Topic modelling

Topic modelling is a form of text mining where a computer program is used to trace word clustering and word co-occurrence within a corpus. These word clusters are considered topics, and the model works on the assumption that each document is made up of a series of topics, each with differing proportions and distributions. However, whilst the algorithm produces the topic model itself, it is the role of the human analyst to assign social meaning to those clusters, and infer what the topic may be. My review of current literature in the field suggests that this is done on the basis of mere eyeballing, whereas my analysis and subsequent discussion throughout this thesis suggests that a much more systematic closer reading of documents would be more beneficial. Not only would this result in the much more accurate labelling of topics, but it would inevitably make topic modelling a much more inviting method to humanist researchers.

Customer experience is of critical importance to the sustained growth of a business. It is necessary to ensure great customer experience, thus maintaining customer satisfaction and loyalty. An approach that intended to develop and improve customer experience is called Customer Experience Management (CEM).  CEM is a strategy used to track, oversee, and organize all interactions to help a business focus on the needs of its customers. This research uses sentiment analysis and topic modeling to analyze the experience of Internet Service Provider customer. The output of this research expected to drive the strategies change in CEM. This research uses data taken from customer tweets on Twitter. It is considering that the data on social media is enormous and unstructured. Therefore, classification using Naive Bayes Classifier applied to assist and expedite in the sentiment analysis process. The classification for sentiment analysis using NBC gained accuracy above 82%. Hence, the classification models using NBC achieve excellent capability for sentiment analysis. To determining topics that often discussed by customers, this research uses the Latent Dirichlet Allocation models for Topic Modeling.

In this article we show how fashion brands communicate with their follower on Instagram. We use  a  continuously  update  dataset  of  68  brands,  more  than  300,000  posts  and  more  than 40,000,000 comments. Starting  with  descriptive  statistics,  we  uncover  different  behavior  and  success  of  the  various brands.  It  turns  out  that  there  are  patterns  specific  to  luxury,  mass-market  and  sportswear brands.  Posting  volume  is  extremely  brand  dependent  as  is  the  number  of comments  and  the engagement of the community.Having  understood  the  statistics,  we  turn  to  machine  learning  techniques  to  measure  the response of the community via comments. Topic models help us understand the structure of their respective community and uncover insights regarding the response to campaigns.Having up-to-date content is essential for this kind of analysis, as the market is highly volatile. Furthermore, automatic data analysis is crucial to measure the success of campaigns and adjust them accordingly for maximum effect.

Social networks analysis studies the interactions among users when using social media. The content provided by social media is composed of essentially two parts: a network structure of users' links (e.g. followers, friends, etc.) and actual data content exchanged among users (e.g. text, multimedia). Topic modeling and sentiment analysis are two techniques that help extracting meaningful information from large or multiple portions of the text: identifying the topic discussed in a text, and providing a value characterizing an opinion respectively. This extracted information can then be combined to the network structure of users' links for further tasks as predictive analytics, pattern recognition, etc. In this paper we propose a method based on graph databases, topic modelling and sentiment analysis to facilitate pattern extraction within social media texts. We applied our model to Twitter datasets, and were able to extract a series of opinion patterns.

Cuando comenzó la pandemia de COVID 19, las plataformas sociales tuvieron un rol central en la producción y acceso a la  información. Este estudio identifica los tópicos de mayor interés y sus sentimientos asociados en Twitter en la conversación pública en lengua española en ese periodo. Asimismo, analiza el rol de Twitter como plataforma social involucrada en la conversación pública,  como medio para la autocomunicación de masas y para amplificar la voz de un conjunto reducido de actores de alta visibilidad. Se recolectaron 231.375 tuits en España y América Latina durante dos meses. Se midieron indicadores de frecuencia y sentimientos, y se agruparon términos para identificar tópicos y determinar el interés de los usuarios sobre estos, mediante métodos digitales y de lenguaje computacional en R. La frecuencia de los principales términos es dinámica a lo largo del período estudiado, lo que sugiere diferentes percepciones de la pandemia. Los tópicos principales refieren a conversaciones en torno a la cantidad de casos, muertos y contagiados, con prevalencia de sentimientos negativos. La muestra analizada corresponde a mensajes generados por usuarios comunes en su gran mayoría, pero una parte de ella ha sido amplificada a gran escala mediante retuits y marcas de favoritos.

