Reflection #14 – [04-17] – [Jiameng Pu]

  1. Lelkes, Y., Sood, G., & Iyengar, S. (2017). The hostile audience: The Effect of Access to Broadband Internet on Partisan Affect. American Journal of Political Science, 61(1), 5-20

Summary:

Over the past fifty years, partisan animus has been increasing, the reach of partisan information sources has been expanding in the meantime. With increasing polarization on the Internet, this paper talks about identifying the impact of access to broadband Internet on affective polarization by exploiting differences in broadband availability brought about by variation in state right-of-way regulations (ROW). Lelkes et al. measured an increase in partisan hostility by collecting data from multiple resources, e.g., Federal Communication Commission (the data on broadband access), National Annenberg Election Studies(the data on the partisan effect). The author concludes that access to broadband Internet increases partisan hostility but is stable across levels of political interest.

Reflection:

The topic itself is intriguing, and I have had a strong feeling that people become more hostility not only in the Partisan Affect but on diverse topics in the community. People are passionate about educating people who have different opinions and values. One interesting discussion in this paper is to explore the relationship between the number of subscribers and the number of providers at the census tract level or at the zip code level. I liked the process of find convincing proxy for the measure needed in the research, which could potentially introduce more interesting findings. Although many county-level factors and indicators are examined, e.g., unemployment rate, median age, the male-to-female ratio, percent black, another view I come up with is that whether new media apps/websites people mainly use in their daily life have a significant effect on partisan polarization, because it’s intuitive that people’s viewpoints are easily influenced by specific subscribed news feed and people also tend to choose news media share the same viewpoints with them.

Besides partisan affect, this paper reminds me of people’s hostility in different communities and fields, because I feel this is the trend in today’s virtual social space regardless of topics and platforms. Like I said before, people nowadays are passionate about educating people who have different opinions and values with them, which induces unprecedented malicious discussion on the Internet and makes the access to broadband internet itself become a relatively narrow viewpoint to explore.

Read More

Reflection #13 – [04-10] – [Jiameng Pu]

[1] Pryzant, Reid, Young-joo Chung and Dan Jurafsky. “Predicting Sales from the Language of Product Descriptions.” (2017).
[2] Hu, N., Liu, L. & Zhang, J.J. “Do online reviews affect product sales? The role of reviewer characteristics and temporal effects”. Inf Technol Manage (2008) 9: 201. https://doi.org/10.1007/s10799-008-0041-2

Summary:

The first paper posits that textual product descriptions are also important determinants of consumer choice. They mine more than 90,000 product descriptions on the Japanese e-commerce marketplace Rakuten and propose a novel neural network architecture that leverages an adversarial objective to control for confounding factors, and attentional scores over its input to automatically elicit textual features as a domain-specific lexicon. They show that how textual features and word narratives can predict the sales of each product. However, the second paper focuses on online product reviews provided by consumers, such as reviewer quality, reviewer exposure, product coverage, and temporal effects.

Reflection:

I really enjoy the first paper, since it’s based on neural network architecture and I’m a neural-nets person, which means I’d like to try many research topics on neural nets and feel neural nets are like black box but also like legos: researchers can feel free both invent some creative components and build parts together according to the need of your tasks. Plus, product description + neural nets is an interesting direction. One thing I’ve never expected is that they combined feature selection task with prediction task inside the proposed neural nets, which I feel great because people always use neural nets to do a lot of similar things. In the experiment, some of classmates mentioned health products do not have any brand loyalty, which I don’t think is an issue. If you are an experienced patient, you would know brand loyalty always exists in every category… I would suggest another two things for this paper: 1. give more intuition to the design of neural network architecture due to its black-box property; 2. I’m curious about whether and how the technique can be apply to other different languages besides Japanese.

