Funções do litsearchr para identificar palavras-chave

Para utilizar as funções deste pacote, precisamos realizar uma pesquisa inicial. Utilizamos a seguinte lista de palavras-chave no Web of Science:

“(atlantic AND (rainforest* OR forest)) OR restinga OR ((mangrove* OR grassland* OR scrub) AND (neotropic OR brazil* OR south america NOT pampa* NOT cerrado)) AND (vegetation OR flor OR plant* OR phyt) AND (tree OR shrub* OR epiphyte* OR liana* OR herb* OR paleoecology OR palynology OR bryophyte* OR pteridophyte* OR gymnosperm* OR angiosperm* OR seed* OR ontogeny OR”seed rain” OR “seed bank” OR “seed banks” OR “ecological succession” OR “secondary succession” OR spore* OR pollen OR community OR assembl* OR biodiversity OR fragment* OR fire OR deforest* OR log* OR “slash-and-burn” OR livestock OR “land conversion” OR “land use” OR exotic OR invas* OR alien OR non-native OR non-indigenous OR “edge effect” OR landscape OR eudicot* OR monocot* OR palm* OR “climate change”)”

Essa pesquisa resultou em 20679 artigos. Os resultados completos foram baixados em .ris e analisados nas funções do pacote. Com todos os nossos arquivos na mesma pasta, podemos usar a função import results para criar um dataframe e juntar todos os arquivos. Depois usar a função remove_duplicates para tirar artigos que tenham exatamente o mesmo título:

setwd("C:/Users/amabi/OneDrive/Área de Trabalho/New folder (2)")


library(litsearchr)

list<-list.files()
list<-list[2:22]
head(list)

[1] "savedrecs (1).ris"  "savedrecs (10).ris" "savedrecs (11).ris"
[4] "savedrecs (12).ris" "savedrecs (13).ris" "savedrecs (14).ris"

data<-import_results(file=list)

Reading file savedrecs (1).ris ... done
Reading file savedrecs (10).ris ... done
Reading file savedrecs (11).ris ... done
Reading file savedrecs (12).ris ... done
Reading file savedrecs (13).ris ... done
Reading file savedrecs (14).ris ... done
Reading file savedrecs (15).ris ... done
Reading file savedrecs (16).ris ... done
Reading file savedrecs (17).ris ... done
Reading file savedrecs (18).ris ... done
Reading file savedrecs (19).ris ... done
Reading file savedrecs (2).ris ... done
Reading file savedrecs (20).ris ... done
Reading file savedrecs (3).ris ... done
Reading file savedrecs (4).ris ... done
Reading file savedrecs (5).ris ... done
Reading file savedrecs (6).ris ... done
Reading file savedrecs (7).ris ... done
Reading file savedrecs (8).ris ... done
Reading file savedrecs (9).ris ... done
Reading file savedrecs.ris ... done

data_dupli <-remove_duplicates(data, field = "title", method = "exact")

Só 33 artigos possuíam títulos duplicados, nossa database final tem 20646 observações. Vamos criar dois conjunto de termos com extract terms: um levando em consideração os títulos e abstracts dos artigos e outro com as palavras-chave dos autores. Os dois consideram que os termos devem aparecer ao menos em 50 artigos e que tenham de 1 a 3 palavras:

title_abstracts<-paste(data_dupli$title, data_dupli$abstract)

racked_keywords<-
  extract_terms(
    text= title_abstracts,
    keywords=NULL,
    method="fakerake",
    min_freq=100,
    ngrams = TRUE,
    min_n=1,
    max_n=3
  )

Loading required namespace: stopwords

real_keywords<-
  extract_terms(
    text=NULL,
    keywords=data_dupli$keywords,
    method="tagged",
    min_freq=100,
    ngrams=TRUE,
    min_n=1,
    max_n=3
  )

Vamos juntar todas os termos em um único objeto e criar uma matriz de “presença e ausência” deles por artigo.

all_keywords<-unique(append(real_keywords, racked_keywords))

BBWO_dfm<-
  create_dfm(elements=data_dupli$title,
             features=all_keywords)

Aqui criaremos uma rede que relaciona os termos com a quantidade de vezes que eles ocorrem. Eles tem que estar presentes em ao menos 10 artigos e ao menos 5 vezes no mesmo documento. Testei aqui também a presença em no mínimo 5 artigos. Como não aumentou muito (2072 para 2360 termos), vou seguir com a segunda opção.