Cultural products are a source to acquire individual values and behaviours. Therefore, the differences in the content of the magazines aimed specifically at women or men are a means to create and reproduce gender stereotypes. In this study, we compare the content of a women-oriented magazine with that of a men-oriented one, both produced by the same editorial group, over a decade (2008-2018). With Topic Modelling techniques we identify the main themes discussed in the magazines and quantify how much the presence of these topics differs between magazines over time. Then, we performed a word-frequency analysis to validate this methodology and extend the analysis to other subjects that did not emerge automatically. Our results show that the frequency of appearance of the topics Family, Business and Women as sex objects, present an initial bias that tends to disappear over time. Conversely, in Fashion and Science topics, the initial differences between both magazines are maintained. Bes...

Patterns for the internet of things (IoT) which represent proven solutions used to solve design problems in the IoT are numerous. Similar to objectoriented design patterns, these IoT patterns contain multiple mutual heterogeneous relationships. However, these pattern relationships are hidden and virtually unidentified in most documents. In this paper, we use machine learning techniques to automatically mine knowledge graphs to map these relationships between several IoT patterns. The end result is a semantic knowledge graph database which outlines patterns as vertices and their relations as edges. We have identified four main relationships between the IoT patterns-a pattern is similar to another pattern if it addresses the same use case problem, a large-scale pattern uses a small-scale pattern in a lower level layer, a large pattern is composed of multiple smaller scale patterns underneath it, and patterns complement and combine with each other to resolve a given use case problem. Our results show some promising prospects towards the use of machine learning techniques to generate an automated repository to organise the IoT patterns, which are usually extracted at various levels of abstraction and granularity.

Understanding the emotions of sharing in social media plays a key role in learning people's thoughts. Knowing the emotion of human beings with developing technology provides benefits in various fields. For example, media, marketing and advertising areas allow people to reflect on their use and idea-specific content. In our study, Latent Dirichlet Allocation (LDA), Latent Semantic Analysis (LSA) and Probabilistic-Latent Semantic Analysis (P-LSA) were used to determine the emotions of individuals from Turkish tweets. In addition, the success of the developed n-stage state of the LDA algorithm in the emotion analysis was compared with the existing methods. The dataset consists of 4000 tweets of 5 different emotions, including angry, fear, happiness, sadness and surprise. All topic modeling methods were modeled for 3 and 5 class datasets and their successes and running times were measured. It has been observed that the developed n-stage LDA method achieves success in terms of running time and performance according to LDA and P-LSA. The most successful and fastest modeled method was LSA.

In the software development process, the requirements traceability in the Software Requirements Specification (SRS) is an important aspect to trace fulfillment of requirement. Generally, the implementation of traceability is done manually, and it requires considerable time and money. Many methods have been proposed to identify traceability automatically, therefore, we conducted a comparative study of several existing methods to find out which method is better. In this study, we made a comparison of 4 methods, namely the Information Retrieval (IR), Ontology, Combination of Information Retrieval with Topic Modeling (LSA or LDA) methods. From the comparison of those methods, we found that the IR method gets the highest accuracy score, but the precision, recall and fl-score are unequal between dataset 1 and dataset 2, while other methods (ontology, IR+LSA, IR+LDA) have more stable metrics. The average score of precision from highest to lowest consecutively is in the IR + LDA method, Ont...

Top line hotels are now shifting into the digital way in how they understand their customers to maintain and ensuring satisfaction. Rather than the conventional way which uses written review or interviews, the hotel is now heavily investing in Artificial Intelligence particularly Machine Learning solutions. Analysis of online customer reviews changes the way companies make decisions in a more effective way than using conventional analysis. The purpose of this research is to measure hotel service quality. The proposed approach emphasizes on service quality dimensions reviews of the top-5 luxury hotel in Indonesia that appear on online travel site TripAdvisor based on section "Best of 2018". In this research, we use a model based on a simple Bayesian classifier to classify each of customer review into one of service quality dimensions. Our model was able to separate each classification properly by accuracy, kappa, recall, precision, and F-measure measurements. To uncover latent topics in the customer opinion we use Topic Modelling. We found that the common issues occur is about responsiveness as it got the lowest percentage compared to others. Our research provides a faster outlook of hotel rank based on service quality to end customer based on a summary of the previous online review.