The pair of paper are perfectly related, because most people can empirically feel that two most important factors influencing their purchase are product description and online reviews. Thus the second paper dives into how online reviews are associated with sales. I feel more difficult to read the second paper with five hypothesis and tons of tables, which are pretty old school, but I’m still impressed by its simplicity and practicability. It seems the paper mainly use data from Amazon.com’s Web Service (AWS), thus I’m a little curious about if the dataset can significantly influence the analysis, because I feel different E-commerce websites truly have distinct styles in online review. For example, online review on Chinese website Taobao is more vivid and customer-engaged, e.g., with tons of pictures and customer conversations in the review section. In that case, I guess researchers probably need to reconsider features of online reviews involved in the analysis. Personally, I’m not that sure which recommendation system, i.e.,  yes/no or 1-star to 5-star scale, because sometimes I feel difficult to decide whether to recommend an item with both merit and demerit, that’s where 5-star scale helps for people who struggle to make choices.

Read More

Reflection #12 – 04/05 – [Jiameng Pu]

Using linguistic and topic analysis to classify sub-groups of online depression communities

The world is in the period of high incidence of different kinds of mental problem. The complex nature of depression, e.g., different presentation of depression among depressed people, makes it difficult for the treatment and prevention. Therefore, the paper focuses on online communities to explore depression based on more and more exchange of information, support, and advice online. Machine learning can help identify topics or issues relevant to those with depression and characterize the linguistic styles of depressed individuals. The paper utilizes machine learning techniques on linguistic features to identify sub-communities in the online depression communities.

Reflection

They mentioned “In this work we have used Live Journal data as a single source of online networks. In fact, Live Journal users could also join other networking services, such as Facebook or Twitter”, which exactly corresponds to my concern — the dataset could be better and more up-to-date, like data from popular social media, e.g., reddit, twitter and facebook. In fact, communities like facebook has conducted such research to detect accounts that might be in depression to carry out necessary psychological treatment. However, Live Journal data is obviously old for this task. Another point I feel confused is the five subgroups of online communities were identified: Depression, Bipolar Disorder, Self-Harm, Grief/Bereavement, and Suicide.. I didn’t see the inner logic how these five subgroups can fully represent downhearted psycholinguistic features, for instance, what’s the difference between self-harm and suicide… What this paper impresses me is the comparison of different classifiers, from SVM to Lasso. In practice, I sometimes feel like researchers have to try different machine learning models, and it’s not that we can always correctly guess which would work better, even when you got some priori experience. I haven’t used LIWC so far, but it’s apparently one of the most widely used tools in all the paper we’ve read. Look forward to trying it out in my future research…

Read More

Reflection #10 – [03/22] – [Jiameng Pu]

An Army of Me: Sockpuppets in Online Discussion Communities

Summary:

People interact with each other on the Internet mainly by discussing mechanism provided by social networks such as Facebook, Reddit. However, sockpuppets created by malicious users badly influence the network environment by engaging in undesired behavior like deceiving others or manipulating discussions. Srijan et al. study sockpuppetry across nine discussion communities. By firstly identify sockpuppets using multiple signals indicating accounts might share the same user, they then characterize their behavior by inspecting different aspects. They find that the behavior of sockpuppets is different from that of ordinary users in many ways, e.g., start fewer discussions, write shorter posts, use more personal pronouns such as “I”. The study contributes towards the automatic detection of sockpuppets by presenting a data-driven view of deception in online communities.

Reflection:

For the process of identifying sockpuppets, the strategy is inspired by Wikipedia administrators who identify sockpuppets by finding accounts that make similar edits on the same Wikipedia article in near-similar time and from same IP address, which makes sense. But for the hyper-parameter, top percentage(5%) of most used IP address, is there any better strategy that can decide the percentage more numerically rather than intuitively? When measuring linguistic traits of sockpuppets, LIWC word categories is used to measure the fraction of each type of words written in all posts, and VADER for sentiment of posts. Up to now, I feel LIWC word categories is powerful and heavily used in social science research, I’ve never used VADER before. In the double life experiment, although they match sockpuppets with ordinary users that have similar posting activity, and that participate in similar discussion, I feel like there is too much uncertainty in the linguistic feature of ordinary users, i.e., different users have different writing style. Then the cosine similarity of the feature vectors for each account would be less convincing.

