<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Yash Mishra on Medium]]></title>
        <description><![CDATA[Stories by Yash Mishra on Medium]]></description>
        <link>https://medium.com/@myash180?source=rss-81e501960f08------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*EB_tfceNt22mBQJH_t0nFg.jpeg</url>
            <title>Stories by Yash Mishra on Medium</title>
            <link>https://medium.com/@myash180?source=rss-81e501960f08------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 22:32:55 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@myash180/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Question-Answering Chatbot Development using XLnet transformer model for Digital/Smart Learning]]></title>
            <link>https://medium.com/@myash180/question-answering-chatbot-development-using-xlnet-transformer-model-for-digital-smart-learning-3d142754655d?source=rss-81e501960f08------2</link>
            <guid isPermaLink="false">https://medium.com/p/3d142754655d</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[xlnet]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[smart-learning]]></category>
            <dc:creator><![CDATA[Yash Mishra]]></dc:creator>
            <pubDate>Sat, 26 Nov 2022 19:49:53 GMT</pubDate>
            <atom:updated>2022-11-26T19:49:53.559Z</atom:updated>
            <content:encoded><![CDATA[<p>Society is becoming more “mobile-first” as a result of digitization (Oracle, n.d.). As messaging apps grow increasingly common, chatbots are becoming more important in this mobility-driven revolution (Oracle, n.d.). Intelligent conversational chatbots are increasingly being utilized as interfaces for mobile applications, and they’re changing the way businesses and customers communicate (Oracle, n.d.). Chatbots enable businesses to communicate with consumers on a personal level without incurring the costs of hiring human representatives (Oracle, n.d.). Chatbots are nothing more than a program that engages in conversation with the user and responds to their questions (Oracle, n.d.). Chatbots, often known as “bots,” can understand natural language and respond to users’ inquiries (Chat Compose, n.d.). These replies, on the other hand, might take the shape of a job, such as visiting a website, exhibiting an image, or watching a movie (Chat Compose, n.d.). Almost everyone has had a conversation with a chatbot at some point in their lives (Chat Compose, n.d.). Chatbots, in most circumstances, save a lot of time and human resources while also providing a superior answer without human mistakes or emotions (Chat Compose, n.d.). This research is centered on the role of chatbots in digital and smart learning, as well as their development utilizing deep learning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/0*GW8JrbxU4XoY-bxF.jpg" /></figure><h4>Transformer Model</h4><p>A transformer is a deep learning model that employs the self-attention process to assign different weights to each element of the input data (Wikipedia, n.d.). Natural language processing and computer vision are two areas where it is widely used (Wikipedia, n.d.).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/940/1*MTd7wI-SL3RzYZNe5dQ_8g.png" /><figcaption>Chatbot Architecture</figcaption></figure><p>Transformers, like recurrent neural networks (RNNs), is designed to analyze sequential input data, such as natural language, and are used for things like translation and text summarization (Wikipedia, n.d.). In contrast to RNNs, transformers handle the entire input at once (Wikipedia, n.d.). The attention mechanism provides context at any point in the input stream (Wikipedia, n.d.). The transformer does not have to handle one word at a time if the incoming input is a natural language phrase, for example (Wikipedia, n.d.). In comparison to RNNs, this allows for more parallelization, resulting in quicker training times (Wikipedia, n.d.).</p><h4>XLNet Model</h4><p>XLNet is a game-changing advancement in natural language processing and language modeling. Language modeling is the process of creating models that can extract long-range contextual information and important statistical qualities from a corpus or body of text sequences. Language models that have been trained may predict the next token in a sequence, and the weights can be used to input into other NLP models. The more semantic information and features that a model can extract from a corpus, the better it will perform on subsequent challenges (Srinivasan, 2019).</p><h4>Chatbot Environment</h4><p>The chatbot environment shown in the figure below is made up of several components. User communications are one of them. They’re a dynamic input that our model can get at any moment. They are made up of a string that represents the user’s real text, along with some metadata structured to have additional information that can describe the text more understandingly. Metadata may contain any images or links sent by the user, the timestamp of the message sent, the platform from which it is sent, etc. The chatbot only has read-only rights and it cannot change the text sent by the user.