Hospitality, as one of the sectors in the tourism industry, continues to show positive trend every year based on its revenues. In order to understand public opinion, the legacy method such as interview, questionnaire, and other time-consuming method is still widely used in this sector. However, this method is less efficient compared to data analytics methodology using available data from online source, such as social media Therefore, we need to conduct a research to see a better method in order to understand the public opinion in hospitality sector. There are many methodologies to support content analysis based on unstructured data to understand public opinion. In this research, we use content analysis methodology which consists of Sentiment Analysis, Topic Modeling, and Text Network Analysis to process 483413 obtained tweets from January 7th until March 28th, 2017. The twitter data is gained using Python and R language. The whole process of the research consists of data collection, data preprocessing, and interpreting the results. We use Rstudio, Jupyter Notebook and Gephi softwares during the process. As a case study in the hospitality sector, we use Airbnb as one of the sharing economy business models that allows other people to share their space and begins to be widely used by travelers around the world. Our objective is to understand public opinion towards Airbnb as online-based sharing economy business model which becomes a disruptive innovation for tourism industry. As the result, we are able to apply the combination of content analysis methods to give a good understanding towards the public opinion. Text Network Analysis gives us the ability to summarise large-scale conversation in very fast and real-time fashion. It provides the knowledge by associate most frequent words used in social conversations. Sentiment Analysis shows the feelings and emotion from people about certain topics. Topic Modelling is able to capture popular topics, thus gives comprehensive understanding on how people react towards a new business model

The paper examines the efficiency of topic models as features for computational identification and conceptual analysis of linguistic metaphor on Russian data. We train topic models using three algorithms (LDA and ARTM – sparse and dense) and evaluate their quality. We compute topic vectors for sen-tences of a metaphor-annotated Russian corpus and train several classifiers to identify metaphor with these vectors. We compare the performance of the topic modeling classifiers with other state-of-the-art features (lexical, morphosyntac-tic, semantic coherence, and concreteness-abstractness) and their different com-binations to see how topics contribute to metaphor identification. We show that some of the topics are more frequent in metaphoric contexts while others are more characteristic of non-metaphoric sentences, thus constituting topic predic-tors of metaphoricity, and discuss whether these predictors align with the con-ceptual mappings attested in literature. We also compare the topical heterogene-ity of metaphoric and non-metaphoric contexts in order to test the hypothesis that metaphoric discourse should display greater topical variability due to the presence of Source and Target domains.

The study focuses on the information flow on twitter during the corona virus outbreak. Tweets related to #coronavirus are studied using sentiment analysis and topic modelling using Latent Dirichlet Allocation post preprocessing. The study concluded that the information flow was accurate and reliable related to corona virus outbreak with minimum misinformation. LDA analysis had identified the most relevant and accurate topics related to corona virus outbreak and sentiment analysis confirmed the prevalence of negative sentiments like fear along with positive sentiments like trust. Governments and Healthcare authorities & institutions effectively utilized to spread accurate and reliable information on twitter.

The Topics2Themes tool, which enables text analysis on the output of topic modelling, was originally developed for the English language. In this study, we explored and evaluated adaptations required for applying the tool to Japanese texts. That is, we adapted Topics2Themes to a language that is very different from the one for which the tool was originally developed. To apply Topics2Themes to Japanese texts, in which white space is not used for indicating word boundaries, the texts had to be pre-tokenised and white space inserted to indicate a token segmentation. Topics2Themes was also extended by the addition of word translations and phonetic readings to support users who are second-language speakers of Japanese. To evaluate the adaptation to a second language, as well as the reading support, we applied the tool to a corpus consisting of short Japanese texts. Twelve different topics were automatically identified, and a total of 183 texts representative for the twelve topics were extracted. A learner of Japanese carried out a manual analysis of these representative texts, and identified 35 reoccurring, fine-grained themes.

In this paper we present primary content analysis results for the Greek Parliament data in the context of Natural Language Processing and  Text  Mining  approaches.  The  raw  minutes  of  the Greek  Parliament  plenary  sessions of  the  last  26 years are  processed  and transformed into a structured and machine readable format, and then clustered based on the analysis of their content using topic modelling techniques. Inspired  by  and  following the  work  of Greene  and  Gross  (2017) for  the  European  Parliament,  we  employ  a  two - layer methodology for  applying  topic  modelling  in  a  Non-negative  Matrix  Factorization  framework  to  a  timestamped  corpus  of  political speeches in order to explore dynamic topics. The results are visualized in various ways (by topic, by time) providing at the same time information about the contribution of each Parliament Member, political party and region (constituency) to each topic, and by extent, the ability to explore how the political and policy agenda has been shaped and evolved in Greece over time.