BBWO_graph<-
  create_network(
    search_dfm = BBWO_dfm,
    min_studies = 10,
    min_occ=5
  )

BBWO_graph2<-
  create_network(
    search_dfm = BBWO_dfm,
    min_studies = 5,
    min_occ=5
  )

Com a rede, podemos usar a função cutoff para indicar um limite de “força” para nós importantes da rede. O método cumulativo para o cutoff encontra o ponto de limite da força do nó em que 70% da força total da rede é capturada.

cutoff<-
  find_cutoff(
    graph=BBWO_graph2,
    method="cumulative",
    percent=0.7,
    imp_method ="strength"
  )

Agora podemos reduzir nossa rede inicial, que tinha muitos termos, com o nosso valor de cutoff. Utilizamos a função reduce graph para isso:

reduced_graph <-reduce_graph(BBWO_graph2, cutoff_strength = cutoff)

Vamos ver nossa rede final:

library(ggraph)

Warning: package 'ggraph' was built under R version 4.2.3

Loading required package: ggplot2

Warning: package 'ggplot2' was built under R version 4.2.3

ggraph(reduced_graph)+
  coord_fixed()+
  expand_limits(x=c(-3, 3))+
  geom_edge_link(aes(alpha=weight))+
geom_node_point(shape="circle filled", fill="white")+
  geom_node_text(aes(label=name), hjust="outward", check_overlap=TRUE)+
  guides(edge_alpha="none")

Using "stress" as default layout

Warning: Using the `size` aesthetic in this geom was deprecated in ggplot2 3.4.0.
ℹ Please use `linewidth` in the `default_aes` field and elsewhere instead.

Podemos extrair os termos na nossa rede final e escrevê-los novamente para realizar uma nova busca, com as funções get_keywords e write_search, respectivamente:

search_terms<-
  get_keywords(reduced_graph)

nova_busca<-write_search(
  search_terms,
  languages="English",
  exactphrase=TRUE,
  stemming=FALSE,
  closure="left",
  writesearch=FALSE
)

[1] "English is written"

