Update text.py

text.py

@@ -8,27 +8,33 @@ Text summarization methods can either be used as an extractive or abstractive mo
 
 <h3><center> Abstractive Model </center></h3>
 
-
 The underlying engines for the Abstractive part are transformer based model BART, a sequence-to-sequence model with a bidirectional (BERT-like) encoder and an autoregressive (GPT-like) decoder. The BART-model was pre-trained by KBLab/bart-base-swedish-cased (link) to learn general knowledge about language. Afterwards, the model was further fine-tuned on two labelled datasets that have been open-sourced:
 
 - [Gabriel/xsum_swe](https://huggingface.co/datasets/Gabriel/xsum_swe)
 - [Gabriel/cnn_daily_swe](https://huggingface.co/datasets/Gabriel/cnn_daily_swe)
 
-To see more in depth regarding the training go to model card: [Gabriel/bart-base-cnn-xsum-swe](https://huggingface.co/Gabriel/bart-base-cnn-xsum-swe).
+To see more in depth regarding the training go to model card: [Gabriel/bart-base-cnn-xsum-swe](https://huggingface.co/Gabriel/bart-base-cnn-xsum-swe). 
+
+The core idea behind the training procedure is sequential adoption through transfer learning, i.e multiple phases for fine-tuning a pre-trained model on different datasets. For more information on this topic read: https://arxiv.org/pdf/1811.01088v2.pdf. 
+
 """
       
 sum_app_text_tab_2= """ 
 <h2><center> 🤗  </center></h2>
 
-The core idea behind the training procedure is sequential adoption through transfer learning, i.e multiple phases for fine-tuning a pretrained model on different datasets. The figure below illustrates how the skill level of the model increases at each step:
+The figure below illustrates how the skill level of the model increases at each step:
 ![alt text2](file/BART_SEQ.png)
 
 The main benefits of transfer learning in general include the saving of resources and improved efficiency when training new models, so feel free to adopt this model for your type of problem! 
 
 <h3><center> Extractive Model  </center></h3>
 
-The extractive models for this app are using sentence-transformer models, which basically is a bi-encoder that determines how similar two sentences are. This type of models convert texts into vectors (embedding) that capture semantic information. Additionally, LexRank, an unsupervised graph-based algorithm, is used to determine centrality scores as a post-process step to summarise. The main idea is that sentences "recommend" other similar sentences to the reader. Thus, if one sentence is very similar to many others, it will likely be a sentence of great importance. The importance of this sentence also stems from the importance of the sentences "recommending" it. Thus, to get ranked highly and placed in a summary, a sentence must be similar to many sentences that are in turn also similar to many other sentences.
+The extractive models for this app are using sentence-transformer models, which basically is a bi-encoder that determines how similar two sentences are. This type of models convert texts into vectors (embedding) that capture semantic information. Additionally, LexRank, an unsupervised graph-based algorithm, is used to determine centrality scores as a post-process step to summarise. The main idea is that sentences "recommend" other similar sentences to the reader. Thus, if one sentence is very similar to many others, it will likely be a sentence of great importance. The importance of this sentence also stems from the importance of the sentences "recommending" it. Thus, to get ranked highly and placed in a summary, a sentence must be similar to many sentences that are in turn also similar to many other sentences. 
 ![alt text3](file/Lex_rank.png)
+
+The figure above showcase how LexRank formats similarity graphs based on all possible sentence combinations sentence similarity from the vector embeddings. Notice that the most "recommended" sentences that are extracted (the right graph) are derived from a threshold value which filters "weaker" connections in the similarity graph.
+For more information on this topic read: https://www.aaai.org/Papers/JAIR/Vol22/JAIR-2214.pdf
+
 """         
 
 abstractive_example_text_1= """Frankrike lås Sebastien Chabal har nämnts för en farlig tackling på Englands Simon Shaw under lördagens VM semifinal i Paris. Simon Shaw lastar av trots att Raphael Ibanez, vänster, och Sebastien Chabal. Sale Sharks framåt kommer att ställas inför en disciplinär utfrågning på måndag efter hans tackling på motsatt andra-rower Shaw noterades genom att citera kommissionär Dennis Wheelahan. Chabal började matchen på ersättningsbänken, men kom i 26: e minuten att ersätta den skadade Fabien Pelous under värd Frankrikes 14-9 nederlag. Om han blir avstängd missar Chabal fredagens tredje och fjärde match på Parc des Princes. Samtidigt, Frankrike tränare Bernard Laporte sade att nederlaget var svårare att ta än Englands 24-7 seger i 2003 semifinalen. "År 2003 var de bättre än oss. I själva verket var de bättre än alla", sade Laporte, som lämnar sin roll att tillträda posten som junior idrottsminister i den franska regeringen. "De var som Nya Zeeland i denna turnering - favoriten, förutom att de gick hela vägen. Den här gången är det svårare för igår var det 50-50." Samtidigt, England -- försöker bli den första nationen att försvara VM-titeln -- avslöjade att stjärna kicker Jonny Wilkinson återigen hade problem med matchbollarna under semifinalen. Flughalvan, som uttryckte sin oro efter att ha kämpat med stöveln mot Australien, avvisade en boll innan han sparkade en vital trepoängare mot Frankrike. "Vi sa det inte förra veckan men en icke-match bollen kom ut på fältet i Marseille som Jonny sparkade," chef för rugby Rob Andrew sade. "Han tänkte inte på det när han sparkade det. Matchbollarna är märkta, numrerade ett till sex. Igår kväll hade de "World Cup semifinal England vs Frankrike" skrivet på dem. På matchkvällen var Jonny vaksam när han sparkade för mål att de faktiskt var matchbollar han sparkade. "Träningsbollarna förlorar tryck och form. Hela frågan förra veckan, arrangörerna accepterade alla sex matchbollar bör användas av båda sidor på torsdagen före matchen. " E-post till en vän."""