Today, with the increase of Internet-based documents, we are presented with many data that need to be processed and evaluated. Media, news and advertising are some of the areas where these data are evaluated. For the news, the classification of people in the media sector is an important problem in terms of time. In this paper, it is aimed to determine which types of news titles belong to. The dataset consists of 4200 Turkish new titles belonging to 7 class labels. In order to determine the types, classical Latent Dirichlet Allocation (LDA), Latent Semantic Analysis (LSA) and Non-Negative Matrix Factorization (NMF) algorithms were used in topic modeling. In addition, the LDA-based n-LDA method was also used. The accuracy of all methods used was measured and compared. NMF was the most successful method for three classes, while for five and seven classes LSA was the most successful method.

The scenario planning process consists of various steps and the initial stages (typically Framing and Scanning), are those that require much time and efforts, because imply the consultation of experts, reading of documents and review of the scientific literature on the subject dealt with. A new approach to speed the Scanning phase – which implies capturing the topics and identify the relevant key factors – is based on finding and analysing corpus of documents and to study the relevant literature in depth. The exponential increase of data within social media and the web is a valuable source (Kayser and Blind, 2017) that is worth exploiting and this work aims at implementing this process through a combination of text mining (Tong and Zhang, 2016) and spatial data mining techniques. For these reasons, this study will focus on the value of both textual and geographic data in the process of constructing a future scenario, reducing the duration and efforts of the Scanning phase. We applied the method on a dataset of COVID-19 tweets (Uhl et al., 2017) and the application shows how to extract the key factors driving the scenario building and how to extract useful information from the georeferenced data.

La “europeización” de las esferas públicas nacionales es una cuestión de crear un espacio mediático discursivo dentro de la UE. Hay fuerzas de convergencia, como redes en el espacio digital. Al mismo tiempo, hay fuerzas contrarias, un creciente nacionalismo y populistas que critican las élites a “Bruselas”. Aparentemente, la visión de una esfera pública europea comparte su suerte con el proyecto europeo: después de años de crisis, el optimismo se convirtió en desilusión. Utilizando la cobertura mediática de las elecciones de la UE de 2019 en siete países europeos (57.943 artículos de Alemania, Hungría, Italia, Polonia, Portugal, la República Checa y el Reino Unido), dibujamos cuadro de un público heterogéneo de la UE. Resulta claramente que la europeización horizontal y vertical requieren interpretaciones diferenciadas. Un alto grado de europeización horizontal indica una cobertura mediática convergente y pro-europea (como Alemania o Portugal), un alto grado vertical puede indicar un público polarizado o un panorama mediático poco libre (como el Reino Unido o Hungría). En cuanto a metodología, el estudio muestra que una combinación de análisis de contenido computacional y cooperación internacional puede adelantar la investigación en el público europeo.

The scenario planning process consists of various steps and the initial stages (typically Framing and Scanning), are those that require much time and efforts, because imply the consultation of experts, reading of documents and review of the scientific literature on the subject dealt with. A new approach to speed the Scanning phase – which implies capturing the topics and identify the relevant key factors – is based on finding and analysing corpus of documents and to study the relevant literature in depth. The exponential increase of data within social media and the web is a valuable source (Kayser and Blind, 2017) that is worth exploiting and this work aims at implementing this process through a combination of text mining (Tong and Zhang, 2016) and spatial data mining techniques. For these reasons, this study will focus on the value of both textual and geographic data in the process of constructing a future scenario, reducing the duration and efforts of the Scanning phase. We applie...

In this paper we present primary content analysis results for the Greek Parliament data in the context of Natural Language Processing and Text Mining approaches. The raw minutes of the Greek Parliament plenary sessions of the last 26 years are processed and transformed into a structured and machine readable format, and then clustered based on the analysis of their content using topic modelling techniques. Inspired by and following the work of Greene and Gross (2017) for the European Parliament, we employ a two - layer methodology for applying topic modelling in a Non-negative Matrix Factorization framework to a timestamped corpus of political speeches in order to explore dynamic topics. The results are visualized in various ways (by topic, by time) providing at the same time information about the contribution of each Parliament Member, political party and region (constituency) to each topic, and by extent, the ability to explore how the political and policy agenda has been shaped an...