nova_busca

[1] "((abundance) AND (amazon) AND (amazonia) AND (amphibia) AND (anura) AND (area) AND (areas) AND (argentina) AND (assemblages) AND (atlantic) AND (\"atlantic forest\") AND (\"atlantic rain forest\") AND (\"atlantic rainforest\") AND (bahia) AND (basin) AND (biodiversity) AND (biology) AND (biomass) AND (birds) AND (brazil) AND (\"brazilian atlantic forest\") AND (carbon) AND (cerrado) AND (climate) AND (\"climate change\") AND (coleoptera) AND (communities) AND (community) AND (conservation) AND (density) AND (diet) AND (dispersal) AND (distribution) AND (disturbance) AND (diversity) AND (dynamics) AND (ecology) AND (ecosystem) AND (ecosystems) AND (edge) AND (evolution) AND (fauna) AND (fire) AND (fish) AND (flora) AND (forest) AND (forests) AND (fragmentation) AND (fragments) AND (frog) AND (\"genetic diversity\") AND (genus) AND (gradient) AND (grassland) AND (grasslands) AND (growth) AND (habitat) AND (history) AND (holocene) AND (hymenoptera) AND (impact) AND (impacts) AND (land) AND (\"land use\") AND (landscape) AND (light) AND (litter) AND (mammals) AND (management) AND (mangrove) AND (model) AND (neotropical) AND (\"new species\") AND (nitrogen) AND (\"northeastern brazil\") AND (pasture) AND (patterns) AND (phenology) AND (plant) AND (plantations) AND (plants) AND (pollen) AND (population) AND (populations) AND (\"protected areas\") AND (quality) AND (record) AND (records) AND (regeneration) AND (region) AND (responses) AND (restinga) AND (restoration) AND (richness) AND (river) AND (savanna) AND (scale) AND (size) AND (\"small mammals\") AND (soil) AND (\"south america\") AND (\"southeastern brazil\") AND (\"southern brazil\") AND (\"species richness\") AND (state) AND (succession) AND (systems) AND (temperature) AND (traits) AND (tree) AND (trees) AND (\"tropical forest\") AND (\"tropical forests\") AND (variability) AND (vegetation) AND (water) AND (ability) AND (action) AND (actions) AND (activity) AND (affect) AND (africa) AND (agricultural) AND (alter) AND (amazonian) AND (america) AND (american) AND (amphibian) AND (analysis) AND (anthropogenic) AND (anuran) AND (apidae) AND (approach) AND (araucaria) AND (assemblage) AND (assess) AND (assessment) AND (\"atlantic coast\") AND (\"atlantic coastal\") AND (\"atlantic forest fragment\") AND (\"atlantic forest fragments\") AND (\"atlantic forest remnant\") AND (based) AND (beetle) AND (biogeographic) AND (biological) AND (biome) AND (boreal) AND (brazilian) AND (\"brazilian atlantic\") AND (\"brazilian atlantic rainforest\") AND (bromeliad) AND (central) AND (change) AND (character) AND (characterization) AND (chemical) AND (climatic) AND (coast) AND (coastal) AND (\"coastal plain\") AND (comparison) AND (complex) AND (composition) AND (condition) AND (conditions) AND (cover) AND (cultural) AND (deciduous) AND (description) AND (develop) AND (development) AND (differ) AND (diptera) AND (drive) AND (driver) AND (drivers) AND (dynamic) AND (early) AND (eastern) AND (\"eastern brazil\") AND (ecologica) AND (ecological) AND (effect) AND (effects) AND (endangered) AND (endemic) AND (environment) AND (environmental) AND (espirito) AND (\"espirito santo\") AND (europe) AND (evidence) AND (factor) AND (factors) AND (feeding) AND (flies) AND (floristic) AND (\"forest biome\") AND (\"forest fragment\") AND (\"forest fragments\") AND (\"forest remnant\") AND (forestry) AND (formation) AND (fragment) AND (fragmented) AND (fruit) AND (function) AND (functional) AND (genera) AND (generation) AND (genetic) AND (geographic) AND (geographical) AND (gerais) AND (glacial) AND (global) AND (grande) AND (grass) AND (ground) AND (group) AND (habit) AND (habitats) AND (historic) AND (historical) AND (hotspot) AND (human) AND (hylidae) AND (implications) AND (indicator) AND (influence) AND (insect) AND (insight) AND (insights) AND (interact) AND (interaction) AND (interactions) AND (island) AND (janeiro) AND (lands) AND (landscapes) AND (large) AND (level) AND (limit) AND (lizard) AND (local) AND (lower) AND (lowland) AND (mammal) AND (marine) AND (matter) AND (method) AND (mid-atlantic) AND (minas) AND (\"minas gerais\") AND (modeling) AND (molecular) AND (montane) AND (morphological) AND (mountain) AND (national) AND (native) AND (natural) AND (network) AND (niche) AND (north) AND (northeast) AND (northeastern) AND (northern) AND (notes) AND (nutrient) AND (occur) AND (occurrence) AND (ocean) AND (orchid) AND (organic) AND (parana) AND (patch) AND (pattern) AND (paulo) AND (phylogenetic) AND (plain) AND (plantation) AND (position) AND (potential) AND (predict) AND (process) AND (product) AND (production) AND (productive) AND (protect) AND (protected) AND (rainforest) AND (range) AND (ratio) AND (recover) AND (regional) AND (relate) AND (related) AND (relation) AND (relations) AND (relationship) AND (relationships) AND (remnant) AND (remnants) AND (reproductive) AND (reserve) AND (resource) AND (response) AND (reveal) AND (review) AND (riparian) AND (rivers) AND (rodent) AND (santa) AND (santo) AND (satellite) AND (season) AND (seasonal) AND (secondary) AND (sediment) AND (seedling) AND (select) AND (serra) AND (serve) AND (sites) AND (small) AND (\"small mammal\") AND (soils) AND (source) AND (sources) AND (south) AND (south-eastern) AND (\"south american\") AND (southeast) AND (\"southeast brazil\") AND (southeastern) AND (southern) AND (spatial) AND (species) AND (specific) AND (stage) AND (stand) AND (states) AND (station) AND (status) AND (stems) AND (stock) AND (stream) AND (streams) AND (structure) AND (study) AND (subtropical) AND (success) AND (system) AND (table) AND (taxon) AND (taxonomic) AND (temporal) AND (terrestrial) AND (threat) AND (threatened) AND (trait) AND (transition) AND (tropical) AND (types) AND (urban) AND (variation) AND (vertebrate) AND (waters) AND (watershed) AND (western) AND (wetland) AND (woody) AND (years))"

Apesar de ser uma ferramenta interessante, muitas palavras-chave que não são do nosso interesse ainda estão sendo capturados com a busca inicial. Além disso, a função write_search sempre utiliza AND (pelo menos nos testes que rodei) e isso impede uma busca efetiva nas plataformas.

Sugiro que usemos a ferramenta na identificação e melhora dos termos de pesquisa, afinal, com a nossa busca inicial, capturamos mais de 20.000 artigos. Abaixo, uma sugestão de nova busca: