If an asker in acute need of answer, do not receiveany answers, due to ill-framing or poor asking styles, it mightcreate negative impact on the asker’s mind and the asker mighteventually leave the platform. This paper has potential to earlydetect such questions and recommend corrections which mightlead to answers from the community. Our system can also helppromoting these lowly visible questions to the more informedanswerers, experts so that they receive adequate number ofresponses, thereby, improving the answer rate

7) Psycholinguistic aspects of question texts—We considerthe LIWC scores from the different categories as featuresfor the model.8) ROUGE-LCS recall of the question text at the end ofthe observation period of the prediction with referenceto the original question text posted by the asker.2) User-Level Editing Activities:1) Total number of: 1) answers added by the question asker;2) answers removed by the question asker; 3) questiontext edits made by the asker; 4) question details editsmade by the asker; 5) comments made by the asker;6) topic added edits made by the asker; and 7) topicremoved edits made by the asker

Character length of a question, number of words in aquestion, fraction of nonfrequent words in a question,and number of function words in a question.2) INV and OOV words in question text—For each ques-tion text, we check whether a word appearing in thequestion text, is an INV word or OOV word by compar-ing with GNU Aspell dictionary. We then consider thefraction of INV words as a feature of our model.3) Presence of n-grams of the question content in Englishtexts—We search for 2, 3, 4 grams of the words fromthe question text in the corpus of 1 million contempo-rary American English words.6We use the presenceof bigrams, trigrams, 4 gm each as features for theprediction model.4) POSDiv of the words in the question. We also use thedifference in POSDiv between the initial question textand the question text after time periodt(observationperiod) as a feature to the model.5) Distribution of LDA topics obtained from questiontexts—For topic discovery from the question corpus,we adopt LDA [43] model, a renowned generativeprobabilistic model for discovery of latent topics in adocument. For a questionqi, we consider all the wordsin that question as a document for the LDA model.We set the number of topics asK=10, 20, 30 andfind outp(topick|Di)for a documentDicontaining allthe words of theith question. Each of thesep(topick|Di)fork=1...Kact as a feature of the model.6) LDA topical diversity—We also compute LDA topi-cal diversity (TopicDiv) of a question (qi) from thedocument–topic distributions obtained above

Open Questions and Their Topics

How Long Does a Question Remain “Open?

4) Our characterization further helps us to predict whethera given question remaining unanswered for a specifictime period

1) We identify and investigate two major kinds of lin-guistic activities on Quora: user level [e.g., basicactivities like posting a question/answer/comment aswell as linguistic styles that involves word/char usage,and part-of-speech (POS) tag usage] and questionlevel [e.g., content, topic associations, and edits for aquestion). Remarkably, many of these activities arefound to have a natural correspondence to the qualitiesthat human judges would consider while deciding if aquestion would remain unanswered (see Table I for a setof motivating examples).2) We perform an extensive measurement study to showthat answerability can be indeed characterized based onthe above-mentioned linguistic activities.3) A central finding is that the language use patterns ofthe users is one of the most effective mechanisms tocharacterize answerability.

Factors to judge if a question wouldremain open include its subjectivity, openendedness, vagueness,ambiguity, and so on. It is difficult to collect such judgments forthousands of questions, requiring automatic framework to dealthe issue of answerability of questions. In this paper, we quantify:1)user-leveland 2)question-level linguistic activities—that cannicely correspond to many of the judgment factors noted earlier,canbeeasily measured for each question postand thatappropriatelydiscriminates an answered question from an unanswered one.Ourcentral finding is thatthe way users use language while writingthe question text can be a very effective means to characterizeanswerability

In order to capture such psycholinguistic aspects of the asker,we use Linguistic Inquiry and Word Count (LIWC) (Pen-nebaker, Francis, and Booth 2001) that analyzes various emo-tional, cognitive, and structural components present in indi-viduals’ written texts

Content of a question text is important to attract people andmake them engage more toward it. The linguistic structure(i.e., the usage of POS tags, the use of Out-of-Vocabularywords, character usage etc.) one adopts are key factors foranswerability of questions

In this work, we show that appropriate quantifi-cation of variouslinguisticactivities can naturally correspondto many of the judgment factors

aquestion would remain open based on factors like if it issubjective, controversial, open-ended, vague/imprecise, ill-formed, off-topic, ambiguous, uninteresting etc

it is important to identify the openquestions and take measures based on the types - poor qualityquestions can be removed from Quora and the good qualityquestions can be promoted so that they get more visibility andare eventually routed to topical experts for better answers

In Quora, the questions with no answers arereferred to as “open questions”. These open questions needto be studied separately to understand the reason behind theirnot being answered or to be precise, are there any character-istic differences between ‘open’ questions and the answeredCopyrightc2017, Association for the Advancement of ArtificialIntelligence (www.aaai.org). All rights reserved.ones

In this study we focus on the answerability of questionson Quora, i.e., whether a posted question shall eventuallyget answered