Performance analysis of large language models in the domain of legal argument mining

Generative pre-trained transformers (GPT) have recently demonstrated excellent performance in various natural language tasks. The development of ChatGPT and the recently released GPT-4 model has shown competence in solving complex and higher-order reasoning tasks without further training or fine-tuning. However, the applicability and strength of these models in classifying legal texts in the context of argument mining are yet to be realized and have not been tested thoroughly. In this study, we investigate the effectiveness of GPT-like models, specifically GPT-3.5 and GPT-4, for argument mining via prompting. We closely study the model's performance considering diverse prompt formulation and example selection in the prompt via semantic search using state-of-the-art embedding models from OpenAI and sentence transformers. We primarily concentrate on the argument component classification task on the legal corpus from the European Court of Human Rights. To address these models' inherent non-deterministic nature and make our result statistically sound, we conducted 5-fold cross-validation on the test set. Our experiments demonstrate, quite surprisingly, that relatively small domain-specific models outperform GPT 3.5 and GPT-4 in the F1-score for premise and conclusion classes, with 1.9% and 12% improvements, respectively. We hypothesize that the performance drop indirectly reflects the complexity of the structure in the dataset, which we verify through prompt and data analysis. Nevertheless, our results demonstrate a noteworthy variation in the performance of GPT models based on prompt formulation. We observe comparable performance between the two embedding models, with a slight improvement in the local model's ability for prompt selection. This suggests that local models are as semantically rich as the embeddings from the OpenAI model. Our results indicate that the structure of prompts significantly impacts the performance of GPT models and should be considered when designing them.

NAG: neural feature aggregation framework for credit card fraud detection

The state-of-the-art feature-engineering method for fraud classification of electronic payments uses manually engineered feature aggregates, i.e., descriptive statistics of the transaction history. However, this approach has limitations, primarily that of being dependent on expensive human expert knowledge. There have been attempts to replace manual aggregation through automatic feature extraction approaches. They, however, do not consider the specific structure of the manual aggregates. In this paper, we define the novel Neural Aggregate Generator (NAG), a neural network-based feature extraction module that learns feature aggregates end-to-end on the fraud classification task. In contrast to other automatic feature extraction approaches, the network architecture of the NAG closely mimics the structure of feature aggregates. Furthermore, the NAG extends learnable aggregates over traditional ones through soft feature value matching and relative weighting of the importance of different feature constraints. We provide a proof to show the modeling capabilities of the NAG. We compare the performance of the NAG to the state-of-the-art approaches on a real-world dataset with millions of transactions. More precisely, we show that features generated with the NAG lead to improved results over manual aggregates for fraud classification, thus demonstrating its viability to replace them. Moreover, we compare the NAG to other end-to-end approaches such as the LSTM or a generic CNN. Here we also observe improved results. We perform a robust evaluation of the NAG through a parameter budget study, an analysis of the impact of different sequence lengths and also the predictions across days. Unlike the LSTM or the CNN, our approach also provides further interpretability through the inspection of its parameters.

On the quality of semantic interest profiles for onine social network consumers

Social media based recommendation systems infer user' interests and preferences from their social network activity in order to provide personalised recommendations. Typically, the user profiles are generated by analysing the users' posts or tweets. However, there might be a significant difference between what a user produces and what she consumes. We propose an approach for inferring user interests from followees (the accounts the user follows) rather than tweets. This is done by extracting named entities from a user's followees using the English Wikipedia as knowledge base and regarding them as interests. Afterwards, a spreading activation algorithm is performed on a Wikipedia category taxonomy to aggregate the various interests to a more abstract and broader interest profile. We evaluate the coverage of followee lists in terms of named entities and show that they provide sufficient input to infer comprehensive semantic interest profiles. Further, we compare the profiles created with the followee-based approach against tweet-based profiles. With over 7 out of 10 items being relevant to the users in our evaluation, we show that the followee-based approach can compete with the state of the art and performs even better in predicting the user's interests than their human friends do.