Uncovering Social Spammers: Social Honeypots + Machine Learning

Summary:

Both web-based social networks (e.g., Facebook, MySpace) and online social media sites (e.g., YouTube, Flickr) rely on their users as primary contributors of content, which made them prime targets of social spammers. Social spammers engage themselves in undesirable behavior like phishing attacks, to disseminate malware and commercial spam messages, etc, which will seriously impact the user’s experience. Kyumin et al. propose a honeypot-based approach for uncovering social spammers in online social systems by harvesting deceptive spam profiles from social networking communities and creating spam classifiers to actively filter out existing and new spammers. The machine learning based classifier is able to identify previously unknown spammers with high precision and a low rate of false positives.

Reflection:

The section of machine learning based classifier impressed me a lot, since it shows how to investigate the discrimination power of our individual classification features apart from only evaluating the effectiveness of classifiers, in which ROC curve plays an important role. Also, AMContent, the text-based features modeling user-contributed content in the “About Me” section, shows me how to use more complicated text feature besides simple data like age, marital status, gender. I’ve never heard of Myspace before but there is still twitter experiment, otherwise I would think this is a weird choice of experiment dataset. For twitter spam classification, we can obviously see the differences in the way they collect account feature, i.e., twitter accounts are noted for their short posts, activity-related features, and limited self-reported user demographics. Thus there is a reminder that feature design varies according to the variation of study subjects.

 

Read More

Reflection #9 – [02/22] – [Jiameng Pu]

De Choudhury, Munmun, Michael Gamon, Scott Counts, and Eric Horvitz. “Predicting depression via social media.” ICWSM13 (2013): 1-10.

Summary & Reflection:

Tens of millions of people around the world each year suffer from depression but global provisions and services for identifying, supporting, and treating mental illness of this nature have been considered as insufficient. Since people’s virtual activities on social media can potentially indicate their mental state to some degree, the paper explores the potential to use social media to detect and diagnose the major depressive disorder in individuals. By compiling a set of Twitter users who report being diagnosed with clinical depression and observing their social media postings over a year preceding the onset of depression, they measure and feed behavioral attributes, such as social engagement, emotion, language and linguistic styles, to a statistical classifier that estimates of the risk of depression. Results indicate there are useful signals for characterizing the onset of depression, which can further instrumental in developing practical detection tools for depression.

I’m pretty impressed by using the Amazon’s Mechanical Turk interface to conduct clinical depression survey, which obviously a great advance that can cover more participates. But for the survey design, there is still one common question that whether the participants can remain objective and honest when answering the questionnaire and providing self-reported information since respondents would Because sometimes the respondent would unconsciously or even consciously hide their true situation. Although I was thinking about how to improve it, I did not think of a better way. But What we need to pay special attention to is that the design of the questions, which should appropriately guide the psychological state of the participants. The problem should not be blunt or irritating. For Measuring Depressive Behavior, I’m impressed by some of the measures such as defining the egocentric social graph, but I’m not convinced by some hypothesis like “Individuals with depression condition are likely to use these names in their posts”. In my intuition, I do not think depression patients will positively desire to receive feedback on their effects during the course of treatment. I also strongly feel that one of the most important things in social science research is actually to alleviate biases existing in many places, e.g., the authors in this paper conduct an auxiliary screening test in addition to the CES-D questionnaire to get rid of noisy responses.

Mitra, Tanushree, Scott Counts, and James W. Pennebaker. “Understanding Anti-Vaccination Attitudes in Social Media.” In ICWSM, pp. 269-278. 2016.

Summary & Reflection:

Public health can be threatened by an anti-vaccination movement which reduces the likelihood of disease eradication. Anti-vaccine information can be disseminated on social media like Twitter, thus Twitter data would help understand the drivers of attitudes among participants involved in the vaccination debate. By collecting tweets of users who persistently hold pro and anti-attitudes, and those who newly adopt anti attitudes towards vaccination, they find that those with long-term anti-vaccination attitudes manifest conspiratorial thinking, mistrust in government, and are resolute and in-group focused in language.

