280 likes | 297 Views
Memory Standardization. Meliton Padilla. Overview. Introduction Related work Methodology Contribution Questions. Abstract. Model the change of memory requirements for cell phones. Related work. Introduction. Methodology. Contribution. Questions. Todays standards. Related work.
E N D
Memory Standardization Meliton Padilla
Overview • Introduction • Related work • Methodology • Contribution • Questions
Abstract • Model the change of memory requirements for cell phones Related work Introduction Methodology Contribution Questions
Todays standards Related work Introduction Methodology Contribution Questions
Original approach • Potential Issues • Noise from multiple posts • Not enough text to generate data • Limited amount of data access Related work Introduction Methodology Contribution Questions
Product reviews Benefits • Less noise • Subject originated • Large sample sizes Related work Introduction Methodology Contribution Questions
Main goal • Extract feature specification from textual reviews • Target memory for multiple devices • Allow product review monitoring to inform when a change needs to be made Related work Introduction Methodology Contribution Questions
Related work Related work Introduction Methodology Contribution Questions
Ganesan, K., Zhai, C., & Viegas, E. (2012, April). Micropinion generation: an unsupervised approach to generating ultra-concise summaries of opinions. InProceedings of the 21st international conference on World Wide Web (pp. 869-878). ACM. Key attributes • Compactness • Summaries should use as few words as possible (between 2-5) • Representativeness • Summaries should reflect major opinions in text • Readability - Summaries should be fairly well formed Related work Introduction Methodology Contribution Questions
Micropinon • A set of short phrases expressing opinions on a specific topic or entity • Leading to a method of also creating reviews on character limited social sites Related work Introduction Methodology Contribution Questions
Example Related work Introduction Methodology Contribution Questions
Issues from textual anaylsis • Different types of grammar • Recreating a new sentence in order to capture original opinion (without using any original text) • How to tell the difference between a factual statement compared to an opinion Related work Introduction Methodology Contribution Questions
solution • Similarity scores: sim(mi,mj) • Measured with Jaccard similarity measure (or cosine) • Allows control redundancy of the same opinion • Readability scores: Sread(mi,mj) - Measure well form structure of phrases (Microsoft Web N-gram) • Representativeness scores: Srep(mi,mj) • Measure how well a phrase represents the opinion from original text • Captured by a pointwise mutual information (PMI) function Related work Introduction Methodology Contribution Questions
Example • Readability scores of phrases Related work Introduction Methodology Contribution Questions
Pang, Bo, and Lillian Lee. "Opinion mining and sentiment analysis." Foundations and trends in information retrieval 2.1-2 (2008): 1-135. Key attributes • Generating feature-based summaries • Distinguishing positive and negative comments • Grouping the data together to make looking for features easier Related work Introduction Methodology Contribution Questions
Example • Each summary should produce Related work Introduction Methodology Contribution Questions
Issues • How to tell if a opinion is positive or negative • Natural language processing techniques • Assuring the feature chosen is relatable to the product and not repeated Related work Introduction Methodology Contribution Questions
Solutions • Wordnet • System that helps find opinion words and frequent features • Part-of-Speech Tagging (POS) • Frequency of nouns, verb, adjective, etc. (Nlprocessor linguistic parser) • Orientation identification for opinion words - Only positive and negative orientations Related work Introduction Methodology Contribution Questions
Example • Using Wordnet to create a positive/negative approach a bipolar cluster Related work Introduction Methodology Contribution Questions
Methodology Related work Introduction Methodology Contribution Questions
Key differences • Focus just on the memory features of a device • Include other electronic devices besides just cell phones, examples such as laptops, mp3s and cameras • Sample current and past reviews to see if a trend can be modeled from the data Related work Introduction Methodology Contribution Questions
Processing techniques • Product reviews and previous data sets Related work Introduction Methodology Contribution Questions
Processing techniques Data is filtered Related work Introduction Methodology Contribution Questions
Processing techniques Steps needed • Collect large amount of data (may be separated by product type) • Extract opinion sentences and sort into a positive/negative category • Keep count of the positive to negative ratio • Use a similarity technique to measure the sweet spot of minimum required memory, in order to have a good product Related work Introduction Methodology Contribution Questions
Processing techniques Potential issues • Getting current reviews from Amazon • Currently provided API to view a current URL review page for 24hours • Comparing different products based on memory capability's • Analyzing textual data Related work Introduction Methodology Contribution Questions
Contribution • Being able to provide a way for consumers or manufacturers an easy method to decide on the memory required Related work Introduction Methodology Contribution Questions
References • [1] Pang, B., & Lee, L. (2008). Opinion mining and sentiment analysis.Foundations and trends in information retrieval, 2(1-2), 1-135. • [2] Ganesan, Kavita. "Micropinions vs. Micro-reviews." Text Mining, Analytics & More:. N.p., n.d. Web. 12 Oct. 2016. • [3] Ganesan, K., Zhai, C., & Viegas, E. (2012, April). Micropinion generation: an unsupervised approach to generating ultra-concise summaries of opinions. InProceedings of the 21st international conference on World Wide Web (pp. 869-878). ACM. • [4] Qadir, A. (2009, September). Detecting opinion sentences specific to product features in customer reviews using typed dependency relations. InProceedings of the Workshop on Events in Emerging Text Types (pp. 38-43). Association for Computational Linguistics