Inhibition of endothelium-derived relaxing factor in A6 cells

The effects of endothelium-derived relaxing factor (EDRF) on Na+ transport in distal renal tubular A6 cells have been studied by inhibition of its synthesis with L-NAME (10(-2) mol/l). Na+ transport was monitored by measuring short-circuit current, cell voltage, transepithelial, apical and basolateral membrane conductances. EDRF production in A6 cells was tested by application of its substrate L-arginine. The blockade of EDRF decreased significantly the Na+ current (11 %), membrane potential (5 mV) and basolateral conductance (33 %), but did not affect the apical membrane conductance. Activation of apical Na+ conductance by dexamethasone incubation (10(-7) mol/l) did not further influence the drop in Na+ current. The involvement of basolateral K+ channels in cell depolarization and in the reduction of basolateral conductance was tested in tissues with elevated basolateral K+/Cl- conductance ratios (by increasing bath osmolarity) and by application of barium (0.5-10(-3) mol/l) a K+ channel blocker. The results showed that the effect of L-NAME on the short-circuit current was more pronounced in A6 cells with increased K+/Cl- conductance ratios, but was almost nullified by barium. Finally, L-arginine fully restored the Na+ current, thus reversing the inhibition induced by L-NAME. We conclude that EDRF is basally released in A6 cells. Inhibition of EDRF by L-NAME directly interferes with Na+ reabsorption. Since apical membrane conductance remains unchanged, the decrease in short-circuit current results from cell depolarization. The latter, together with the drop in basolateral conductance, might reflect inactivation of K+ channels.

Effect of dexamethasone on sodium channel block and densities in A6 cells

The association (ON) and dissociation (OFF) rates of either positively charged amiloride or its uncharged analogue, CDPC (6-chloro-3, 5-diaminopyrazine-2-carboxamide), with the apical Na+ channel protein of renal A6 cells were analysed during exposure to the synthetic glucocorticoid, dexamethasone, using noise analysis. These rates were further used to reach specific conclusions about single-channel current, channel density and open probability of the channel in the absence of the blocker. Short-term exposure (3 h) to 10(-7) mol/l dexamethasone at the basolateral side increased the short-circuit current, Isc by 85%, without a change in the ON and OFF rates of the interaction between amiloride and the Na+ channel. A longer incubation (24 h) with dexamethasone tripled the current with a notable increase in the ON rate of the interaction between amiloride and the and channel. The OFF rate remained constant. The effects of dexamethasone on the rate constants of the reaction of amiloride with the channel did not match with the expected changes in membrane potential. On the other hand, ON and OFF rates of the interaction between neutral CDPC and the channel were not influenced by a 24-h incubation with dexamethasone. Further calculations disclosed that the gain in macroscopic current after a 24-h incubation with dexamethasone might be explained by an increase in Na+ channel density, and, to a lesser extent, by a rise in single-channel current. This all occurred without a change in the fraction of time spent by the channel in the conducting state in the absence of the blocker.

Effect of ouabain on membrane conductances and volume in A6 cells

The present study reports the effect of a reduction in the Na(+)-transport rate on cell volume. A decrease in transport rate was achieved by inhibition of the basolateral Na+/K+ pump with ouabain. Cultured A6 cell monolayers were short-circuited and exposed to ouabain at the basolateral surface. In one series of experiments, cells were impaled with microelectrodes to measure cell voltage, apical fractional resistance and thus derive membrane conductances. Another set, A6, served for cell height measurements. Ouabain decreased short-circuit current (Isc), which is an index of transepithelial Na+ transport: the reduction in transport rate varied from 26 to 79% within 10 min. Equivalent circuit analysis revealed a 20% decrease in apical membrane conductance (ga), whereas basolateral membrane conductance (gb) increased by 66%. A decrease in cell voltage (12 mV) together with drop in ga during ouabain may account for the reduction in Isc. The rise in gb is mainly due to a gain in Cl- conductance which increased from 114 to 613 microS/cm2, compatible with activation of Cl- channels. All of this occurs without a detectable change in cell height. We may conclude from these data that inhibition of Na+ exit by ouabain is quickly compensated by a decrease in apical Na+ entry and an increase in basolateral Cl- conductance. Constant cell volume during ouabain implies that the total cell solute is essentially unchanged.