By comparing linguistic styles, topics of interest and social characteristics of over 3 million tweets from Twitter, Mitra et al categorize users into 3 group: anti-vaccines, pro-vaccines, and joining-anti vaccine cohort. The data collection process involves 2 main phases where they extracted tweet sample from the Twitter Firehouse stream in the first phase and built a classifier to classify the collected posts as pro-vaccine and anti-vaccine tweets. The MEM model that extracts dimensions along which users express themselves seems pretty interesting and it should be a good tool in other potential areas such as personalized recommendation functionality of the social platform since it can capture clusters of co-occurring words which can identify linguistic dimensions that represent psychologically meaningful themes.

Read More

Reflection #8 – [02/20] – [Jiameng Pu]

Bond, Robert M., Christopher J. Fariss, Jason J. Jones, Adam DI Kramer, Cameron Marlow, Jaime E. Settle, and James H. Fowler. “A 61-million-person experiment in social influence and political mobilization.” Nature 489, no. 7415 (2012): 295.

Summary & Reflection:

Since traditional human behavior mainly spread by face-to-face social networks, which is difficult to measure social influence effects in it. But for online social networks, it might be a possible way to evaluate social interaction effects in it and the paper conducts a randomized controlled trial of political mobilization messages to identify social influence effects. Since the act of voting in national elections is a typical behavior that spreads through networks, the paper conducted a randomized controlled trial with all users in Facebook who access the website on the day of US congressional elections in 2010. All the users are randomly assigned to three groups — a ‘social message’ group, an ‘informational message’ group or a control group. The results imply that it’s not rigorous to say previous research suggested that online messages do not work, which possibly caused by small conventional sample sizes. The political mobilization messages can directly influence political self-expression, information seeking and real-world voting behavior of millions of people. In addition, the experiment measuring indirect effects that spread from person to person in the social network suggest strong ties are instrumental for people’s behavior in human social networks.

When measuring the direct effects of online mobilization by assigning users into three groups, they measure acts of political self-expression and information seeking. Personally, I don’t think the whole experiment design is rigorous and valid enough for several reasons. Firstly, there is a huge imbalance in the sample size of the social and informational message groups, i.e., 60,055,176 versus 611,044. I don’t think such a massive difference can be ignored, for instance, how do we make sure 611,044 users in group 2 are sufficient enough to represent the whole user community? Secondly, I’m not convinced that information seeking is a good indicator of people’s political positivity. If a person clicks “I voted” button, it would be very likely that he or she will not click the polling-place link because they’ve voted.

Kramer, Adam DI, Jamie E. Guillory, and Jeffrey T. Hancock. “Experimental evidence of massive-scale emotional contagion through social networks.” Proceedings of the National Academy of Sciences 111, no. 24 (2014): 8788-8790.

Summary & Reflection:

Emotional states can be contagious, which leads people to experience the same emotions without their awareness. The paper test whether emotional contagion occurs outside of in-person interaction between individuals by reducing the amount of emotional content in the News Feed on Facebook. It turns out that emotions expressed by other users on Facebook influence our own emotions, constituting experimental evidence for massive-scale contagion via social networks.

For the experiment design, they use Linguistic Inquiry and Word Count software word counting system to determine whether posts are positive or negative, which I don’t think is sufficient. This is like sentiment analysis by only counting positive and negative words. How about positive sentences expressed by the double negative? It would classify sentences with sarcasm into positive posts, but we all know that language expression is a pretty complex phenomenon. From my perspective, we may solve this problem by applying classic models of polarity analysis on posts. My another concern would be whether people update positive posts are actually happy in real life. There are many examples that people’s attitude showed in their social media does not necessarily represent their real mood or life status, sometimes people even pretend to be positive. This would raise another question about the definition of emotional contagion.

Read More

Reflection #7 – [02/13] – Jiameng Pu

Niculae, V., Kumar, S., Boyd-Graber, J., & Danescu-Niculescu-Mizil, C. (2015). Linguistic harbingers of betrayal: A case study on an online strategy game. arXiv preprint arXiv:1506.04744.