Startups are entrepreneurial organisations that aim to develop a scalable and disruptive business.
However, these small ventures operate in an environment of extreme uncertainty. The startup economy
takes place in the present but is directed towards the future. This chapter critically investigates in online
and offline realms the circulation of imagined futures that create causal links to bridge the gap between the
present economic scenario and potential futures in the Italian startup food economy. This work adopted
a mixed-method approach framed in a qualitative exploratory strategy which was designed to integrate
qualitative techniques and digital methods. This work concludes by highlighting the co-evolutionary
process between online and offline realms. On the one hand, online narratives allow economic actors
to perform in radical uncertain economic contexts, while, on the other hand, the offline practices give
legitimacy and credibility to these potential future scenarios.

Journal editors of all types of scholarly publishing face various issues and choices in order to support the development of the journals they lead and manage. One particular issue is what strategic choices can be made to enhance the visibility, accessibility and impact of research published in their journal as much as possible. Another issue is how to provide potential authors of the journal with useful information that will support them with their publication choice.

The objective of this study is twofold. First, it aims to provide an approach that can be used by journal editors to identify topical trends in scholarly-led publishing in their journal, in order to better inform potential future authors of the journal. In particular, we analyse what factors impact the trends and research cooperations over time with a view to research topics and thematic streams, regions, and the research methods employed by authors in previous publications. Second, this study derives potential factors responsible for specific journal developments, such as the influence of open access principles, guest editors, indexing systems, research cooperations and topics addressed by the journal on the visibility, accessibility and impact of the journal.

To attain the above-mentioned objectives, this study uses data-mining methods to analyse all the articles published in one journal in particular, namely the (diamond(/platinum) open access e-journal for eDemocracy and Open Government called “JeDEM” (see www.jedem.org) that was set up in 2009. The methods used are a combination of data mining (such as the text mining, topic modelling, k-means clustering, social network analysis and community detection) of journal content and metadata with further qualitative interpretation of results from a journal management perspective. The qualitative part confirms or challenges the data analysis part, particularly with view to potential outliers or developments that cannot be explained by quantitative data analysis alone. Combined with internal knowledge from the journal management perspective, we are able to provide an interpretative component and are to relate the trends emerging from the data to strategic decisions or publication details.

The results are as follows. First, our study shows the most prominent research topics of the studied journal, and their evolution over time. Second, the research methods employed by authors publishing in the journal are identified, as well as the research cooperations established through publications in the journal. Third, the crucial factors such as indexing, communication with the community and changes in journal management are derived.

The developed approach was found to be useful to gain insights concerning journal developments, and might also be used by editors of other journals to determine their journal strategy. For example, using data mining methods, editors can analyse whether the topics included in their call for papers match the topics on which the journal publishes or whether it needs adaptations to better manage the expectations of future authors of the journal. Thus, our results might also help authors with their publication choices and tracing the evolutionary stages of the studying scientific problem. Finally, our study derives further research questions aiming at achieving a critical assessment of scholarly developments within the publishing sphere over time.

We have seen that some of social opinion polarization leads to the breakup of the relationship, some in the scale of small communities, but others can divide large organizations or even a nation. The legacy methodology to answer the root cause of opinion polarization in the society commonly use random sampling and questionnaire approach. This approach generally expensive in term of time and money. On the contrary, we have the opportunity to employ big data approach using social media data. Big data methodology provides us the rich source to investigate several questions, such as: how social opinion polarization formed, dynamic social network mechanism during time-windows observation, identification of dominant actors and communities. The power of big data approach lies in the number of data analyzed, where the more data involved in the process, the more accurate to describe the population condition. Today, computing power is no longer become an obstacle to process large volume, fast, and variety data, thus the observation of social opinion polarization process is possible. In this research, we answer the social opinion polarization root cause by using topic modelling methodology. The dynamic social network mechanism is measured using social network properties. Identification of influential actors and communities are provided by social network analysis metric and methodology. By answering three major questions above, we are able to explain of opinion polarization and its growth along the time, both for their topology structure and conversational content. The process knowledge gives us insight to how and when the separation takes place. As a case study, we use two massive adverse political campaign in Indonesia regarding the opinion of pro and contra the incumbent president to continue his presidentship in 2019 presidential election. We benefited of Indonesian habit to produce user-generated content in term of post and conversation in social media. We collect Twitter data with the following information: the observation duration is 10 days from April 27 th to May 2 nd , 2018, data collected based on several opposite hashtags such as #2019GantiPresiden #Jokowi2Periode and many more, we acquire 24097 and 418256 tweets from pro and contra movement. The benefit of this study is to avoid the danger of deepening gap between opponent communities by implementing right strategy to prevent that from happen.