K+ recirculation in A6 cells at increased Na+ transport rates

Homocellular regulation of K+ at increased transcellular Na+ transport implies an increase in K+ exit to match the intracellular K+ load. Increased K+ conductance, gK, was suggested to account for this gain. We tested whether such a mechanism is operational in A6 monolayers. Na+ transport was increased from 5.1 +/- 1.0 microA/cm2 to 20.7 +/- 1.3 microA/cm2 by preincubation with 0.1 mumol/l dexamethasone for 24 h. Basolateral K+ conductances were derived from transference numbers of K+, tK, and basolateral membrane conductances, gb, using conventional microelectrodes and circuit analysis with application of amiloride. Activation of Na+ transport induced an increase in gb from 0.333 +/- 0.067 mS/cm2 to 1.160 +/- 0.196 mS/cm2 and tK was reduced to 0.22 +/- 0.01 from a value of 0.70 +/- 0.05 in untreated control tissues. As a result, gK remained virtually unchanged at increased Na+ transport rates. The increase in gb after dexamethasone was due to activation of a conductive leak pathway presumably for Cl-. Increased K+ efflux, IK, was a consequence of the larger driving force for K+ exit due to depolarization at an elevated Na+ transport rate. The relationship between calculated K+ fluxes and Na+ transport rate, measured as the Isc, is described by the linear function IK = 0.624 x INa -0.079, which conforms with a stoichiometry 2:3 for the fluxes of K+ and Na+ in the Na+/K(+)-ATPase pathway. Our data show that homocellular regulation of K+ in A6 cells is not due to up-regulation of gK.

Osmotic swelling and membrane conductances in A6 cells

Hyposmotic basolateral perturbations (-30 mosmol/kg) in cultured renal layers (A6) increased basolateral membrane conductance more than 2-fold within 10 min; the increase was partly due to upregulation of K+ conductance, but other conductive pathways were also activated. The raise in apical membrane amiloride-sensitive Na+ conductance was less pronounced; it appears to be due to secondary effects.

Basolateral membrane conductance in A6 cells: effect of high sodium transport rate

Conductance of apical and basolateral membranes in short-circuited cultured renal distal cells (A6) was determined using microelectrodes. Epithelia were pre-incubated with 0.1 mumol/l dexamethasone in the presence of 4 mumol/l amiloride to prevent increase in apical Na+ entry. Omission of amiloride increased the Isc from 5.7 to 27.6 microA/cm2 due to the rise in apical membrane conductance from 21 to 595 microS/cm2. Apical fractional resistance decreased from 0.89 to 0.40 and cells depolarized from -52 to -4 mV. Basolateral membrane conductance, which was 320 microS/cm2 at partially inhibited transport, was not significantly altered during the first 2 min following establishment of high transport activity; it started to increase thereafter reaching a more than threefold higher value of 1324 microS/cm2 within 12 min. The gain cannot be explained by increase in partial K+ conductance. Disappearance of the conductance after reduction of basolateral Cl- or in the presence of the Cl- channel blocker 5-nitro-2-(3-phenylpropylamino)benzoate indicates a Cl- conductance, which appears to be activated by depolarization.

Voltage dependent membrane conductances in cultured renal distal cells

Cultured Na(+)-transporting epithelia from amphibian renal distal tubule (A6) were impaled with microelectrodes and analyzed at short-circuit and after transepithelial voltage perturbation to evaluate the influence of voltage on apical and basolateral membrane conductances. For equivalent circuit analysis, amiloride was applied at each setting of transepithelial potential. At short-circuit, apical and basolateral membrane conductances averaged 88 and 497 microS/cm2, respectively (n = 10). Apical membrane conductance, essentially due to Na(+)-specific pathways, decreased after depolarization of the apical membrane. The drop was considerably larger than predicted by the Goldman-Hodgkin-Katz (GHK) constant-field equation. This suggests decrease in permeability of the apical Na+ channels upon depolarization. Basolateral membrane conductance, preferentially determined by K+ channels, increased after hyperpolarization of the basolateral membrane. This behavior is contrary to the prediction of the GHK constant field equation and reflects inward rectification of the K+ channels. The observed rectification patterns can be valuable for maintenance of cellular homeostasis.

Apical and basolateral conductance in cultured A6 cells

Confluent monolayers of the cultured renal distal tubule cell line (A6) were impaled with microelectrodes under short-circuit conditions. Specific membrane conductances were calculated from equivalent circuit equations. Transport properties of the apical and basolateral membranes were investigated during control conditions and short-term increases in basolateral potassium concentration [K+] from 2.5 to 20 mmol/l, with or without 0.5 mmol/l Ba2+ at the basolateral side. As in most other epithelia, the apical membrane represents the major resistive barrier. Transcellular, apical and basolateral membrane conductances (gc, go and gi respectively), obtained from 22 acceptable microelectrode studies, averaged 61, 80 and 292 microS/cm2, respectively. There was a highly significant correlation between short-circuit current (Isc) and go, whereas gi was unrelated to Isc. The Isc, which averaged 4.1 microA/cm2, was almost completely blocked by amiloride. This was associated with fast hyperpolarization; the intracellular potential (Vsc) increased from -69 to -83 mV and the fractional apical resistance rose to nearly 100%. Using the values of Vsc during amiloride at normal and high [K+], an apparent transference number for K+ at the basolateral membrane of 0.72 can be calculated. This value corresponds with the decrease in gi to about 25% of the control values after blocking the K+ channels with Ba2+. The nature of the remaining conductance is presently unclear. The cellular current decreased during high [K+] and Ba2+, in part resulting from reduction of the electrochemical gradient for apical Na+ uptake due to the depolarization.(ABSTRACT TRUNCATED AT 250 WORDS)

Regulation of ion conductance in frog skin by isoproterenol

The effect of isoproterenol on apical and basolateral membrane conductance in principal cells of short-circuited frog skin was analyzed using microelectrodes. Isoproterenol (10(-6) mol/l) increased the apical membrane conductance in addition to stimulating Cl- conductive pathways outside the principal cells. The effect on apical Na+ channels explains the increase in amiloride sensitive short-circuit current. Basolateral membrane conductance increased only slightly. Steady-state I/V relationships of the basolateral membrane indicate that the inward rectification of basolateral membrane K+ channels was not altered.

Dual effect of barium on basolateral membrane conductance of frog skin

The effect of Ba2+ on basolateral membrane conductance (gi) in isolated frog skins was analysed. Response patterns were different in tissues with high and low spontaneous intracellular potential. At high (negative) potentials, serosal Ba2+ inhibited gi as is expected of a potent K+ channel blocker, whereas in tissues with low potential, gi remained unchanged or even increased after Ba2+. The direction of change in gi was also dependent on the magnitude of gi under control conditions. Decrease of gi was only observed at high gi in the control period. In contrast, gi increased if control values of gi were below 0.5 mS/cm2. In tissues with spontaneously low intracellular potential, an inhibitory effect of Ba2+ on gi could be induced by hyperpolarization of the basolateral membrane with transepithelial voltage perturbation. Under these conditions, voltage-dependent, inward rectifying K+ channels are activated, which are Ba2(+)-sensitive. Furthermore, hyperpolarization of the basolateral membrane potential (Vi) during Ba2+ rapidly decreased gi. These results suggest that Ba2+, in addition to blocking K+ channels, activates (presumably unspecific) basolateral membrane channels. This dual effect, which is obvious in tissues with low spontaneous gi, might similarly exist in tissues with high control gi. Identification, however, is virtually impossible due to the large decrease in potassium conductance.

Base induced hyperpolarization of the cell potential in HCO3- free perfused Necturus renal proximal tubules

Short-term peritubular alkalinization from 7.5 to 8.5 hyperpolarized (-8.8 mV) the basolateral membrane potential (V1) in HCO3- free Herpes buffered Necturus renal proximal tubule cells. This sustained base induced hyperpolarization (BIH) was associated with an increase in the peritubular apparent transference number for potassium (tK+). The apparent transference number for potassium (tK+) was estimated at pH 7.5 and 8.5 by raising peritubular K+ from 2.5 to 10 mmol/l. tK+ increased linearly as V1 hyperpolarized, whereas tK+ measured in the presence of peritubular Ba2+ at pH 7.5 and 8.5 was nearly zero. However, the BIH persisted in the presence of barium at the peritubular, luminal or both sides of the epithelium. Moreover this BIH was also accompanied by a small hyperpolarization (-0.4 mV) of the transepithelial membrane potential (V3) in the absence or presence of peritubular and/or luminal Ba2+. Therefore we conclude that BIH must originate from additional mechanisms other than an increase in peritubular or luminal potassium conductance.