Summary:

The paper explores linguistic cues that indicate fickle interpersonal relations, like close friends becoming enemies. Since data that define the relationship between friends or enemies are not extensively accessible, researchers turn to a war-themed strategy game in which friendships and betrayals are orchestrated primarily through language. By studying dyadic interactions in the game and analyzing languages under cases that players form alliances and betray each other, they characterize subtle signs of imminent betrayal in players’ conversation and examine temporal patterns that foretell betrayal. From conversation scenarios in Diplomacy (the war-themed strategy game), we can actually see that betrayer would unconsciously reveal their planned treachery, meanwhile, the eventual victim can rarely be able to notice these signals.
They find that if the balance of conversational attributes like positive sentiment, politeness, and structured discourse shows sudden changes, imminent betrayal would happen in the future conversation. Then researchers provide a framework for analyzing communication patterns and explore linguistic features that are predictive of whether friendships will end in betrayal. They also discuss how to generalize methods to other domain and how automatically predicting relationships between people can help advance the study of trust and relationships using computational linguistics.

Reflection:

“Despite people’s best effort to hide it, the intention to betray can leak through the language one uses.” This reminds me of another idea that we may use the same strategy to detect people’s relationship pattern. Under normal circumstances, there are different relationship patterns when people getting along with each other. For example, some are balanced relationships like friends, colleagues, and relatives, while others are unbalanced relationships like leaders and subordinates, professors and students. Through the conversation content between people, we can extract the linguistic features following the same direction such as sentiment, argumentation, and discourse, politeness, and talkativeness, to predict the possible relationship between people. By analyzing the patterns of interpersonal relationships, we can have a deeper understanding of the current status of people’s life or whether the patterns will change over time, which is a more macro-sociological issue to figure out.
Apart from detecting the relationship pattern, we may use specific semantic features to study much other potential information hidden in the conversation, such as trust, familiarity, and intimacy between people.
Since the intuition of this paper is that a stable relationship should be balanced, it makes sense that all the predictions of Betrayal in the paper is based on signal an imbalance in the communication patterns of the dyad. However, my concern is whether these mentioned semantic features could provide a complete and efficient predictive analysis. Are there other available properties? e.g., Humor, straightforwardness. Or, in addition to detecting the imbalance of both sides in conversation, we may analyze the change of speech mode of the betrayer according to the timeline, i.e., imbalance of speech pattern before and after the decision to betray. This can solve difficult problems, for example, some people in nature are not as polite as the other party of the conversation.
In addition to the logistic regression which is often used for binary classification, support vector machine(SVM) is another classic algorithm that can be used for classification. As they have different advantages, we can design a control group in the experiment to choose the best classifier. Similarly, the semantic features can also be experimented in a controlled manner, so as to select the optimal combination of linguistic features that are most efficient for predicting betrayal.

Read More

Reflection #6 – [2/8] – Jiameng Pu

Danescu-Niculescu-Mizil, C., Sudhof, M., Jurafsky, D., Leskovec, J., & Potts, C. (2013). A computational approach to politeness with application to social factors. arXiv preprint arXiv:1306.6078.

Summary:

The paper focuses on exploring the content politeness on social platforms like Wikipedia, Stack Exchange. They conduct linguistic analysis exhibiting politeness on the corpus of requests and extract positive and negative politeness strategies from the context. In the experiment, two classifiers, a bag of words classifier (BOW) and a linguistically informed classifier (Ling.), are compared to illustrate the effectiveness of their theory-inspired features. They also show the relationship between politeness levels and social power/reputation. In general, we see a negative correlation between politeness and power on Stack Exchange or Wikipedia. They also add that there are differences in politeness levels across factors of interest, such as communities, geographical regions, and gender.

Reflection:

In the paper, they use a bag of words classifier as a strong baseline for the new classification task. However, it didn’t mention whether the BOW classifier is the state-of-the-art, thus the advancement of BOW needs to be discussed or some other classifiers are expected to be listed in the experiment. With a more effective classifier, where can we take advantage of it to benefit users or virtual communities? For instance, we may attach users in the community with a ‘politeness’ feature, hostile users will be automatically muted or banned if their politeness is lower than the floor value. We can also use it to evaluate the politeness of online customer service extensively used in the electronic business, which would benefit the environment of online shopping.

