Appunti Digital marketing

Social Media Sentiment Analysis for Universities' Evaluation

The SA system accepts learning diaries as input and splits them on the basis of the date entry. Afterward, the system performs three fundamental steps: emotion extraction from the diary, negative/positive attributes' identification, topics' extraction. As for the output, the system returns charts that visualize the emotional states identified and their respective evolution over time.

As regards the identification of emotions, that is one of the most important tasks faced by the system: it compares each sentence, obtained during the splitting pre-processing step, against the NRC word-emotion lexicon, that was manually annotated into eight emotion categories, according to Plutchik's eight basic emotions (i.e., joy, sadness, fear, anger, anticipation, surprise, disgust and trust).

Tummel & Richert propose an approach for the analysis of opinions published on Twitter, as a complementary source for evaluating.

universities; they used data related to 9 German Institutes of Technology (TU9) by collecting all tweets matching word from a list of keywords containing combinations of the universities names and titles belonging to them for a time spanning a winter semester. The dataset has been processed in three steps: filtering, feature extraction and sentiment analysis out-and-out. A second analysis allowed the authors to monitor sentiments about daily events and activities at each university. To carry on this task, all classified tweets are divided on daily basis, in order to analyze specific temporal points, either positive or non-positive, using word frequencies estimates for each point. MOOC's Sentiment Analysis MOOCs suffer in terms of satisfaction of students' needs whose most direct consequence is the high dropout rate. Shen & Kuo, in their work, introduced an approach that allows to draw descriptive statistics about trends of MOOCs on Twitter, allowing the observation of.

1. participants’ activities on Twitter as well as identifying the right time to post or analyze tweets.
2. The approach also analyzes public sentiment in Twitter towards MOOC learning, looking for related tweets, obtaining insights about perceptions of MOOC learning.
3. Finally, analyzing positive and negative MOOCs related retweets, the approach identifies influencers of these retweets in order to allow MOOCs analysts recommending MOOC related messages to the influencers.
4. In order to obtain also the sentiment, positive or negative, that the top influencer disseminates or he is going to do, it is performed a sentiment analysis on the retweet.

LADEL Architecture

LADEL consists of four main tunable modules:

1. a) collection module,
2. b) cleaning module,
3. c) word cloud module,
4. d) the sentiment opinion module.

An important role in the architecture of LADEL is played by the employing of dashboards (from which it derives its name) that can be tuned depending on the task we are dealing with.

a) Collection

module: the main component is the searchTwitter() function which returns tweets containing the search string. The corpus of tweets thus collected are stored in object data so that they can be easily processed when necessary. b) Cleaning module, as a canonical cleaning process dirt html links were removed, second retweets were removed, third # and @ and then all the punctuation, numbers and unnecessary spaces were removed. c) Word cloud module, word clouds visualize the weights of the words present. In particular, for sentiment analysis purposes, word clouds allow for an understanding of the kind of sentiments present in the collection/corpus of items and why these items belong to one or more emotion categories. d) Sentiment opinion module, there are several approaches for sentiment analysis and those based on supervised machine learning are among the most performative and adaptable to a given task because of the availability of tagged dataset, as well as the possibility of labelling and

training new models for unexplored domains. Here 3 ML algorithms are employed to classify sentiments. The core of this step is represented by the boolean match of each word w.r.t. the positive or negative opinion-lexicon.

E-commerce websites services versus buyers' expectations: an empirical analysis of the online marketplace

With the growth of online shopping, the buyers are faced with information and cognitive overload, entailing worse buyers' decisions. Various decision aids, more and more implemented as web services, aim at reducing this overload. Often, they implement compensatory strategies that enable desirable and undesirable values of a product attribute to compensate each other. → increasing the number of options beyond a handful can lead to poor choices, decreasing satisfaction people rely more on heuristics or non-compensatory strategies than rationality to arrive at decisions and purchases.