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/939/1*QBWrNhfn_BOPzM9txCZlbQ.png" /><figcaption>Chatbot Environment</figcaption></figure><p>Another part of the chatbot environment it is the trained model. It is the model that we trained using the Transformer XLNet model and the dataset collected. It is capable of handling the user and giving all the answers without human intervention. It can give suggestions to users on the basis of their queries in order to help them in every possible way.</p><h4>Dataset Used for Training</h4><p>For the implemented chatbot training, the Stanford Question Answering Dataset has been used. The Stanford Question Answering Dataset (SQuAD) is a reading comprehension dataset comprised of questions submitted by crowd workers on a collection of Wikipedia articles, with each question’s response consisting of a text segment, or span, from the relevant reading passage, or the question being unanswerable. The SQuAD has 2 versions where the v1.1 overall dataset size is 94.06 MiB and has been split into ‘train’ and ‘validation’ whereas ‘train’ has 87599 examples and ‘validation’ has 10570 examples. Similarly, v2.0 has an overall dataset size is 148.54 MiB and has been split into ‘train’ and ‘validation’ whereas ‘train’ has 130319 examples and ‘validation’ has 11873 examples (TensorFlow, n.d.).</p><p>The figure below shows the squad v2.0 structure of the data in which it is the present and the first sample of the dataset. There are columns of title, which have the title of the context, context, from which the answer is provided, question, id, and answers with the text and its starting point in that particular context.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/940/1*fmL-4zWPnkM3wVyGGVJ3LA.png" /><figcaption>The data structure of Squad v2.0 with the first sample</figcaption></figure><h4>Data Structure Used</h4><p>In order to access the data easily, a simple data structure is used. Usually, a recommended database for the Question Answering chatbot is SQL. For the shorter version, instead, we have used a JSON file with a demo dataset in it.</p><p>As shown in Figure below, layer 1 of the JSON is comprised of two keys, namely, “subject_name” and “topics”. “subject_name” represents the name of the subject as shown in the demo dataset as “Geography”, “History”, etc whereas “topics” represents the topics of which the content is present in that particular subject.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/619/1*J39iwB-Vqod1uVqNB3QERQ.png" /><figcaption>Layer 1 of the Data structure used</figcaption></figure><p>The figure below shows the nested layer or Layer 2 of the demo dataset. It comprises 2 keys i.e., “topic_name” and “description”. Here, “topic_name” represents the topic in that particular subject, and “description” represents the content related to that particular topic of that subject.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/940/1*Bpb8-RQuFjJkAOYIGDWeOw.png" /><figcaption>Nested Layer of Data structure used</figcaption></figure><p>The description is only string data, that can be directly used as context. But in many cases, there can be nested topics and they can be used as a list format in the description itself. The script running can take care of the list of dictionaries as well.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/940/1*vDBr59P6WiRZ1rvHM9OJgQ.png" /><figcaption>Another Nested Layer of Data structure used</figcaption></figure><h4>Fine Tuning</h4><p>As mentioned before, the huggingface is a platform where various NLP pre-trained algorithm models can be found. The algorithm used in this research is the XLnet algorithm. XLnet is a pre-trained version of the Transformer-XL model that maximizes the expected likelihood across all permutations of the input sequence factorization in order to learn bidirectional contexts using an autoregressive technique.</p><p>For our use, the algorithm was used XLnet but the pre-trained model used was “jkgrad/xlnet-base-squadv2”. XLnet was jointly developed by Google and CMU and fine-tuned on SQuAD2.0 for question-answering downstream tasks. The training result metrics of the pre-trained model were as follows:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/648/1*ous-Cw2FTWsT5A6Hsrg3Xg.png" /><figcaption>Training Result Metrics of jkgrad/xlnet-base-squadv2</figcaption></figure><p>As shown in the Figure above, the base model of XLNet was fine-tuned with an epoch value of <strong>“4”</strong>. The epoch value is the number that represents the number of times we want to train the model with the same dataset. In order to get a significant result and value, the model needs to be trained with a greater number of epochs. Hence, we fine-tuned the model with an epoch value of <strong>“16”</strong>.</p><h4><strong>Steps to fine-tune the model</strong></h4><ol><li>Load the dataset:</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/595/1*0TDPGL1LmWQ2HWU8uyGquw.png" /></figure><p>2. Load the tokenizer:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/940/1*_KDe-5Y0BBzAwbkmyzesYQ.png" /></figure><p>3. Load the model:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/940/1*UDRlQ2S87QDVYMF0CVEpSQ.png" /></figure><p>4. Tokenize (encode) the dataset:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/411/1*W9UuWSi-RCR586Gzsw7SWQ.png" /></figure><p>5. Fine-tune the model using the encoded value:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/586/1*Hk-wL15WwS-LsQTfj3palw.png" /></figure><p>The overall code can be found <a href="https://github.com/tron135/xlnet_chatbot/tree/main"><em>here</em></a><em> </em>which is used for fine-tuning. Here, the script demands high-end RAM and GPU availability in order to run the fine tune code. Due to the unavailability of high-end devices, the XLNet mode is trained with 8 epochs twice. Which will complete the target epoch, instead of running it for 16 epochs at once.