 

Voigt, R., Camp, N. P., Prabhakaran, V., Hamilton, W. L., Hetey, R. C., Griffiths, C. M., … & Eberhardt, J. L. (2017). Language from police body camera footage shows racial disparities in officer respect. Proceedings of the National Academy of Sciences, 201702413.

Summary:

Similar to the politeness-level difference analyzed in the above paper, one of the most non-negligible topics is the respect-level between different races(even gender, nationality). The paper collects footage from body-worn cameras, we extract the respectfulness of police officer language toward white and black community members by applying computational linguistic methods on transcripts. Disparities are shown in the speaking way of officers toward black versus white community members.

Reflection:

For the data collection, the source of datasets is not geographically diverse. Given that the issue of racial discrimination in different regions may not be of different levels of severity, I think it might be a better choice to collect footage in more than one place in the United States, not just Oakland, which would provide a holistic point of view of how officer treat white and black drivers. On the other hand, the race of officer is a key control factor since I’m curious about how black officer treats black and white drivers in their routine traffic stops, which is not discussed in detail though.
I noticed that the data they extracted were footage from body-worn cameras, transcripts were mainly used in the research. With the development of computer vision technology, extracting footage for the frame detection as an additional feature may help us to better understand the relation pattern of between police officers and drivers. For example, the frequency of conflicts between officers and drivers is also an important measure of respect degree.

Questions:

  • For these topic-similar paper, is there a clear difference between being polite and being respectful?
  • Since it is not easy for social scientists to measure how police officers communicate with the public, researchers use body-worn cameras to collect data. However, with body-worn cameras capturing interactions every day, how do we guarantee police officers would act exactly the same as the way when they don’t wear cameras? Based on this, keeping data collection procedure agnostic to police officers might be an alternative choice for researchers?

Read More

Reflection #5 – [02/06] – Jiameng Pu

Garrett, R. K. (2009). Echo chambers online?: Politically motivated selective exposure among Internet news users. Journal of Computer-Mediated Communication, 14(2), 265-285.

Summary:

The Internet offers users abundant access to information from around the world, which expose people to more diverse viewpoints, including both similar perspective and attitude-challenging information. With so much information, voice comes up that whether the Internet leads people to adopt increasingly insular political news exposure practices, i.e., “echo chambers” and “filter bubbles”, which refer to cases that individuals tend to be exposed only to information from like-minded individuals or materials similar to previous reading due to the influence of algorithms the news sites employed. This paper tests the viewpoint by conducting a web-administered behavior-tracking study. They find that the effect of opinion-challenging information is small and only marginally significant, which means worry that the Internet will lead to an increasingly fragmented society appears to have been overstated.

Reflection:

The paper shows readers how to conduct an exhausted and comprehensive experimental design when studying a specific problem, from selecting the research subjects to designing the experimental setting/workflow, e.g., using a purpose-built software administrates the study. Appropriate groups should be chosen to study according to the problem itself — two partisan online news services, left AlterNet and right WorldNet Daily, are able to bring immediate benefits to the study with the inherent engagement in selective exposure. Another thing I am concerned about is that there are not enough samples participating in the experiment, i.e., a total of 727 subjects (358 Alternet readers and 369 WorldNetDaily readers respectively), which possibly means the generalization of our observation result is not universal enough. Although the author mentions in the limitation that “when viewing these results is that the sample is not representative of the U.S. population at large”, I think it is very difficult to represent groups with hundreds of subjects, not to mention specific group like U.S. population at large. Why do not we find some volunteers on university campuses or research institutes, e.g., students and faculties, which is easier to get more volunteers?

Bakshy, E., Messing, S., & Adamic, L. A. (2015). Exposure to ideologically diverse news and opinion on Facebook. Science, 348(6239), 1130-1132.

Summary:

Based on the same problem background as the first paper, e.g., “echo chambers” and “filter bubbles”, the paper attempts to explore how this modern online Internet environment will influence people’s exposure to diverse content. Past empirical attempts turn out to be limited because of difficulties in measurement and show mixed results. To overcome previous limitations, they use a large, comprehensive Facebook dataset to measure the extent of ideological homophily and content heterogeneity in friend networks. They found that individual choice plays a stronger role than algorithmic ranking does in content exposure.

