AI) tools could play a key role in helping human moderators understand how such hate topics are evolving online. Latent Dirichlet Allocation (LDA) has been shown to be able to identify hate topics from a corpus of text associated with online communities that promote hate. However, applying LDA to each day’s data is impractical since the inferred topic list from the optimization can change abruptly from day to day, even though the underlying text and hence topics do not typically change this quickly. Hence, LDA is not well suited to capture the way in which hate topics evolve and morph. Here we solve this problem by showing that a dynamic version of LDA can help capture this evolution of topics surrounding online hate. Specifically, we show how standard and dynamical LDA models can be used in conjunction to analyze the topics over time emerging from extremist communities across multiple moderated and unmoderated social media platforms. Our dataset comprises material that we have gathered from hate-related communities on Facebook, Telegram, and Gab during the time period January-April 2021. We demonstrate the ability of dynamic LDA to shed light on how hate groups use different platforms in order to propagate their cause and interests across the online multiverse of social media platforms.
{"title":"Dynamic Latent Dirichlet Allocation Tracks Evolution of Online Hate Topics","authors":"R. Sear, R. Leahy, N. J. Restrepo, Y. Lupu, N. Johnson","doi":"10.54364/aaiml.2022.1117","DOIUrl":"https://doi.org/10.54364/aaiml.2022.1117","url":null,"abstract":"Not only can online hate content spread easily between social media platforms, but its focus can also evolve over time. Machine learning and other artificial intelligence (AI) tools could play a key role in helping human moderators understand how such hate topics are evolving online. Latent Dirichlet Allocation (LDA) has been shown to be able to identify hate topics from a corpus of text associated with online communities that promote hate. However, applying LDA to each day’s data is impractical since the inferred topic list from the optimization can change abruptly from day to day, even though the underlying text and hence topics do not typically change this quickly. Hence, LDA is not well suited to capture the way in which hate topics evolve and morph. Here we solve this problem by showing that a dynamic version of LDA can help capture this evolution of topics surrounding online hate. Specifically, we show how standard and dynamical LDA models can be used in conjunction to analyze the topics over time emerging from extremist communities across multiple moderated and unmoderated social media platforms. Our dataset comprises material that we have gathered from hate-related communities on Facebook, Telegram, and Gab during the time period January-April 2021. We demonstrate the ability of dynamic LDA to shed light on how hate groups use different platforms in order to propagate their cause and interests across the online multiverse of social media platforms.","PeriodicalId":373878,"journal":{"name":"Adv. Artif. Intell. Mach. Learn.","volume":"17 4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133376872","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：