</p><p>The figure below shows the last 8-epoch model results with training loss and validation loss. As shown in the Figure below, the training loss is 0.193900 which is very good. It might reduce mode if trained with varieties of datasets and more epochs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/481/1*H4ezICujo-29jprfY_OoYw.png" /></figure><p>If you wanna check the detailed version with results, then it can be found <a href="https://github.com/tron135/xlnet_chatbot/blob/main/st20193024%20CIS7017.pdf"><em>here</em></a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3d142754655d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Business Analysis on Tesco 2022]]></title>
            <link>https://medium.com/@myash180/business-analysis-on-tesco-2022-3c416398fac1?source=rss-81e501960f08------2</link>
            <guid isPermaLink="false">https://medium.com/p/3c416398fac1</guid>
            <category><![CDATA[brexit]]></category>
            <category><![CDATA[in-depth-analysis]]></category>
            <category><![CDATA[business-process-analysis]]></category>
            <category><![CDATA[tesco]]></category>
            <dc:creator><![CDATA[Yash Mishra]]></dc:creator>
            <pubDate>Wed, 23 Nov 2022 19:51:38 GMT</pubDate>
            <atom:updated>2022-11-23T19:51:38.003Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>Tesco Plc is the largest food retailer in the United Kingdom and is one of the major British retailers, ranking third behind Walmart in the United States and Carrefour in France in the retail industry. It sells roughly 4000 different culinary items (Euromonitor, 2010).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/940/1*baoi9Oj36v3I0_iojpTxNQ.png" /><figcaption>Tesco Procedure</figcaption></figure><p>Tesco has expanded globally since the early 1990s, with operations in 11 other countries in the world (Wikipedia, 2022). The company pulled out of the US in 2013, but as of 2018 continues to see growth elsewhere (Wikipedia, 2022). Since the 1960s, Tesco has diversified into areas such as the retailing of books, clothing, electronics, furniture, toys, petrol, software, financial services, telecoms, and internet services (Wikipedia, 2022). In the 1990s Tesco repositioned itself from being a downmarket high-volume low-cost retailer, attempting to attract a range of social groups with its low-cost “Tesco Value” range (launched 1993) and premium “Tesco Finest” range (Wikipedia, 2022).</p><p>Since Tesco is well established and locally well known, an industry this big took a big hit due to the pandemic. According to the Balance sheet of Tesco PLC on Yahoo Finance, Tesco’s assets were in a constant increase from 2018–2019–2020 (Yahoo Finance, 2022). But in 2021, it took a great hit with a loss of 13.304% (Yahoo Finance, 2022). Tesco has a large presence in West Europe (accounting for 78.4% of sales in 2019) and the severe impact of COVID-19 in this region causing stockpiling early on, has driven sales for the grocer (AP News, 2021). To cope with swelled demand for online grocery, Tesco has mobilized its logistic network in the UK and increased the number of delivery slots it offers.</p><h3>Analysis</h3><h4>1. Strategic analysis using Michael Porter’s Five Force Model</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/634/1*pJahnkkpkt89nz1Y6dSJ2w.png" /><figcaption>Michael Porter’s Five Forces Model</figcaption></figure><p>1.1. Competitive rivalry within Industry</p><p>The intensity of competitive rivalry within the industry is moderate to high for Tesco because of the community’s increased need and supply of commodities to meet fundamental necessities. As a result of this growth, a number of groups and corporations have seized the market in order to give multiple platforms for clients to purchase from, resulting in severe competition among the established players. Tesco is the consistent market leader in terms of grocery market share with a 27 percent share as of December 2020 followed by Sainsbury’s with 15.7%, ASDA with 14.1%, Morrisons with 10.3%, ALDI with 7.7%, and The Cooperative with 6.3%. As a result, the presence of such well-known names in the sector makes the operating atmosphere more competitive.</p><p>1.2. Bargaining Power of Buyers</p><p>The intensity of the bargaining power of buyers is high because of the availability of the various vendors and the buyers can buy the products from wherever they like. On top of it, the prices are comparable hence the buyer has a significant amount of power here. Customers’ awareness of product prices and offers has increased as a result of online channels, which influences customer purchasing behavior. Hence it is hard to predict customer behavior.</p><p>1.3. Threats of New Entrants</p><p>Looking as much Tesco has established in the past few decades, a newcomer retailer is very less likely to be a threat to Tesco. Hence the intensity of this threat is low. It might be a moderate threat to other retailers with new entrants considering a few factors such as convenient accessibility, locality, specialty, etc.