Reflection:

One heuristic point to take away is to classify stories as either “hard” (such as national news, politics, or world affairs) or “soft” content (such as sports, entertainment, or travel) by training a support vector machine, which reduces the data size while keeping the most efficient part of the dataset since “hard” content tends to be more controversial than soft content. However, some of the “soft” content is also practicable and appropriate for the task, i.e., hot and controversial topics in sports and entertainment fields. Thus the author may consider augmenting the size of the dataset to find out the result.

Same as the vague distinction between exposure and consumption mentioned in the limitation analysis, I also think how to clarify the line among conservative, neutral, liberal? Can we possibly give a uniform and appropriate criteria for the classification of those three categories?

Read More

Reflection #4 – [01/30] – [Jiameng Pu]

  • A parsimonious language model of social media credibility across disparate events

Summary:

With social media’s dominance of people’s acquisition of news and events, the credibility of content on different platforms tend to be less rigorous than that of content on traditional journalists. In this paper, the author conducts research on analyzing the credibility of content on social media and proposes a parsimonious model that can map language cues to perceived levels of credibility. The model is presented based on examining the credibility corpus of Twitter messages corresponding to 15 theoretically grounded linguistic dimensions. It turns out that there are considerable indicators of events’ credibility in the language people use.

Reflection:

The paper inspires me a lot by leading me to go through a whole research process from idea creation to idea implementation. In the data preparation phase, it’s a very common task to annotate with ordinal values on the content we are studying, the proportion-based ordinal scale PCA was used in this paper can help compromise extreme conditions, which is like a trick that I can take away and try in other studies. The author uses logistic regression as the classifier, I think neural networks should also be a good choice to make classifications. Specifically, neural networks used for classifying usually has inputs of feature number and outputs of class number. Potentially, neural networks might help us derive better-classifying performance, which then helps the analysis of feature contribution.

  • The Promise and Peril of Real-Time Corrections to Political Misperceptions

Summary:

Computer scientists create real-time correction systems to label inaccurate political information with the purpose of warning users of inaccurate information. However, the author is skeptical about the efficiency of the real-time correction strategy and leads an experiment to compare the effects of real-time correction to non-real-time correction. In the design phase of the comparative experiment, the researchers conduct a control variable method and assess participants’ perceptions of facts through questionnaires. Then they construct a linear regression model to analyze the relationship between belief accuracy of participates and the correction strategy. The paper concludes that real-time corrections are modestly more effective only among individuals predisposed to reject the false claim.

Reflections:

This paper conducts a comparative study about effects of immediate and delayed corrections on readers’ belief accuracy. Generally, one of the most important parts of comparative research is to design a reasonable and feasible experimental scheme. Although lots of big data research needs to collect ready-made data for preprocessing, some research requires researchers themselves to “produce” data. Thus the design scheme of getting data has a significant impact on subsequent experiments and analysis.
In the first survey-based step for data collection, choice of survey samples, setting of control variables, and evaluation methods are main points that can greatly affect the experimental results, such as the diversity of participants (race, gender, age), the design of delayed correction, and the design of the questionnaire.

Particularly, in order to achieve the delay correction, the author employed a distraction task—participates were asked to complete a three-minute image-comparison task. Although this task can achieve the purpose desired, this is not the only strategy we can perform. For example, the duration of the distraction task may have a different impact on the participants’ cognition of the news facts, so researchers can try multiple durations to observe whether there is different impact. In analyzing section, linear regression is one of the most common models used in result analysis. However, for some complex issues without a strict rule, the error of a linear regression model is potentially larger than that of a nonlinear regression model. Nonlinear regression with appropriate regularization is also an option to choose.

Question:

  1. As analyzed in the limitation section, although the author tries to make the best possible experimental design, there are still many design decisions that affect the experimental results. How can we do this to minimize the error?
  2. Intuitively, it is more proper to study mainstream reading group on the Internet, the average age of the study object is too large under this circumstances.

Read More