The inquiry revealed the existence of a gap between what E-Commerce

websites/sellers provide and whatE-Commerce buyers want, findings suggest the use of non-compensatory strategies when implementing E-Commerce websites, in order to bridge the existing gap Information and cognitive overload • Kirsh identified four primary causes of cognitive overload: 1) Too much information supply; 2) Too much information demand; 3) Constant multi-tasking and interruption; 4) Inadequate workplace infrastructure. • Chen proposed the role of information filtering mechanisms to examine the effect of information load→on subjective state towards decision rich information leads to a perception of high informationoverload and lead costumers to a worse subjective state towards decision. • Arana and Leon suggest that information filtering tools as mentioned above, may have influences onrelieving but are not the panacea to the phenomenon of information and cognitive overload, wherenovice consumers may face a more serious information overload problem than experienced.

The role of heuristics

The Nobel laureate Herbert Simon demonstrated that an organism requires simple choice mechanisms (i.e., heuristics) to satisfy its several needs. People, in situations involving uncertainty, rely more on heuristics than rationality to reach a decision.

Heuristics – Valerdi: simple, efficient rules, hard-coded, either by evolutionary processes or learned, which have been proposed to explain how people make decisions, come to judgments, and solve problems, typically when facing complex problems or incomplete information.

• Gigerenzer and Goldstein, using a computer simulation, compared heuristic algorithms and various rational inference mechanisms (e.g., multiple regression) and showed that the heuristic ones matched, or out-performed, all competitors in inferential speed and accuracy.
• Papatla and Liu also investigated the role that both search engines and infomediaries play in bringing unplanned consumers and showed that the former play a stronger

role than the latter. According to a Jupiter Research report, 33% of buyers often or sometimes make unplanned purchases, demonstrating the need for more heuristic-oriented services implemented on E-commerce sites in order to meet specific buyers' needs. Compensatory and non-compensatory strategies: In general, the implementation of non-compensatory strategies displays a list of options that meet a predetermined set of thresholds set by the decision maker. For example, a website service allowing the buyer to enter text for searching or choosing from a list of criteria (e.g., shopping by categories or by price) are implementations of non-compensatory strategies that do not enable trading or a low value of an attribute with a high value of another. The use of non-compensatory strategies may result in the elimination of some alternatives based on criteria set on one attribute, even if such options may be very attractive on other attributes. Compensatory strategies enable desirable and

undesirable values of a product attribute or service to compensate for each other: a website displaying an array of rows of products with columns for various attributes; another example is external ratings of products.

Gudigantala et al compared compensatory versus non-compensatory strategies on different grounds. First, they compared the two strategies surveying five studies from literature, using variables such as satisfaction, decision quality, effort and confidence.

The main findings are a complete absence of compensatory strategies even if the survey of the literature shows that implementation of these strategies, as web decision support systems, would provide better decision quality, satisfaction, and confidence to buyer, reducing his effort as well.

Sellers Online Services: A Literature Survey

Findings are based on 3 kinds of evidence according to web/shopping decision support system, that implement non-compensatory strategies, are preferred by buyers on the online

1. marketplace.Implementing those strategies could help bridge the gap between sellers and buyers in the online marketplace.
2. Hauble and Trifts observed that buyers often seem unable to evaluate all possible alternatives in depth, so they tend to use a two-stage process to make a decision:
   1. Screen a large set of available products identifying a subset of the most promising alternatives;
   2. Evaluate the subset in depth, comparing products based upon important attributes and, finally, they decide.
3. The authors consider the support of "interactive tools", or "decision aids", aiming at assisting buyers in performing each of the above tasks, to be valuable.
4. The results showed that decision aids have strong and beneficial effects on both the quality and the efficiency of purchase decisions, allowing buyers to make better decisions, with less effort.
5. Chen et al. proposed an extended model with the aim of considering the roles of information filtering mechanisms, online shopping

Experience and perceived information overload, in order to analyze the effects of information load on subjective state towards decision. Experiments, performed using a simulated online seller, showed three kinds of results:

1. Buyers perceive higher information overload while accessing large amount of information online.
2. The individual differences must be taken into account; individuals with different information processing abilities, and internal information filtering mechanisms, perceive differently the information overload.
3. More product information is not favorable both for buyers and sellers.

Causal map is widely used to describe expert knowledge in specific domains, they may help buyers to consider quantitative and qualitative factors simultaneously. Experiments show that a recommendation mechanism may benefit from using a causal map.

Sellers and Buyers on the Online Marketplace. The authors developed a form to be used by the researchers for analyzing websites to record the

servicesthey provide and they posted an online questionnaire to get an idea of the other side of the coin (buyersexpectations).