</p><p>1.4. Bargaining Power of Suppliers</p><p>The intensity of bargaining power of the suppliers is low to moderate since Tesco has a vast number of suppliers, so it can choose from whichever supplier is giving the product cheaper with less effort. Hence the supplier does not have much of an option and the intensity decreases the supplier. A high number of vendors fight for a limited amount of space in retail stores. The increased supply capacity makes it challenging for suppliers to have an effect on the successful growth of the company, which allows them to hold specific authority over them. Thus, the bargaining power of suppliers is kept low to moderate.</p><p>1.5. Threat of Substitute Products</p><p>Tesco provides a wide choice of products to its clients. It also sells products that are alternatives for the majority of the items. If there are substitutes for Tesco products, they are either of poor quality or of great quality with a hefty price tag. Therefore, the intensity of the threat of substitute products is low.</p><h4>2 Resource Audit:</h4><p>A resource audit is a strategy analysis technique that is used to investigate both tangible and intangible resources (Vliegenthart, 2021). The results of a resource audit serve as a starting point for forming a strategy, initiating change, or developing products and services. Tangible resources are further classified into Physical Resources, Financial Resources, and Human Resources (Vliegenthart, 2021).</p><p>2.1 Physical Resources:</p><ul><li>Currently, 4,673 stores are operational worldwide out of which 4,008 are across the UK (Blázquez, 2021).</li><li>23 warehouses across the UK (Rahman, 2020).</li></ul><p>2.2 Financial Resources:</p><ul><li>In the financial year of 2021, Tesco’s annual revenue amounted to 53.17 billion British pounds in the United Kingdom and the Republic of Ireland (Tesco PLC, 2022).</li><li>Whereas, Tesco’s international annual revenue amounted to 4.71 billion British pounds (Tesco PLC, 2022).</li></ul><p>2.3 Human Resources:</p><p>· 367,361 employees across the globe (Blázquez, 2021).</p><p>2.4 Intangible Resources:</p><ul><li>6th most valuable brand in the UK (Statista, 2021)</li><li>Top in the grocery market share in the UK (Statista, 2021)</li><li>4th leading food and beverage retailer in Europe in 2020 (Retail Index, 2020)</li></ul><h4>3. CATWOE Analysis</h4><p>CATWOE is a technique that provides a framework for defining and analyzing business stakeholder perspectives. The mnemonic stands for Customer, Actor, Transformation, Worldview, Owner, and Environment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gUlk1NOQnM7pOvmRxHQ4nQ.png" /><figcaption>CATWOE Analysis of Tesco</figcaption></figure><p>For further detailed analysis, kindly refer to the research paper on <em>Business Process Analysis on Tesco 2022</em> which can be found <a href="https://github.com/tron135/research_paper/blob/main/Business%20Process%20Analysis%20on%20Tesco%202022.pdf">here</a>.</p><h4>4. Comments, Observation, and Recommendation</h4><p>On the basis of what we have observed so far, Tesco has gone through a hard time in the past few years. The number of stores all over the world has decreased by 3000 in the past 4–5 years. According to the strategy analysis done using Michael Porter’s Five Forces Model and Resource Audit, Tesco is a fast-growing company and still can expand a lot in the UK and ROI (Regions Of Ireland) itself. Although Tesco has exited many countries, the annual revenue, and the quality of services has not decreased. The chance of losing from the competition from the following major retailers in the market will still be there even though Tesco is far ahead as mentioned in section 1.1. Still, the majority of the products from Tesco with its own brand name are unbeatable. Tesco worked really hard to complete the needs of its customers even in the times of Covid and managed to maintain its annual revenue in the same year.</p><p>According to CATWOE analysis, we realized the important perspectives of some of the crucial stakeholders and prioritized them with the help of the power/interest grid (mentioned in the detailed analysis). We came to know that in Covid times, even with the Government policies and regulations, Tesco managed to deliver the products at an efficient cost regardless of the lack of suppliers. It is important that Tesco develop a model in such a way that it is able to earn profits as well as not compromise customer needs. From all this, we can easily say that Tesco really cares about its customers and always tries to give the best service possible.</p><p>UML use case diagrams showcase how Tesco has worked in order to maintain its clientele without complicating the purchasing procedure. Even a new customer can buy products easily with the help of self-guided systems. Further detailed representation is given in the UML class diagram and UML sequence diagrams.</p><p>Overall, Tesco is a reliable industry in which a customer can put his trust. Tesco has its products in almost every industry nowadays some of which include Tesco banks, Tesco sim cards, Tesco mobile phones, etc</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3c416398fac1" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>